Today, the popularity of 3D videos is increasing significantly. This trend can be attributed to their immersive appeal and lifelike experience. In an era dominated by the widespread distribution of digital content, data integrity, and ownership, all of these elements are of crucial importance. In this context, the practice of traitor tracing, closely related to Digital Rights Management (DRM), facilitates the identification and tracking of unauthorized users who have violated copyright in order to share illegal copyright-protected content. In this paper, we propose a solution to this problem, we introduce an innovative traitor tracing approach focused on 3D video, with a particular focus on the DIBR (Depth Image-Based Rendering) format, which can be vulnerable to an Interleaving attack strategy. For this purpose, we develop a new phylogeny tree construction method designed to combat collusion attacks. Our experimental evaluations demonstrate the effectiveness of our proposed approach particularly when applied to long fingerprinting codes. Compared to Tardos’ approach, our method delivers very good results, even for a large number of colluders.

In multimedia forensics, most research focuses on analyzing individual assets, such as images or videos, to verify their authenticity by analyzing their intrinsic properties. However, provenance analysis examines multiple assets collectively, tracing their histories through pairwise relationships. This approach is beneficial in identifying traces of manipulation and understanding the evolution of digital content. In this research, we apply provenance analysis to detect and track malicious users engaged in collusion attacks by analyzing their fingerprinting codes. Specifically, the correlations we consider are not generic media content similarities, but rather statistical and structural relationships between the embedded fingerprinting codes of different multimedia assets. By quantifying these correlations, our method can reliably identify individuals responsible for generating unauthorized copies and distinguish between different collusion strategies. Furthermore, it allows us to classify and estimate the specific type of collusion attack employed, providing deeper insights into the strategies used by attackers, thereby aiding in the development of more robust countermeasures and enhancing the resilience of multimedia security systems.

In the computer vision field, object recognition becomes a very active field of interest. In this research, we proposed a very deep learning approach namely Boosted DenseNet. Our method used the BCNN architecture boosted by conducting reinforced dense blocks. These reinforced dense blocks consisted of a similar convolutional layers numbers as BCNN boosted by adding MBA Activation and Concatenated Rectified Linear Unit functions. In addition, our Boosted DenseNet was improved by applying boosted convolutional layers which offer a deeper network with the same number of parameters. Besides, our method has been reinforced by adding Generalizing Pooling layer. Generalizing pooling, which combined pooling operations within a hierarchical tree structure, replaced the max-pooling layer in reinforced dense blocks. Experiments on CFAR-10 and Pascal VOC 2007 have outperformed the state-of-the-art approaches and demonstrated the robustness of our model for object recognition tasks.

Cancer is a critical global public health problem with meager median survival. It is therefore quite essential to detect this disease at an early stage to improve diagnostic results and consequently avoid serious complications. For this purpose, various researchers have implemented automated methods with the use of different medical imaging modalities. Accordingly, the expansion of deep learning techniques grants opportunities to enhance diagnosis, cure, and prevention. In this study, a diagnostic system for accurate classification of ultrasound breast abnormalities based on the powerful ResNet-50 CNN is proposed with the aim of providing early detection of breast cancer decease. The contribution of this work lies in the novel approach taken to improve the performance of the ResNet50 model in the classification of ultrasound breast cancer images. Transfer learning allows for the model to leverage pre-existing knowledge, while the application of data augmentation techniques enhances the diversity and quality of the training data. Additionally, the optimization of the batch size as a hyperparameter ensures that the model is able to effectively learn from the training data, leading to improved accuracy and efficiency in the classification process. This approach is crucial in the early detection and treatment of breast cancer. Quantitative and qualitative evaluations have been detailed in this study using Breast Ultrasound Dataset BUSI. Our presented work shows interesting results in terms of accuracy, specificity, sensitivity, and AUC which exceed the performance of other compared works. Moreover, the proposed method helps boost the clinical diagnosis of breast cancer. It may integrate a radiologist network, allowing them to constantly follow up on the patient's medical history.

The performance and power of Wavelet Neural Networks (WNNs) rely on the proper structure of the WNN. In this study, a hybrid approach is suggested to build a Wavelet Neural Network. We design a network based on the genetic algorithm (GA), the Gram-Schmidt Selection (GSS) method, and the multilibrary wavelet function (MLWF). The initialization of the WNN is performed by the GSS method, which aims to select the candidate library wavelet functions (MLWF) to develop the WNN. The GA is used to solve the structure and the learning of the WNN and the GSS algorithm is applied to select the important wavelets. The GA was used to calculate the suitable values of the Network parameters, to solve the structure and learning of the WNN. In this research, the GSS method is used to determine a set of best wavelets whose centres and dilation parameters are used as initial values for subsequent training. Experimental tests proved that the proposed approach is very efficient and accurate.

In order to contribute to safe sharing of 3D mul-tiresolution meshes, a new approach of crypto-watermarking is proposed. This approach is based on the use of RSA algorithm and spherical harmonics transform. The aim being is to increase the integration rate while maintaining mesh quality on the one hand and extract correctly inserted image on the other hand. To embed data, host mesh is decomposed using spherical harmonics transform. Resulting coefficients are watermarked twice to insert the grayscale image already encrypted using RSA algorithm. Finally, watermarked mesh is reconstructed through the use on 10% of coefficients already calculated. Found results prove that our approach is able to insert a high amount of data without influencing the mesh quality. The application of the most popular attacks does not prevent a correct extraction of data already inserted. Our algorithm is,then,robust against these attacks.

Fires are the most devastating disasters that the world can face. Thereby, it is crucial to exactly identify fire areas in video surveillance scenes, to overcome the shortcomings of the existing fire detection methods. Recently, deep learning models have been widely used for fire recognition applications. Indeed, a novel deep fire detection method is introduced in this paper. An improved fire model based on tiny-YOLOv3 (You Only Look Once version 3) network is developed in order to enhance the detection accuracy. The main idea is the tiny-YOLOv3 improvement according to the refined proposed training hyperparameters. The generated model is trained and evaluated on the constructed and manually labeled dataset. Results show that applying the proposed training heuristics with the tiny-YOLOv3 network improves the fire detection performance with 81.65% of mean Average Precision (mAP). Also, the designed model outperforms the related works with a detection precision of 97.6%.

High frequency oscillations (HFO) from, MEG (magnetoencephalography) and intracerebral EEG are considered as effective tools to identify cognitive status and several cortical disorders especially in epilepsy diagnosis. The aim of our study is to evaluate stationary wavelet transform (SWT) technique performance in efficient reconstruction of pure epileptic high frequency oscillations, reputed as biomarkers of epileptogenic zones: generators of inter ictal epileptic discharges, and offhand seizures. We applied SWT on simulated and real database to detect non-contaminated HFO by spiky element. For simulated data, we computed the GOF of reconstruction that reaches for all studied constraint (relative amplitude, frequency, SNR and overlap) a promising results. For real data we used time frequency domain to evaluate SWT robustness of HFO reconstruction. We proved that SWT is an efficient filtering technique for separation HFO from spiky events. Our results would have an important impact on the definition of epileptogenic zones.

Locating electromagnetic sources among magnetoencephalography (MEG) allows the definition of responsible generators of excessive discharges in epilepsy. Source localization of MEG biomarkers is considered as a diagnostic aid for neurologists (pre surgical investigation of epilepsy ). Several techniques are proposed to resolve forward and inverse problem of source localization. Our goal in this study is to compare three distributed methods of inverse problem: MNE, sLORETA and dSPM in defining networks connectivity of spiky epileptic events. We used a pre processing chain to evaluate the rate of epileptic spikes connectivity among MEG for five pharmaco resistant patients and two groups of spiky events. For each inverse technique, we calculated the cross correlation between active sources, in fact dsPM shows the highest level of connectivity, MNE gives also a connectivity between the entire active sources but with lowest rate then sLORETA finally dsPM depicts a low number of connection between active regions. These results promote the combination of several localization methods during the investigation of epileptogenics zones.

