ABSTRACT The impact of rapid urbanisation, industrialisation, lack of awareness and lifestyle changes has resulted in an increase in the prevalence of diabetes and its complications such as diabetic macular edema (DME). It is the most common cause of blindness, characterised by an abnormal rise in the level of fluid in the macula. It affects the keenest vision in severe cases. In this paper, we propose an automatic grading of DME to help ophthalmologists diagnose the condition timely and early. The proposed system consists of an ensemble deep neural network using a residual convolution block followed by global average pooling and a dense layer. The proposed ensemble model consists of two models, namely model 1 and model 2. The proposed model has been evaluated on the publicly available benchmark ISBI IDRiD dataset. Our proposed model outperforms its competitive models in the IDRiD competition with an accuracy of 0.94. Moreover, it achieves a ROC‐AUC of 0.98. Gradient‐weighted Class Activation Mapping is also constructed to ensure that our proposed model is accurate.

This paper presents a binary coded evolutionary computational method inspired from the evolution in plant genetics. It introduces the concept of artificial DNA which is an abstract idea inspired from inheritance of characteristics in plant genetics through transmitting dominant and recessive genes and Epimutaiton. It involves a rehabilitation process which similar to plant biology provides further evolving mechanism against environmental mutation for being better and better. Test of the effectiveness, consistency, and efficiency of the proposed optimizer have been demonstrated through a variety of complex benchmark test functions. Simulation results and associated analysis of the proposed optimizer in comparison to Self-learning particle swarm optimization (SLPSO), Shuffled Frog Leap Algorithm (SFLA), Multi-species hybrid Genetic Algorithm (MSGA), Gravitational search algorithm (GSA), Group Search Optimization (GSO), Cuckoo Search (CS), Probabilistic Bee Algorithm (PBA), and Hybrid Differential PSO (HDSO) approve its applicability in solving complex problems. In this paper, we have shown effective results on thirty variables benchmark test problems of different classes.

In the area of medical imaging, automation of medical image classification or recognition is an active field of research. Over the past decade, due to the popularity and usability of artificial neural networks, it is becoming the norm to achieve this automation by deep neural networks. In toxicology based research, zebrafish has become a key model organism in phenotypical imaging and drug discovery. Detection of complex patterns and phenotypes in zebrafish embryo during preclinical or clinical trials is a standard part of the drug discovery cycle. Currently this classification task of phenotypes is mostly experts based. In this work, we propose a fine-tuned convolution neural network (CNN) based model for automated classification of different phenotypical changes observed due to the toxic substance in the zebrafish embryo. We demonstrated the ability of CNN model as well as a fine-tuned CNN based model to classify different deformation in an embryo with high accuracy. Such automated medical imaging model can be used extensively by experts in the area of toxicology and drug discovery.

The perceptual adaptation of the image (PAI) is introduced by inspiration from Chevreul–Mach Bands (CMB) visual phenomenon. By boosting the CMB assisting illusory effect on boundaries of the regions, PAI adapts the image to the perception of the human visual system and thereof increases the quality of the image. PAI is proposed for application to standard images or the output of any image processing technique. For the implementation of the PAI on the image, an algorithm of morphological filters (MFs) is presented, which geometrically adds the model of CMB effect. Numerical evaluation by improvement ratios of four no-reference image quality assessment (NR-IQA) indexes approves PAI performance where it can be noticeably observed in visual comparisons. Furthermore, PAI is applied as a postprocessing block for classical morphological filtering, weighted morphological filtering, and median morphological filtering in cancelation of salt and pepper, Gaussian, and speckle noise from MRI images, where the above specified NR-IQA indexes validate it. PAI effect on image enhancement is benchmarked upon morphological image sharpening and high-boost filtering.

Modern data is increasingly complex. High dimensionality, heterogeneity and independent multiple representations are the basic properties of today's data. With increasing sources of data collection, a single object can have multiple representations, which we call views. In this paper we propose a multiview classification technique, which uses fuzzy mapping to obtain maximum similarity between an object and nearest multiview centroids. Our fuzzy mapping based approach obtains a unit L1 hyperplane as a common space for each view. To establish the efficacy of our proposed method we present experimental comparisons with number of baselines on two synthetic and two real-world data sets.

Image representation through local descriptors is a research hotspot in computer vision. In this letter, we propose a novel local image descriptor based on intensity permutation and zone division. The oFAST detector is first employed to detect keypoints with orientations, and then steered patterns are applied to sample rotation-invariant points within the local keypoint patch. In the step of local patch description, intensity permutation and zone division are implemented to construct our descriptor, with the advantages of inherent robustness and invariance to monotonic brightness changes. Our proposed algorithm performed well in the experiments on benchmark dataset for descriptor evaluation.

Traffic sign recognition (TSR) is an essential research issue in the design of driving support system and smart vehicles. In this paper, we propose a permutation-based image feature to describe traffic signs, which has an inherent advantage of illumination invariance and fast implementation. Our proposed feature LIPID (local image permutation interval descriptor) employs interval division and zone number assignment on order permutation of pixel intensities, and takes the zone numbers as the descriptor. A comprehensive performance evaluation on German Traffic Sign Recognition Benchmark (GTSRB) dataset is carried out, which reveals the great performance of our proposed method. Experiment results exhibit that our feature outperforms some state-of-the-art descriptors, showing a potential prospect in TSR applications.

Image representation through local descriptors is the basis of numerous computer vision applications. In the past decade, many local image descriptors such as SIFT and SURF have been proposed, yet algorithms requiring low memory and computation complexity are still preferred. Binary descriptors such as BRIEF have been suggested to satisfy this demand, showing a comparable performance but much faster computation speed. In this paper, we propose a novel local image descriptor, LIPID, which employs intensity permutation and interval division to yield an effective performance in terms of speed and recognition. Our method is inspired by LUCID, proposed by Ziegler and Christiansen [8]. An extensive evaluation on the well-known benchmark datasets reveals the robustness and effectiveness of LIPID as well as its capability to handle illumination changes and texture images.

Electronic medical records provide us with an enormous amount of data with vast potential. If properly analyzed, medical data can be converted to knowledge that improves treatment, uncovers unexpected associations, and supports the personal experience of doctors and nurses, allowing them to make more informed decisions. Medical data is often generated by monitoring across a period of time, whether new data arrives quickly or slowly, consistently or sporadically, but many data mining methods are not designed to consider the temporal aspect of a data set. Besides the extra dimension of time, medical processes often involve interaction between many attributes at once, complicating the discovery of relevant patterns and associations. Specialized methods to interpret medical data can improve the quality of knowledge extracted from it. Tensors are appropriate data structures to represent our data in a multi-dimensional format, taking into account the relationship between many dimensions at once. We can further segment our data into discrete temporal chunks, creating a sequence of tensors. By applying dynamic tensor analysis to our tensor sequence, we can reveal patterns and associations within our data set and capture their change over time. This information can be developed into medical knowledge that can be used to support future treatment.

Neuropathic pain (NeuP) is often more difficult to treat than other types of chronic pain. The ability to predict outcomes in NeuP, such as response to specific therapies and return to work, would have tremendous value to both patients and society. In this work, we propose an adaptive clustering algorithm using the Neuropathic Pain Scale (NPS) to develop a set of standard patient templates. These templates may be useful in studying treatment response in NeuP. The approach is evaluated on 108 subjects' baseline data and results demonstrate the efficacy of utilizing neuropathic pain scale (NPS) metrics and our proposed method.

This paper formulates the automated tagging of media as a multilabel classification problem. An induction method for creating a multilabel decision tree is presented, then applied to the problem of identifying emotions associated with music clips. The results are compared to two other multilabel classification approaches (CLUS [26] and RAKEL [24]), and the proposed approach is shown to produce better results using a variety of performance measures.