Fast mesh compression is becoming a requisite in several applications such as medical imaging and video games. Graphics Processing Units (GPUs) are recently becoming massively parallel devices for Single Instruction, Multiple Data (SIMD) computing, addressing hence greater implementation challenges. Transformation and Quantization (TQ) is considered the second highest workload part of the wavelet-based mesh coding. Therefore, its acceleration will further improve the overall processing speed of the coding. In this paper, an OpenCL (Open Computing Language) acceleration of TQ is proposed. The Butterfly Wavelet Transform (BWT) based on the unlifted scheme is adopted in the transformation method while the embedded deadzone quantization is employed for the wavelet quantization. A chunk rearrangement process is applied for the computation of the neighborhood information needed for the Butterfly subdivision stencils. Accordingly, every chunk proceeds independently the prediction of the wavelet coefficients and their quantization. The key insights behind the proposed TQ method on GPU are a smart memory management and an efficient memory data mapping. Extensive experimental assessments demonstrate the effectiveness of our GPU implementation in terms of memory and runtime costs while preserving the rate distortion performance of the state-ofthe-art Bitplane coder.

The current study presents a new 3D mesh deformation process using multi‐mother wavelet neural network architecture, which relies on genetic algorithm and multiresolution analysis. Classic forming algorithms begin with a predetermined network architecture that may be insufficient or too complicated. In addition, the solving of wavelet neural network training problems is described by their perceived inability to escape local optima. The main objective of the authors' proposed approach is that it prevents both the insufficiency and local minima by integrating the genetic algorithm; their wavelet network is used as an approximation tool to align the features of mesh to have efficient deformation processes. Also, such meshes are especially expensive to transmit and are awkward to deform. For this reason, they propose to use multiresolution analysis to decompose a surface geometry into several levels of detail in order to work only with the approximation coefficient at a chosen decomposition level. Hierarchical triangle mesh representations provide access to a triangle mesh at the desired resolution without omitting any information. The experimental results showed the validity of the generalisation ability and the efficiency of their suggested multi‐mother wavelet network architecture based on genetic algorithm and multiresolution analysis for 3D mesh modelling and deformation.

Recently, object recognition has been a very active field of interest. The success of the deep learning based methods in recognizing objects has encouraged recent works to follow this approach. In this paper, we propose a very deep recurrent convolutional neural network approach for object recognition. Our approach uses a very deep convolutional neural network reinforced by integrating recurrent connections to the convolutional layers. Besides, the pooling step has been improved by using two main techniques: the Generalizing Pooling and the spatial pyramid pooling. The Generalizing pooling, that replaces the maxpooling layer commonly used in convolutional neural network, combines pooling operations within a hierarchical tree structure. The Spatial Pyramid Pooling which enables the removal of the fixed size constraint of input image has been conducted. In addition, the data augmentation technique has been used to strengthen the training process. Experiments on three object recognition benchmarks dataset: Pascal VOC 2007, CIFAR-10 and CIFAR-100, have shown the success of our approach.

Remembering an event or a meeting, recalling the face or the name of a person, keeping in mind what we ate or the place of a lost object is sometimes a difficult task. The human memory has its limits. In order to go beyond these limits, researchers developed sensors and wearable cameras to capture individual's experiences. This trend called lifelog has recently been the subject of several panels, workshops and benchmarks. By analyzing the lifelog tasks of these events more closely, we notice that there are still challenges in managing, analyzing, indexing, retrieving, summarizing and visualizing the captured data. In this work, we present a multilevel deep learning-based processing for lifelog image retrieval enhancement. Our proposed approach is based on five phases in which we use deep learning at several levels. The first phase consists of data pre-processing based on low-level image features to filter out irrelevant, noisy and blurred images. In the second phase, we detect and cross high-level image features using pre-trained CNN to enhance the metadata image description. Then, we manage a semantic segmentation based on the WU-Palmer measure similarity. This segmentation is performed to limit the search area and to control better the runtime and the complexity. The fourth phase consist in analyzing the query using LSTM to match concepts with queries. The final phase which based on doc2sequence aims at retrieving the images that is answering the query.

Despite a great deal of progress in face recognition technologies, current solutions are still vulnerable to spoof attacks. In fact, it is easy to access digital replicas of facial biometric information from readily available photos, videos and 3D masks. The literature contains several face anti spoofing methods that try to detect whether the face in the front of the recognition system is real or an artificial replica. However, these methods are not robust and require many improvements since they are sensitive to lightening conditions and pose variations. In order to address these issues, we propose a novel face anti spoofing method based on Multi Color Convolutional Neural Network (CNN) architecture named DeepColorFASD. Our approach investigates the effect of space colors (RGB, HSV and Y CbCr) on CNN architectures and proposes a fusion based voting method for face anti spoofing. In addition, we also explain the resulting feature maps visualizations. We evaluate our system through an experimental study using CASIA FASD: a well-known face anti spoofing database. The results using this challenging database demonstrate that our solution performs better than recent works as measured by Half Total Error Rate (HTER) and ROC curve.

Recently, a variety of approaches have been enriching the field of remote sensing (RS) image processing and analysis. Unfortunately, existing methods remain limited to the rich spatiospectral content of today's large data sets. It would seem intriguing to resort to deep learning (DL)-based approaches at this stage with regard to their ability to offer accurate semantic interpretation of the data. However, the specificity introduced by the coexistence of spectral and spatial content in the RS data sets widens the scope of the challenges presented to adapt DL methods to these contexts. Therefore, the aim of this paper is first to explore the performance of DL architectures for the RS hyperspectral data set classification and second to introduce a new 3-D DL approach that enables a joint spectral and spatial information process. A set of 3-D schemes is proposed and evaluated. Experimental results based on well-known hyperspectral data sets demonstrate that the proposed method is able to achieve a better classification rate than state-of-the-art methods with lower computational costs.

Speaker emotion recognition is considered among the most challenging tasks in recent years. In fact, automatic systems for security, medicine or education can be improved when considering the speech affective state. In this paper, a twofold approach for speech emotion classification is proposed. At the first side, a relevant set of features is adopted, and then at the second one, numerous supervised training techniques, involving classic methods as well as deep learning, are experimented. Experimental results indicate that deep architecture can improve classification performance on two affective databases, the Berlin Dataset of Emotional Speech and the SAVEE Dataset Surrey Audio-Visual Expressed Emotion.

With the rapid development of Remote Sensing acquisition techniques, there is a need to scale and improve processing tools to cope with the observed increase of both data volume and richness. Among popular techniques in remote sensing, Deep Learning gains increasing interest but depends on the quality of the training data. Therefore, this paper presents recent Deep Learning approaches for fine or coarse land cover semantic segmentation estimation. Various 2D architectures are tested and a new 3D model is introduced in order to jointly process the spatial and spectral dimensions of the data. Such a set of networks enables the comparison of the different spectral fusion schemes. Besides, we also assess the use of a “noisy ground truth” (i.e. outdated and low spatial resolution labels) for training and testing the networks.

Deep Convolutional Neural Network (DCNN) can be marked as a powerful tool for object and image classification. However, the training stage of such networks is highly consuming in terms of storage space and time. Also, the optimization is still a challenging subject. In this paper, we propose a fast DCNN based on smart dropout and layer skipping. The proposed approach led to improve the speed of the testing stage as well as image classification accuracy. This was possible thanks to three key advantages: First, the rapid way to compute the features using Fast Beta Wavelet Transform. Second, the proposed intelligent dropout method is based on whether or not a unit is efficiently and not randomly selected. Third, it is possible to classify the image using efficient units of earlier layer(s) and skipping all the subsequent hidden layers directly to the output layer. Our experiments were performed on CIFAR-10 and MNIST datasets and the obtained results are very promising.

Serious games have been used for several years in order to offer continuous and professional training to the companies employees. These games have unevenly affected different fields, they are less present in certain areas (industry) than others (medical, service, …) because of inherent specificity. In this article, we propose an approach to facilitate the implementation of a playful scenario dedicated to industrial trainings. This approach is based on multifaceted resource modelling. It is also illustrated by two examples of serious games: in a steel manufacturing company and in the medical domain.

Serious games are becoming an increasingly used alternative in technical/professional/academic fields. However, scenario development poses a challenging problem since it is an expensive task, only devoted to computer specialists (game developers, programmers…). The ultimate goal of our work is to propose a new scenario-building approach capable of ensuring a high degree of deployment and reusability. Thus, we will define in this paper a new generation mechanism. This mechanism is built upon a model driven architecture (MDA). We have started up by enriching the existing standards, which resulted in defining a new generic meta-model (CIM). The resulting meta-model is capable of describing and standardizing game scenarios. Then, we have laid down a new transformational mechanism in order to integrate the indexed game components into operational platforms (PSM). Finally, the effectiveness of our strategy was assessed under two separate contexts (target platforms) : the claroline-connect platform and the unity 3D environment.