This paper presents a method for automatic event detection in videos. The method requires no prior definition of events or their modeling and is thus suitable for unsupervised learning of events in a variety of applications. The presented approach relies on firing patterns of a collection of random local detectors in the field of view of a camera. The firing patterns of the detectors are clustered through a modified vector quantization approach. The resulting clusters are annotated, if needed to assign a meaningful description to each clustered group of firing patterns. The resulting codebook can be used for future detection of events of interests. We demonstrate our approach using data from two different applications.

This paper proposes a new active learning approach, confidence-based active learning, for training a wide range of classifiers. This approach is based on identifying and annotating uncertain samples. The uncertainty value of each sample is measured by its conditional error. The approach takes advantage of current classifiers' probability preserving and ordering properties. It calibrates the output scores of classifiers to conditional error. Thus, it can estimate the uncertainty value for each input sample according to its output score from a classifier and select only samples with uncertainty value above a user-defined threshold. Even though we cannot guarantee the optimality of the proposed approach, we find it to provide good performance. Compared with existing methods, this approach is robust without additional computational effort. A new active learning method for support vector machines (SVMs) is implemented following this approach. A dynamic bin width allocation method is proposed to accurately estimate sample conditional error and this method adapts to the underlying probabilities. The effectiveness of the proposed approach is demonstrated using synthetic and real data sets and its performance is compared with the widely used least certain active learning method.

We present a moving object recognition system based on the characteristics of the object's edges. Our approach is formulated as an edge detection problem within two consecutive frames of a video sequence. The framework is evaluated in a retail environment for detecting dollies entering or leaving the back door of a store and in opening or closing a cashier's cash drawer. The system is trained manually with pre-recorded video sequence of a dolly entering a building and a cash drawer opening and closing. Once trained, the system is shown to be able to detect dollies and cash drawer actions reliably. The framework can be used to detect objects of any defined edge characteristics as a part of an event detection system for small retail businesses.

This paper presents a shape-based retrieval system and its application to infrared satellite images. A complete system is presented, from region extraction of a full hemisphere scan to the actual retrieval mechanism. After region extraction, polygonal approximation is applied to the region shape, and local features of the polygons are hashed to provide an association space. This space becomes the indexing structure through which retrieval takes place. Although the indexing stage, containing region extraction and polygonal approximation, is slow, the actual retrieval is very fast. On average, retrieval of a query shape from a database of 1965 shapes takes 0.7 seconds for the more reduced representation, and 2.8 seconds for the less reduced representation consisting of 1914 shapes. The overall design is good for a moderately sized database, and extensions could be made to apply the method to a massive database. The results show that the approach performs well, and that there is a substantial speed benefit for using the local association hashing method.

This paper discusses the use of a combination of support vector machine and decision tree learning for recognizing four emotions in speech, which are neutral, angry, lombard, and loud. The base features selected were pitch, derivative of pitch, energy, speaking rate, formants, band-widths, and Mel frequency cepstral coefficients. Three methods of combining learned support vector machine and decision tree classifiers were proposed, namely, minimum misclassification, maximum accuracy, and dominant class. Using the Speech Under Simulated and Actual Stress database, the average accuracy from the minimum misclassification, maximum accuracy, and dominant class methods were 72.4%, 70.8%, 71.3% respectively as opposed to 63.9% and 67.4% which were obtained by using support vector machine and decision tree learning alone.

This paper describes a speaker detection system using cross-modal association methods. Four association approaches are designed using linear and nonlinear association models. Speaker detection experiments were conducted to compare the approaches

We present the preliminary results of a data mining study of a production line involving hundreds of variables related to mechanical, chemical, electrical and magnetic processes involved in manufacturing coated glass. The study was performed using two nonlinear, nonparametric approaches, namely neural network and CART, to model the relationship between the qualities of the coating and machine readings. Furthermore, neural network sensitivity analysis and CART variable rankings were used to gain insight into the coating process. Our initial results show the promise of data mining techniques to improve the production.

Talking face detection is important for videoconferencing. However, the detection of the talking face is difficult because of the low resolution of the capturing devices, the informal style of communication and the background sounds. In this paper, we present a novel method for finding the talking face using latent semantic indexing approach. We tested our method on a comprehensive set of home video conferencing sessions with a very high detection rate. Our experiments show that the LSI method accuracy degrades gracefully in a noisy environment as opposed to the correlation method which simply fails in presence of noise.