Recognizing the same person across multiple potentially non-overlapping cameras, known as Person re-identification, is a fundamental challenging task in Computer Vision. This is due to the important challenges that it proposes, like large view angle, pose, background clutter and occlusion and low resolution. Most of existing approaches rely on brute-force matching between pedestrian local descriptors and thus suffer from low computational efficiency. To address this issue, we present a new perspective for person re-identification based on a histogram encoding scheme [1] that assigns a global signature to each pedestrian image and thus simplifies the matching process. For that, an extended weighted version of the traditional Fisher vector encoding scheme is proposed. This is achieved by incorporating the Topological location of the encoded descriptors in the encoding process. Thus, two main contributions are proposed. (1) By designing a super Fisher vector representation, we aim to improve both the rate and the speedup of the person matching process. (2) By weighting the Fisher vector encoding scheme, we aim to remove noisy and busy background clutters surrounding a person throughout the Topological weight. Experimental results made on two challenging datasets, the VIPeR dataset, the CUHK03 dataset and the Market-1501 dataset, prove the effectiveness of the proposed method.

The MOOCs (Massive Open Online Courses) represent a category in the frame of TEL (Technology Enhanced Learning) particularly fashionable today since they allow the largest number of learners to access specific teachings. However, the principle of proposing very sequential and linear pedagogical paths is not attractive enough. In fact, the low success rate shows that it is necessary to maintain the motivation and attractiveness of the learner according to his learning progress. In this context, we propose in this article, an approach concerning paths' flexibility within a MOOC that can meet the needs of the learner. In fact, our proposed approach relies on a retroaction cycle around the learners' global representation (skills, sociological characteristics, practices). The elaborated proposals in this paper were experimented in a new MOOC platform: Claroline Connect.

Speech and music discrimination task is considered among the most important tools in several multimedia applications. In this paper, we propose a twofold approach for speech/music discrimination: in the first side, we consider a relevant and rich set of features then in the second one, we adopt mid-term level statistics. The used set of features involves musical descriptors as well as speech and cepstral descriptors. According to the retrieved results, standard deviation metric was elected as the best mid-term level statistic parameter. The SVM classifier has provided the higher accuracy value among the set of employed classifiers. And when combined to the standard deviation statistical parameter, it has reached a satisfying accuracy percentage higher than 99%. Thus, the proposed scheme has achieved promising classification performance thanks to the discriminating abilities and diversity of the used features besides to the statistics on mid term level.

The MOOCs are environments that permit a massive provision of mediatized contents. Despite the success of this phenomenon, the course drop out rate remains very high. In this paper, we want to address the issue of the attractiveness of these contents, because even though they are rich, the proposed resources are rarely complex. However, the provision of complex resources, allowing a practical experience, facilitates the acquisition of knowledge and the global attractiveness of MOOCs. Most of the TEL standards formalize pedagogical resources which can be complex in their structure but simple in their use. In this article, we propose a modeling of complex resources based on SOA to facilitate their integration in a MOOC platform. The proposed model has been tested in a MOOC platform (Claroline Connect) with two types of complex resources: a remote laboratory and a serious game.

The skyline, defined as the line separating the sky from other objects on the ground, could provide unique and useful information for a variety of applications. This line was used as a key data, especially, for geo-localization and aerial robotic applications. The particular shape and the geometric features of a skyline may be the identity of the landscape itself. The skyline, once well extracted, could show the silhouette of a famous tower, the mountain peaks, or the landscape topography. In this paper, we proposed a geometric description of the extracted skyline from landscapes. Based on some geometric descriptors, we tried to pick up practical measurements for each skyline. The first proposed approach was the straight lines' classification to differentiate between urban and natural landscapes from their horizon line. The second one is the curvature analysis using a Curvature Scale Space descriptor. This descriptor was used to enhance the first one and to distinguish between natural part and buildings in the same skyline. The results obtained from these geometric description tools were very competitive and they will be the inputs for a classification process.

The main purpose of transfer learning is to resolve the problem of different data distribution, generally, when the training samples of source domain are different from the training samples of the target domain. Prediction of salient areas in natural video suffers from the lack of large video benchmarks with human gaze fixations. Different databases only provide dozens up to one or two hundred of videos. The only public large database is HOLLYWOOD with 1707 videos available with gaze recordings. The main idea of this paper is to transfer the knowledge learned with the deep network on a large dataset to train the network on a small dataset to predict salient areas. The results show an improvement on two small publicly available video datasets.

Local feature detection is a fundamental module in several mobile vision applications such as mobile object recognition and mobile visual search. The effectiveness and the efficiency of a local feature detector decide to what extent it is suitable for a mobile application. Over the past decades, several local feature detectors have been developed. In this paper, we are interested in FAST (Features from Accelerated Segment Test) local feature detector for its efficiency. However, FAST detector shows poor robustness against both scale and rotation changes. Therefore, we aim at enhancing FAST robustness against both scale and rotation changes while maintaining good efficiency. To this end, we propose a Scalable and Oriented FAST-based local Feature detector (SOFF). A comprehensive comparison against FAST detector and its variants is performed on benchmark datasets. Experimental results demonstrate that SOFF detector outperforms other FAST-based detectors in many cases. Furthermore, it is efficient to compute, thereby suitable for mobile vision applications.

In recent years, Computer vision has become a very active field. This field includes methods for processing, analyzing, and understanding images. The most challenging problems in computer vision are image classification and object recognition. This paper presents a new approach for object recognition task. This approach exploits the success of the Very Deep Convolutional Neural Network for object recognition. In fact, it improves the convolutional layers by adding recurrent connections. This proposed approach was evaluated on two object recognition benchmarks: Pascal VOC 2007 and CIFAR-10. The experimental results prove the efficiency of our method in comparison with the state of the art methods.

Triangular surface are now widely used for modeling three-dimensional object, since these models are very high resolution and the geometry of the mesh is often very dense, it is then necessary to remesh this object to reduce their complexity, the mesh quality (connectivity regularity) must be ameliorated. In this paper, we review the main methods of semi-regular remeshing of the state of the art, given the semi-regular remeshing is mainly relevant for wavelet-based compression, then we present our method for re-meshing based trust region spherical geometry image to have good scheme of 3d mesh compression used to deform 3D meh based on Multi library Wavelet Neural Network structure (MLWNN). Experimental results show that the progressive re-meshing algorithm capable of obtaining more compact representations and semi-regular objects and yield an efficient compression capabilities with minimal set of features used to have good 3D deformation scheme.

With the great popularity of the photo sharing site Flickr, the research community is involved to produce innovative applications in order to enhance different Flickr services. In this paper, we present a new process for diverse visual suggestions generation on Flickr. We unify the social aspect of Flickr and the richness of Wikipedia to produce an important number of meanings illustrated by the diverse visual suggestions which can integrate the diversity aspect into the Flickr search. We conduct an experimental study to illustrate the effect of the fusion of the Wikipedia and Flickr knowledge on the diversity rate among the Flickr search and reveal the evolution of the diversity aspect through the returned images among the different results of search engines.

3D multiresolution mesh compression systems are still widely addressed in many domains. These systems are more and more requiring volumetric data to be processed in real-time. Therefore, the performance is becoming constrained by material resources usage and an overall reduction in the computational time. In this paper, our contribution entirely lies on computing, in real-time, triangles neighborhood of 3D progressive meshes for a robust compression algorithm based on the scan-based wavelet transform(WT) technique. The originality of this latter algorithm is to compute the WT with minimum memory usage by processing data as they are acquired. However, with large data, this technique is considered poor in term of computational complexity. For that, this work exploits the GPU to accelerate the computation using OpenCL as a heterogeneous programming language. Experiments demonstrate that, aside from the portability across various platforms and the flexibility guaranteed by the OpenCL-based implementation, this method can improve performance gain in speedup factor of 5 compared to the sequential CPU implementation.

Speaker gender identification is considered among the most important tools in several multimedia applications namely in automatic speech recognition, interactive voice response systems and audio browsing systems. Gender identification systems performance is closely linked to the selected feature set and the employed classification model. Typical techniques are based on selecting the best performing classification method or searching optimum tuning of one classifier parameters through experimentation. In this paper, we consider a relevant and rich set of features involving pitch, MFCCs as well as other temporal and frequency-domain descriptors. Five classification models including decision tree, discriminant analysis, nave Bayes, support vector machine and k-nearest neighbor was experimented. The three best perming classifiers among the five ones will contribute by majority voting between their scores. Experimentations were performed on three different datasets spoken in three languages: English, German and Arabic in order to validate language independency of the proposed scheme. Results confirm that the presented system has reached a satisfying accuracy rate and promising classification performance thanks to the discriminating abilities and diversity of the used features combined with mid-level statistics.

Extreme learning Machine is a well known learning algorithm in the field of machine learning. It's about a feed forward neural network with a single-hidden layer. It is an extremely fast learning algorithm with good generalization performance. In this paper, we aim to compare the Extreme learning Machine with wavelet neural networks, which is a very used algorithm. We have used six benchmark data sets to evaluate each technique. These datasets Including Wisconsin Breast Cancer, Glass Identification, Ionosphere, Pima Indians Diabetes, Wine Recognition and Iris Plant. Experimental results have shown that both extreme learning machine and wavelet neural networks have reached good results.

Person re-identification is a fundamental challenging task in Computer Vision that consists on recognizing the same person across multiple potentially non-overlapping cameras. This importance is due to the important challenges that it proposes like pose, background clutter and occlusion, illumination changes and low resolution. Also, most of the existing approaches rely on brute-force matching between pedestrian local descriptors and consequently, suffer from low computational efficiency. So, to address this issues, we present a new perspective for person re-identification based on a histogram encoding scheme that assigns a global signature to each pedestrian image and thus, simplifies the matching process. The main contribution of this paper is the design of an extended weighted version of the traditional Fisher vector (FV) encoding scheme. This is achieved by incorporating the Topological location of the encoded descriptors CN, CHS and 15-d in the encoding process and then combining the obtained Topological weighted histograms in order to form our proposed descriptor. The super Fisher vector representation has improved both the rate and the speedup of the person matching process, while weighting the FV encoding scheme by the Topological weight helped out to remove the noisy and busy background clutters surrounding the pedestrians in the images. Besides, Retinex transform was applied in order to handle the problem of illumination variations. Experimental results made on three challenging datasets, the VIPeR dataset, the CUHK03 dataset and the Market-1501 dataset, prove the effectiveness of the proposed method.

This paper presents a structure of the Wavelet Neural Networks used to classify the DNA sequences. The satisfying performance of the Wavelet Neural Networks (WNN) depends on an appropriate determination of the WNN structure optimization problem. In this paper we present a new method to solve this problem based on Genetic Algorithm (GA) and the Least Trimmed Square (LTS). The GA is used to solve the structure and the learning of the WNN and the LTS algorithm is applied to select the important wavelets. First, the scale of the WNN is managed by using the time-frequency locality of wavelet. Furthermore, this optimization problem can be solved efficiently by Genetic Algorithm as well as the LTS method to improve the robustness. The performance of the Wavelet Networks is investigated by detecting the simulating and the real signals in white noise. The main advantage of this method can guarantee the optimal structure of the WNN. The experimental results have indicated that the proposed method (WNN-GA) with the k-means algorithm is more precise than other methods. The proposed method has been able to optimize the wavelet neural network and classify the DNA sequences. Our goal is to construct a predictive approach that is highly accurate results. In fact, our approach allows avoiding the complex problem of form and structure in different groups of organisms. The experimental results are showed that the WNN-GA model outperformed the other models in terms of both the clustering results and the running time. In this study, we present our system which consists of three phases. The first one is the transformation, is composed of two sub steps; the binary codification of the DNA sequences and the Power Spectrum Signal Processing. The second step is the approximation; it is empowered by the use of the Multi Library Wavelet Neural Networks (MLWNN). Finally, the third one is the clustering of the DNA sequences, is realized by applying the algorithm of the k-means algorithm.

In multimedia distribution platforms, one of the main challenges is to provide an efficient and accurate tracing process despite the lack of information about the colluders' strategy. Indeed, the original Tardos tracing performance is considered as suboptimal because of its agnostic behavior and conservative accusation regardless the collusion strategy. The Expectation Maximization algorithm has shown to be an efficient solution to estimate the collusion channel and thus to tune the Tardos accusation functions. In this paper, we explore the impact of this algorithm in a group-based tracing scheme to deal with the computational costs and the invariance of the Tardos accusation performance. The tracing scheme we propose benefits from a twofold accusation process. Indeed, in a first time, it is based on a two-level tracing strategy which consists in tracing guilty groups in a first level with the Boneh Shaw tracing code and in retrieving at least one colluder in accused groups with Tardos code in the second level. This strategy has reduced efficiently the decoding process of the Tardos code. The main shift we propose in the second level is to apply the Expectation Maximization algorithm to be tightly tied to collusion yielded by colluders and hence to find the more accurate Tardos accusation functions. The performance of the resulting tracing scheme is evaluated according to different criteria and promising results have been achieved when compared to the existing tracing schemes proposed in the literature.

A novel proposed approach, collaborative representation-based classification, has been developed for face recognition and recently used in image classification task owing to its simplicity and effectiveness. The major drawback of this method is the neglect of the spatial structure among the image representations. Inspired by the success of this technique and motivated by the power of spatial information in improving the image representation, we suggest in this paper a novel collaborative approach named spatial collaborative representation based classification. After applying the feature encoding and the pooling method, we exploit the global manifold structure of the image by applying the spatial pyramid representation. After that, two successive steps are required in order to obtain the label category for each image. In the first step we apply the standard collaborative method for each histogram generated at each pyramid level. The second stage aims to combine efficiently the image reconstruction results in order to predict the category label.

Color is a significant visual characteristic for both human vision and computer processing. Global color descriptors characterize an image by its color distribution or histogram, and discard information about object location as well as content of different colors. In this paper, we proposed a local color descriptor based on indexed matrix wavelet analysis and fuzzy decision support system (FDSS), which we called “Fuzzy Indexed Color (FIC)”. First, the indexed map of each image is analysed using Fast wavelet transform to capture the most relevant color feature content for each color channel R, G and B. Then, a FDSS is proposed for image matching in order to get more flexibility and reliability in making decision. The proposed FIC was evaluated using Google color, ebay data, Soccer and Flower datasets; and the results are very promising.

The innovative brand “The internet of Me” is a recent research area that highlight the prevalence of personalization across the internet and focuses on the user habits and actions tracked from his interaction with the web content. This paradigm presents an efficient way to define the user experience, preferences useful in e-commerce, marketing, social and search purpose. In this paper we are interested in the integration of the advances of user profile in information retrieval systems by predicting the user preferences from his user interaction with social content. These data are exploited and modeled to be used for tailoring the query interpretation for the user and delivering more relevant and accurate results for the user. As for as the experiments, our proposed approach of user profile modeling shows promising results applied for the personalization of a defined set of ambiguous queries retained by thirty users having different profiling characteristics.

Developing an automatic arabic sign language recognition system is of great importance, it can be used as a communication means between hearing-impaired and other people.

Recognizing the same person across multiple potentially non-overlapping cameras, known as Person Re-identification, is a fundamental challenging task in Computer Vision. This is due to the important challenges that it proposes, like view point and pose change, background clutter and occlusion, low resolution and illumination variations. Most of the existing approaches rely on brute-force matching between local descriptors and thus suffer from low computational efficiency. Therefore, we present a new perspective for person re-identification based on a histogram encoding scheme. For that, an extended weighted version of the traditional Fisher vector encoding scheme is proposed. This is achieved by incorporating the Salience of the encoded descriptors and their topological location in the encoding process. Experimental results made on three challenging datasets (VIPeR, CUHK03 and Market-1501 dataset), prove the effectiveness of the proposed method.

The success of the deep learning and specifically learning layer by layer led to many impressive results in several contexts that include neural network. This gave us the idea to apply this principle of learning on wavelet network because it is an active research topic at the moment. This paper present our approach for image classification by the combination of two techniques of learning: the wavelet network and the deep learning. We try to classify images in a supervised way following by an unsupervised learning using the principle of autoencoder. Experiments on two databases COIL-100 and MNIST show that our approach gives good results for the two classifiers that we used.

Analysis of material microstructure images have been an important research topic. In fact, multiple pertinent information can be extracted from material texture images such as inclusions, fraction surfaces, heterogeneous components, crystallographic planes orientation, porosity and so on. Actually, this information depends from the amount of noise included in the image. Thus, it is very important to enhance methods exploited for image de-nosing in such cases. In fact, images issued from scanned electronic microscopy as well as computed tomography are used in multiple security applications such as baggage screening and in the encrypted domain, it used to protect privacy of outsourced data in cloud computing. In this paper, we propose a comparative study between two popular techniques used in image de-noising that are; Sure-let wavelet technique and weighted bilateral filtering. Our goal is to grasp the versatility of those methods in de-noising microstructure composite material images. Results of both outputs are well discussed. In our case, we found that sure-let wavelet de-noising gives better output results quantitatively and qualitatively.

Image processing acquires more and more importance because of its wide spread in industrial applications such as; Biometrics, Information Fusion, Image Registration, Image Mosaic, Image Indexing, Retrieval, Image and Video Coding, Motion Detection and Tracing, Feature Extraction and so on. Image de-noising is an important step towards multiple image processing exploitation. In this paper, firstly, we are giving an overview of the most relevant methods applied in image de-nosing. Then, we are describing a bilateral filtering technique and its application to de-noise microstructure images of a metallic composite material. Analysis of de-noised material microstructure images using bilateral filtering techniques is discussed. Results show that using the fast bilateral filtering variant and the robust bilateral filtering method provide good output processed images.

In this paper, we propose a feature-based classification framework for Alzheimer's disease (AD) recognition using Tensor Diffusion Imaging (DTI). The main contribution consists in considering the visual pattern of water molecules diffusion in the most involved region in AD (hippocampal area). We use the Circular Harmonic Functions (CHFs) and the Bag-of-Visual-Words approach to build an AD related-signature. The experiments were accomplished first with a subset of participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and then with the DTI scans of a French epidemiological study: ”Bordeaux-3City”. Experimental results demonstrate that our features-based method applied on the MD maps is able to capture the AD-related atrophy and then classify between AD subjects.

This paper presents a novel hand posture recognizer based on separator wavelet networks (SWNs). Aiming at creating a robust and rapid hand posture recognizer, we have contributed by proposing a new training algorithm for the wavelet network classifier based on fast wavelet transform (FWN). So, the contribution resides in reducing the number of WNs modeling training data. To make that, inspiring from the adaboost feature selection method, we thought to create SWNs (n-1 WNs for n classes) instead of modeling each training sample by its wavelet network (WN). By proposing the new training algorithm, the recognition phase will be positively influenced. It will be more rapid thanks to the reduction of the number of comparisons between test images WNs and training WNs. Comparisons with other works, employing universal hand posture datasets are presented and discussed. Obtained results have shown that the new hand posture recognizer is comparable to previously established ones.

In last years, the emergence of 3D shape in face recognition is due to its robustness to pose and illumination changes. These attractive benefits are not all the challenges to achieve satisfactory recognition rate. Other challenges such as facial expressions and computing time of matching algorithms remain to be explored. In this context, we propose our 3D face recognition approach using 3D wavelet networks. Our approach contains two stages: learning stage and recognition stage. For the training we propose a novel algorithm based on 3D fast wavelet transform. From 3D coordinates of the face (x,y,z), we proceed to voxelization to get a 3D volume which will be decomposed by 3D fast wavelet transform and modeled after that with a wavelet network, then their associated weights are considered as vector features to represent each training face . For the recognition stage, an unknown identity face is projected on all the training WN to obtain a new vector features after every projection. A similarity score is computed between the old and the obtained vector features. To show the efficiency of our approach, experimental results were performed on all the FRGC v.2 benchmark.

Driving security is an important task for human society. The major challenge in the field of accident avoidance systems is the driver vigilance monitoring. The lack of vigilance can be noticed by various ways, such as, fatigue, drowsiness and distraction. Hence, the need of a reliable driver’s vigilance decrease detection system which can alert drivers before a mishap happens. In this paper, we present a novel approach for vigilance estimation based on multilevel system by combining head movement analysis and eyes blinking. We have used Viola and Jones algorithm to analyse head movement and a classification system using wavelet networks for eyelid closure measuring. The contribution of our application is classifiying the vigilance state at multi level. This is different from the binary-class (awakening or hypovigilant state) existing in most popular systems.

Image classification is an important task within the field of computer vision. In this paper we propose a new wavelet network classifier (WNC) based on the cascaded architecture. This classifier is characterized by its new learning approach and its novel architecture which brings a novel robust test way. So, our contributions in this paper reside in two major points. The first one is the proposition of a new training algorithm which overcomes lacuna detected in the latest version of WN learning approach. Hence, our new approach creates separator WNs discriminating classes (n -- 1 WNs to classify n classes) instead of creating a WN for each training image. This contribution makes very rapid the classification process by reducing the number of comparisons between test images WNs and training WNs. The second contribution is the proposition of a novel architecture which brings a new test approach radically different to those employed in ancient WN versions. By the new architecture which is based on the cascade notion, we aim at reducing the number of kernels employed in the approximation of test images. Experiments, using well known benchmarks, show that our new classifier is very robust and rapid compared to already existing ones.

The problem addressed in this paper is the classification and recognition of Arabic handwritten characters. As a solution, we present a Neural Network (NN) architecture based on Fast Wavelet Transform (FWT) and Adaboost algorithm. FWT is used to extract character's features, based on Multi-Resolution Analysis (MRA) at different levels of abstraction. These features are used to calculate inputs of hidden layer. After this first step, the features are filtered, using Adaboost algorithm, to select the best corresponding ones to each shape of input characters. The reported results are tested on Arabic handwritten characters dataset with 6000 characters. The classification rate for the different groups of characters are 93.92%. Additionally, the speed of the classification algorithm is tested and reported.

Sparse coding has proved its efficiency in the image classification task. However, its major drawback is the discarding of the spatial context information that can be extracted from the image. Therefore, we propose in this work a novel sparse coding method called Laplacian sparse coding based on the integration of topological information in the encoding process. This is achieved by embedding the similarities between local region visual phrases into the objective function of the classical Laplacian sparse coding. Experimental results made on several datasets prove the efficiency of the proposed method.

Social and Mobile are the two very characterizing trends of the Internet. Subsequently, the volume of photos with rich social, textual and contextual information increases exponentially either on mobile devices or social networks. Performing an efficient and effective mobile Image Search over social photo collection is therefore a crucial challenge. Indeed, capture the complex connections among social photos is as important as speeding up similarity search at large scale. This paper present a generic Mobile Image Search framework with hypergraph hashing. On the mobile side, users are enabled to formulate whether visual, textual or vocal queries. On the server side, we start by modeling complex connections that may exist among photos and social features using an hypergraph. To accelerate the nearest neighbor search over the hypergraph, a spectral hashing is performed. Namely, each hypergraph vertex is mapped to a binary string without loss of similarity. For unseen items in the hypergraph, a query-adaptive supervised learning is carried out to learn binary strings based on the query type. We report the initial results over NUS-WIDE collection which show that the proposed framework is promising in the field of Mobile Image Search.

According to the ever development of multimedia distribution systems, more than one technique was proposed in the literature to address the copyright protection issue. One key technique was to propose a fingerprinting system based on traitor tracing process to retrieve back the traitorous users who can operate in the mid-way. Some previous works agree upon the fact that users belonging to the same group have more probability to collude together. Several researchers in the tracing traitor field agree upon the fact that constructing a group-based fingerprint should enhance the detection rates of the fingerprinting system. In this paper, we propose to generate a fingerprint having the group property by using a clustering algorithm. We propose to construct a group-based fingerprint according to a DCT-based audio watermarking technique which has proven good robustness and inaudibility results. To show the impact of the classifying algorithm, a set of experimental tests are conducted to check two relevant criteria: the capacity and the security of the group-based fingerprint.

This paper proposes a face detection system based on the skin color, the Gabor filter and the neural network. The use of Gabor filters and neural networks for face recognition is not new. However, the principal focus of the proposed paper is the implementation of skin color selection prior to Gabor filters and neural networks on order to reduce computation time. First, we analyze the skin color to extract skin areas which have an important probability to be faces. This technique robust to the lighting variation allows extracting, from an image, skin areas. We utilize this method to avoid wrong detection and to help the system detect the face in the right areas and minimize the research time. Second, to extract features, we propose a technique using the Gabor filter applied on the localized skin space. Finally, the vectors of the face features obtained by the Gabor filter are used as the input of a neural network classifier which classifies an input image pixel as a face or nonface pixel. Some results are shown to approve our approach efficiency.

Speech recognition is a specialized pattern recognition task with several applications such as vocal command system, dictating machines, and understanding systems. In recent years, research on pattern recognition has increased by developing various methods and algorithms for different applications. In this paper, we proposed a novel training algorithm based on the fast Beta wavelet transform for speech recognition. This approach has many advantages compared to other algorithms. The majority of the old approaches need to inverse matrix, which can be computationally intensive. However, the new algorithm is computed by the iterative application of fast wavelet transform to compute connection weights. To highlight our approach, we compared its experimental results to those of the old ones.

In this paper, we present a comparative study between some well-known knowledge structures, applied in the domain of textual query disambiguation, which are Wikipedia Miner, WordNet and BabelNet. We provide an idea about our proposed approach: online Wikipedia-based query disambiguation. Based on different types of the Wikipedia pages, the proposed approach shows promising results when testing a set of thirty ambiguous queries.

In the last decade, the extensive evolution of 3D graphic applications has induced developpers to enhance the 3D mesh compression techniques. Therefore, 3D models are compressed, transmitted and rendered more and more in real-time and with high quality. Moreover, many out-of-core algorithms are proposed to process complex objects. In this paper, we review the major existing technologies for the single-rate and progressive 3D mesh compression. Then, representative out-of-core approaches are surveyed in details. Finally, we represent some parallel schemes exploiting the evolution of many-core GPU architecture.

Due to the incessant explosion of the multimedia documents amount, the use of metadata is becoming crucial to facilitate the retrieval and the management of these audiovisual contents. Metadata creation is highly time and resources consuming even if the process is automatically done. Thus, video browsing systems uses existing metadata files generally jointed to the corresponding video for efficient semantic multimedia content retrieval. However, missing the metadata file renders the related video useless. So, in this paper, a new strategy for video characterization is described by embedding the metadata information using a blind watermarking technique. Consequently, browsing systems can use the beforehand indexed content just by extracting the corresponding metadata.

Face emotion recognition is one of the most important and rapidly advanced active research areas of computer science. A new method for facial expression recognition based on wavelet network classifier is proposed in this paper. It allows us the detection of six basic emotions other than the neutral one: (Joy, surprise, sadness, anger, fear and disgust) The process is composed of three principle steps: face detection, features extraction and classification. The effectiveness of our proposed algorithm is experimentally demonstrated by using well-known test database: the extended cohen-kanade database.

This paper presents the structured Fisher vector encoding method, a new video representation which yields an improved model to classical FV for human action recognition. Our proposed representation is based on local structural organization of features by building graphs of trajectories. It preserve more information in feature encoding process by local spatial pooling and refining the representation in the global pooling. Local spatio-temporal information are exploited by presenting the relationships among video trajectories as local graphs of trajectories using a multi-scale Delaunay triangulation. Experiments using the human action recognition datasets (Hollywood2 and HMDB51) show the effectiveness of the proposed approach.

Road safety is an important challenge for human society. Fatigue at the wheel is a serious problem causing thousands of road accidents each year. The major issue in the field of driving security systems is driver vigilance control. In this paper, we present a new method for driver's vigilance level measurement using a multi parameter system based on head movement estimation, eyes blinking analysis and face emotion recognition based on wavelet networks classification system.

The bag of words approach describes an image as a histogram of visual words. Therefore, the structural relation between words is lost. Since graphs are well adapted to represent these structural relations, we propose, in this paper, an image classification framework which draws benefit from the eff iciency of the graph in modeling structural information and the good classification performances given by the bag of words method. For each image in the dataset, a graph is created by modeling the spatial relations between dense local patches. Thus, we obtain a graph dataset. From the graph dataset, we select the most frequent subgraphs to construct the bag of subgraphs (BoSG) and we associate to each image a subgraph histogram that describes its visual content. For experiments, we have used the two challenging datasets: 15 Scenes and Pascal VOC 2007. Experimental results show that the proposed method outperforms the bag of words and the spatial pyramid models in terms of recognition rate.

Modelling spatial information of local features is known to improve performance in image categorisation. Compared with simple pairwise features and visual phrases, graphs can capture the structural organisation of local features more adequately. Besides, a dense regular grid can guarantee a more reliable representation than the interest points and give better results for image classification. In this study, the authors introduced a bag of dense local graphs approach that combines the performance of bag of visual words expressing the image classification process with the representational power of graphs. The images were represented with dense local graphs built upon dense scale‐invariant feature transform descriptors. The graph‐based substructure pattern mining algorithm was applied on the local graphs to discover the frequent local subgraphs, producing a bag of subgraphs representation. The results were reported from experiments conducted on four challenging benchmarks. The findings show that the proposed subgraph histogram improves the categorisation accuracy.

Over the last years, relevance re-ranking has been an attractive research, aiming to re-order the initial image search result list by which relevant ones should be at the top ranking list and irrelevant ones should be pruned. In this paper, we propose to explore two population-based meta-heuristic algorithms, which are Particle Swarm optimization(PSO), and Cuckoo search(CS), in order to solve the relevance re-ranking problem as a constrained regularisation framework. By doing so, we define two reranking processes, refereed as APSO-Rank and CS-Rank that converge to the optimal ranked list. Results are further provided to demonstrate the effectiveness and performance of these two reranking processes.

This paper focuses on the issue of emotion recognition. It describes an emotion recognition system based on facial expression which contains four steps: detection of face's elements, localization of feature points, their tracking during a movie and facial expression classification. The first step is realized by the famous Viola and Jones algorithm. To localize feature points we have developed an automatic and easy method. To track them we used the optical flow. Finally the classification step is based on wavelet network using Fast Wavelet Transform FWT. The experimental results demonstrated the efficiency of our system.

Recognizing video events has been a very active field of interest. The diversity of videos captured in complex environments and under difficult conditions makes the event recognition a challenging task. In this paper, we present a video event recognition method which exploits the power of graphs for representing the structural organization of the features and the success of the Bag-of-Words approach. Our method combines the Scale Invariant Feature Transform and the Space-Time Interest Point features to characterize the video. To model the spatio-temporal relations among these features, a graph-based representation is used for each video. Then, the video is indexed based on a histogram of frequent sub-graphs. To evaluate our method, we have used the Columbia Consumer Video dataset. The experimental results show the efficiency of the proposed method.

This paper presents an improved approach to face recognition, called Regularized Shearlet Network (RSN), which takes advantage of the sparse representation properties of shearlets in biometric applications. One of the novelties of our approach is that directional and anisotropic geometric features are efficiently extracted and used for the recognition step. In addition, our approach includes a module based on regularization theory (RSN) to control the trade-off between the fidelity to the data (gallery) and the smoothness of the solution (probe). In this work, we address the challenging problem of the single training sample per subject (STSS). We compare our new algorithm against different state-of-the-arts method using several facial databases, such as AR, FERET, FRGC, FEI, CK. Our tests show that the RSN approach is very competitive and outperforms several standard face recognition methods.

To generate the visual codebook, a step of quantization process is obligatory. Several works have proved the efficiency of sparse coding in feature quantization process of BoW based image representation. Furthermore, it is an important method which encodes the original signal in a sparse signal space. Yet, this method neglects the relationships among features. To reduce the impact of this issue, we suggest in this paper, a Laplacian Tensor sparse coding method, which will aim to profit from the relationship among the local features. Precisely, we propose to apply the similarity of tensor descriptors to create a Laplacian Tensor similarity matrix, which can better present in the same time the closeness of local features in the data space and the topological relationship among the spatially near local descriptors. Moreover, we integrate statistical analysis applied to the local features assigned to each visual word in the pooling step. Our experimental results prove that our method prevails or exceeds existing background results.

We propose in this paper a spatio-temporal pyramid representation (STPR) of the video based Accordion image. The Accordion image allows the pixels having a high temporal correlation to be put in space adjacency. The STPR introduces spatial and temporal layout information to the local SIFT features computed on the Accordion image. It consists in applying firstly, a temporal pyramid decomposition on the video to divide it into a sequence of increasingly finer temporal blocks and secondly in performing a spatial pyramid representation on the Accordion images relative to the temporal blocks. The Multiple Kernel Learning approach is used to combine the multi-histograms coming from different Spatio-Temporal Pyramid levels. Experiments using the human action recognition datasets (Hollywood2 and Olympic sports) show the effectiveness of the proposed approach.

This paper investigates a new approach for face, gender and race classification, called multi-regularized learning (MRL). This approach combines ideas from the recently proposed algorithms called multi-stage learning (MSL) and multi-task features learning (MTFL). In our approach, we first reduce the dimensionality of the training faces using PCA. Next, for a given a test (probe) face, we use MRL to exploit the relationships among multiple shared stages generated by changing the regularization parameter. Our approach results in convex optimization problem that controls the trade-off between the fidelity to the data (training) and the smoothness of the solution (probe). Our MRL algorithm is compared against different state-of-the-art methods on face recognition (FR), gender classification (GC) and race classification (RC) based on different experimental protocols with AR, LFW, FEI, Lab2 and Indian databases. Results show that our algorithm performs very competitively.

This work is in the field of human-computer communication, namely in the field of gestural communication. The objective was to develop a system for gesture recognition. This system will be used to control a computer without a keyboard. The idea consists in using a visual panel printed on an ordinary paper to communicate with a computer.

This paper aims at developing a novel approach for speech recognition based on wavelet network learnt by fast wavelet transform (FWN) including a fuzzy decision support system (FDSS). Our contributions reside in, first, proposing a novel learning algorithm for speech recognition based on the fast wavelet transform (FWT) which has many advantages compared to other algorithms and in which major problems of the previous works to compute connection weights were solved. They were determined by a direct solution which requires computing matrix inversion, which may be intensive. However, the new algorithm was realized by the iterative application of FWT to compute connection weights. Second, proposing a new classification way for this speech recognition system. It operated a human reasoning mode employing a FDSS to compute similarity degrees between test and training signals. Extensive empirical experiments were conducted to compare the proposed approach with other approaches. Obtained results show that the new speech recognition system has a better performance than previously established ones.

This paper presents an improved approach to face recognition, called Regularized Shear let Network (RSN), that takes advantage of the sparse representation properties of shear lets in biometric applications. The main novelty of our approach is the efficient extraction of geometric features based on the properties of the shear let decomposition, a multiscale directional method which is especially designed to capture directional and anisotropic information in multidimensional data. To further improve the performance of our face recognition algorithm, we include a regularization step to control the trade-off between the fidelity to the data (gallery) and smoothness of the solution (probe). In this work, we focus on the challenging problem of the single training sample per subject (STSS). We compare our new algorithm against different state-of-the-arts method using several facial databases including AR, FERET, FRGC, FEI and CK Our tests show that our RSN algorithm is very competitive and outperforms several state-of-the-art face recognition methods.

This paper presents a novel approach for face recognition (FR) based on a new multiscale directional approach, called Shear let Network (SN), and on a recently emerged machine learning paradigm, called Multi-Task Sparse Learning (MTSL). SN aims to extract anisotropic features from an image in order to efficiently capture the facial geometry (shear face), MTSL is used to exploit the relationships among multiple shared tasks generated by changing the regularization parameter to make the optimization convex. We compare our algorithm, called Sparse Multi-Regularized Shear let Network (SMRSN), against different state-of-the-art methods on different experimental protocols with AR, ORL, LFW, FERET, FRGC v1 and Lab2 databases. Our tests show that the SMRSN approach yields a very competitive performance and outperforms several standard methods of FR.

This paper focuses on the study of modified constructive training algorithm for Multi Layer Perceptron “MLP” which is applied to face recognition applications. In general, constructive learning begins with a minimal structure, and increases the network by adding hidden neurons until a satisfactory solution is found. The contribution of this paper is to increment the output neurons simultaneously with incrementing the input patterns. In fact, the proposed algorithm started with a small number of output neurons and a single hidden-layer using an initial number of neurons. During neural network training, the hidden neurons number is increased while the Mean Square Error “MSE” threshold of the Training Data “TD” is not reduced to a predefined parameter. The output neurons number is increased as the input patterns are incrementally trained until all patterns of Training Data “TD” are presented and learned. The proposed algorithm is applied in the classification stage in face recognition system. For the feature extraction stage, a biological vision-based facial description, namely Perceived Facial Images “PFI” is applied to extract features from human face images. The proposed approach is tested on the Cohn-Kanade Facial Expression Database. Compared to the fixed “MLP” architecture and the constructive training algorithm, experimental results clearly demonstrate the efficiency of the proposed algorithm.

This paper presents a new cascaded hybrid Wavelet Network Classifier (CHWNC) designed for hand gesture recognition in real time applications. This paper contains two key contributions. The first is the amelioration of our previous works in the classification domain employing wavelet networks (WN). Precisely, by ameliorating the training way of the latest wavelet network classifier (WNC) version by representing each training class by one WN instead of creating a WN for each training image. This contribution makes very rapid the test phase by reducing the number of comparisons between test images WNs and training WNs. The second contribution is the proposition of a new wavelet network architecture including the cascade notion which decomposes the WN on a set of stages. The new architecture has as aim not only to make recognitions robust and rapid but also to reject as fast as possible gestures which must not be considered by the system (spontaneous gestures). Experiments, based on a well known hand posture dataset, show that our method is very robust and rapid compared to already existing ones.

The instrumentation and physiological patient factors bound to the patient in-vivo complicate the treatment of the images during a medical examination. Thus, they can contribute to the generation of artifacts in the resulting images. The artifacts degrade the quality of the images and can lead, in certain cases, to a bad diagnosis. For this reason, the appeal to the use of the simulation techniques allows to evaluate and improve the devices of acquisition and image processing. The models (called phantoms) are important tools to simulate human anatomy and physiology and to allow the evaluation of acquisition methods and image analysis. Thereby, the simulation offers a way of great importance to evaluate and improve medical techniques of acquisition devices, treatment and the reconstruction of images in-vitro.

This study presents a modified constructive training algorithm for multilayer perceptron (MLP) which is applied to face recognition problem. An incremental training procedure has been employed where the training patterns are learned incrementally. This algorithm starts with a small number of training patterns and a single hidden‐layer using an initial number of neurons. During the training, the hidden neurons number is increased when the mean square error (MSE) threshold of the training data (TD) is not reduced to a predefined value. Input patterns are trained incrementally until all patterns of TD are learned. The aim of this algorithm is to determine the adequate initial number of hidden neurons, the suitable number of training patterns in the subsets of each class and the number of iterations during the training step as well as the MSE threshold value. The proposed algorithm is applied in the classification stage in face recognition system. For the feature extraction stage, this paper proposes to use a biological vision‐based facial description, namely perceived facial images, applied to extract features from human face images. Gabor features and Zernike moment have been used in order to determine the best feature extractor. The proposed approach is tested on the Cohn‐Kanade Facial Expression Database. Experimental results indicate that a good architecture of neural network classifier can be obtained. The effectiveness of the proposed method compared with the fixed MLP architecture has been proved.

In this study, the authors propose an image authentication algorithm in the DCT domain based on neural networks. The watermark is constructed from the image to be watermarked. It consists of the average value of each 8 × 8 block of the image. Each average value of a block is inserted in another supporting block sufficiently distant from the protected block to prevent simultaneous deterioration of the image and the recovery data during local image tampering. Embedding is performed in the middle frequency coefficients of the DCT transform. In addition, a neural network is trained and used later to recover tampered regions of the image. Experimental results shows that the proposed method is robust to JPEG compression and can also not only localise alterations but also recover them.

This paper presents a virtual object control method of augmented reality scene. We have based on control approach on speech recognition. The idea came from human-machine interaction. The speech recognition system is based on wavelet network. In this paper, we have briefly described the used toolkit to do with the augmented reality. Then, we present the speech recognition approach the training and recognition approach. Finally, we present the results.

The encroachment of Internet and Peer to Peer networks has really facilitated our daily lives and works but it contributes to another dangerous phenomenon which is copying a digital content without having authorization, called piracy. To handle this phenomenon, several techniques of tracing traitors were proposed, by combining in the same time a fingerprinting technique to a watermarking one. In this paper, we first present basic notions for multimedia traceability framework: Anti Collusion code (ACC) and the watermarking technique. We show a study of available tracing traitors' schemes and we propose a comparison of accusation ability and computational costs of these techniques. Next we describe our future contribution in this target.

Elastic Bunch Graph Matching EBGM is a face recognition algorithm that is distributed with CSU's Evaluation of Face Recognition Algorithms System. The algorithm recognizes novel faces by first localizing a set of landmark features and then measuring similarity between these features. Both localization and comparison uses Gabor jets extracted at landmark positions. In order to improve the performance of the face recognition system[7][8], we have associated Beta filters to the EBGM technique. This choice of Beta filters is advanced by the performance of these functions in many applications of classification and pattern recognition.

In this paper we describe an automated process of object detection and recognition in real time video scene having a fix background. The objective is to provide a real time system which can detect and recognize any new object introduced in the video. This detection is performed based on an adaptive Gaussian Model which re-estimates the background model permanently. Each detected object is then described with a local detector and descriptor of key points SURF chosen as it is invariant, robust and distinctive. Finally the proposed process leads to a highly performed identification in a database of images based on the structure of the Vocabulary Tree. This paper presents also experimental results to evaluate the performance of the algorithms which confirm the high performance and the robustness of our approach.

Illegal copying of multimedia files has become a very common practice. Indeed, with the rapid development of means of communication, sharing, copying and illegal downloading have become a very easy handling action, at everybody's reach. The magnitude of this continuously increasing phenomenon may have a significant economic impact since it induces a marked loss on turnover. To cope with this huge problem, it becomes necessary to control video traffic and ensure traceability. Thus, each user receives a personalized media release, containing a personal identifier inserted through a robust watermarking technique. If this copy is redistributed illegally, we are able to trace the dishonest user who can be prosecuted. This expresses an urgent need for implementing a reliable fingerprinting scheme with high performances. In this context, we present in this paper a hierarchical fingerprinting system based on Tardos code in order to reduce computational costs required for the pirates' detection. Both theoretical analyses and experimental results are provided to show the performance of the proposed system.

A commonly accepted fact in the biometrics related domain is that fusing multiple classifiers for decision making generally leads to improved recognition performance. Meanwhile, the search for an optimal fusion strategy remains extraordinarily complex since the cardinality of the space of possible fusion schemes is exponentially proportional to the number of competing classifiers. In this paper, we propose a mixture of gated experts for 3D face recognition using an ensemble of 24 different classifiers. The mixture of gated experts is optimized using a Simulated Annealing-based algorithm. It automatically selects and fuses the most relevant similarity measurements. The experimental results of 3D face recognition achieved on the FRGC v2.0 dataset illustrate the effectiveness and stability of the proposed method. Additionally, as a learning-based method, it also has a good robustness to the variations of training database.

In this paper, a novel learning algorithm of wavelet networks based on the Fast Wavelet Transform (FWT) is proposed. It has many advantages compared to other algorithms, in which we solve the problem in previous works, when the weights of the hidden layer to the output layer are determined by applying the back propagation algorithm or by direct solution which requires to compute the matrix inversion, this may cause intensive computation when the learning data is too large. However, the new algorithm is realized by iterative application of FWT to compute the connection weights. Furthermore, we have extended the novel learning algorithm by using Levenberg–Marquardt method to optimize the learning functions. The experimental results have demonstrated that our model is remarkably more refreshing than some of the previously established models in terms of both speed and efficiency.

This paper deals with an experiment which proves that wavelet networks are capable for 3D objects modeling. To prove this, we will propose a new structure of wavelet network founded on several mother wavelets families. This new structure is in some ways similar to the classic wavelet networks but it admits some originality. Actually, wavelet network basically uses dilations and translations versions of only one mother wavelet to construct the network. The proposed structure uses several mother wavelets, in order to maximize best wavelets selection probability. An algorithm to construct this structure is presented. First, 3D object model vertices and their corresponding normal values are used to create a training set. Then, an improved Orthogonal Least Squares method version is applied to optimize wavelet selection for every mother wavelet. Some simulation results will describe the proposed wavelet network performance employing several types of Polywogs as mother wavelets.

Classifier fusion is considered as one of the best strategies for improving performance of general purpose classification systems. On the other hand, fusion strategy space strongly depends on classifiers, features and data spaces. As the cardinality of this space is exponential, one needs to resort to a heuristic to find a sub-optimal fusion strategy. In this work, we present a new adaptive feature and score level fusion strategy (AFSFS) based on adaptive genetic algorithm. AFSFS tunes itself between feature and matching score level, and improves the final performance over the original on both levels, and as a fusion method, it does not only contain fusion strategy to combine the most relevant features so as to achieve adequate and optimized results, but also has the extensive ability to select the most discriminative features. Experiments are provided on the FRGC database showing that the proposed method produces significantly better results than the baseline fusion methods.

Image classification is an important task in computer vision. In this paper, we propose a supervised method for image classification based on a fast beta wavelet networks (FBWN) model. First, the structure of the wavelet network is detailed. Then, to enhance the performance of wavelet networks, a novel learning algorithm based on the Fast Wavelet Transform (FWTLA) is proposed. It has many advantages compared to other algorithms, in which we solve the problem of the previous works, when the weights of the hidden layer to the output layer are determinate by applying the back propagation algorithm or by direct solution which requires to compute matrix inversion, this may be intensive computation when the learning data is too large. However, the new algorithm is realized by the iterative application of FWT to compute connection weights. In the simulation part, the proposed method is employed to classify images. Comparisons with classical wavelet network classifier are presented and discussed. Results of comparison have shown that the FBWN model performs better than the previously established model in the context of training run time and classification rate.

3D shape of face has recently emerged as a major research in face biometrics. However, while it is reputed to be relatively invariant to lighting conditions and pose, one still needs to cope with facial expression variations for a reliable face recognition solution and running time of the matching algorithms for fast identification software. We present in this paper our solutions to overcome these limitations. We propose a new method of 3D facial recognition based on wavelet networks. Firstly, depth image is preprocessed in order to crop the useful area of the face image. Secondly, a compact and representative biometric signature is produced by means of wavelet networks. Finally, the matching of two faces is made by computing Euclidean distance between their two corresponding signatures. To show the efficiency and accuracy of our approach, a subset taken from FRGC v2 dataset is used to made evaluations.

This paper describes a blind video watermarking system invariant to geometrical attacks. Our scheme embeds different parts of a single watermark into different shots of a video under the wavelet domain. A multi resolution motion estimation algorithm (MRME) is adopted to preferentially allocate the watermark to coefficients containing motion. In addition, embedding and extraction of the watermark are based on the relationship between a coefficient and its neighbors. Experimental results show that inserting watermark where picture content is moving is less perceptible. Further, it shows that the proposed scheme is robust against common video processing attacks.

In this paper, we propose a novel digital video watermarking scheme based on multi resolution motion estimation and artificial neural network. A multi resolution motion estimation algorithm is adopted to preferentially allocate the watermark to coefficients containing motion. In addition, embedding and extraction of the watermark are based on the relationship between a wavelet coefficient and its neighbor's. A neural network is given to memorize the relationships between coefficients in a 3x3 block of the image. Experimental results show that embedding watermark where picture content is moving is less perceptible. Further, it shows that the proposed scheme is robust against common video processing attacks.

Enterprises are evolving towards a more agile, dynamic and adaptive organisation that can make quick responses to the market and customer requirements. This carried out an increasing need for enterprises to get involved in collaboration strategies. Moreover, new IT organisation, namely Service Oriented Architectures (SOA), can be introduced to implement opened and agile information system. To align the enterprise strategy and the information support system organisation, we present a cooperation model based on SOA, called service oriented enterprise. Thanks to a multi-level process organisation, simple combination and filtering rules can be applied to build dynamically customised distributed processes on demand. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

With the rapid development of multimedia applications, and the fact that a large quantity of multimedia data is distributed via the Internet network, the protection of video becomes a major problem. It is therefore necessary to protect them. This can be done using video watermarking methods. To be really efficient, the mark has to be resistant to malicious attacks. Methods developed for still images cannot be efficiently used for video. Effectively, if a mark is embedded in only one image of the video, a temporal filtering along the motion displacement can easily removed it without significant degradation of the video. It is therefore necessary to define specific techniques for video. In this context, we present in this paper a video watermarking method based on the use of video mosaicing. We show that the proposed method is resistant to temporal filtering attack