Multimodal information processing has received considerable attention in recent years. The focus of existing research in this area has been predominantly on the use of fusion technology. In this paper, we suggest that cross-modal association can provide a new set of powerful solutions in this area. We investigate different cross-modal association methods using the linear correlation model. We also introduce a novel method for cross-modal association called Cross-modal Factor Analysis (CFA). Our earlier work on Latent Semantic Indexing (LSI) is extended for applications that use off-line supervised training. As a promising research direction and practical application of cross-modal association, cross-modal information retrieval where queries from one modality are used to search for content in another modality using low-level features is then discussed in detail. Different association methods are tested and compared using the proposed cross-modal retrieval system. All these methods achieve significant dimensionality reduction. Among them CFA gives the best retrieval performance. Finally, this paper addresses the use of cross-modal association to detect talking heads. The CFA method achieves 91.1% detection accuracy, while LSI and Canonical Correlation Analysis (CCA) achieve 66.1% and 73.9% accuracy, respectively. As shown by experiments, cross-modal association provides many useful benefits, such as robust noise resistance and effective feature selection. Compared to CCA and LSI, the proposed CFA shows several advantages in analysis performance and feature usage. Its capability in feature selection and noise resistance also makes CFA a promising tool for many multimedia analysis applications.

MOTIVATION: We implement a data mining technique based on the method of Independent Component Analysis (ICA) to generate reliable independent data sets for different HIV therapies. We show that this technique takes advantage of the ICA power to eliminate the noise generated by artificial interaction of HIV system dynamics. Moreover, the incorporation of the actual laboratory data sets into the analysis phase offers a powerful advantage when compared with other mathematical procedures that consider the general behavior of HIV dynamics. RESULTS: The ICA algorithm has been used to generate different patterns of the HIV dynamics under different therapy conditions. The Kohonen Map has been used to eliminate redundant noise in each pattern to produce a reliable data set for the simulation phase. We show that under potent antiretroviral drugs, the value of the CD4+ cells in infected persons decreases gradually by about 11% every 100 days and the levels of the CD8+ cells increase gradually by about 2% every 100 days. AVAILABILITY: Executable code and data libraries are available by contacting the corresponding author. IMPLEMENTATION: Mathematica 4 has been used to simulate the suggested model. A Pentium III or higher platform is recommended.

A coarse-to-fine scheme for the detection of sideview faces in color images is proposed in this paper, which extends the current state of the art of face detection research. The input image can be of complex scene, containing cluttered background and confusing objects. The system consists of four stages, each of which is a refinement of the previous one, namely: (1) skin-tone detection by color, (2) region and edge preprocessing with morphological operations and length filtering, (3) face candidate region selection based on normalized similarity value and (4) final verification using hidden Markov models. Encouraging experimental results have been obtained, due to the utilization of multiple features of the input image and the conjunction of employment of various image processing and pattern recognition techniques. Besides providing the ability to detect faces other than frontal-view, our work has 3 original contributions, including the Normalized Similarity Value (NSV) to detect the presence of a given curve pattern, the iterative partition process to segment the object from confusing extraneous regions for higher detection accuracy and the exploration of the use of HMM to recognize objects in images.

Digital video is becoming widely available in compressed form, such as a motion JPEG or MPEG coded bitstream. In applications such as video browsing or picture-in-picture, or in transcoding for a lower bit rate, there is a need to downscale the video prior to its transmission. In such instances, the conventional approach to generating a downscaled video bitstream at the video server would be to first decompress the video, perform the downscaling operation in the pixel domain, and then recompress it as, say, an MPEG, bitstream for efficient delivery. This process is computationally expensive due to the motion-estimation process needed during the recompression phase. We propose an alternative compressed domain-based approach that computes motion vectors for the downscaled (N/2xN/2) video sequence directly from the original motion vectors for the N/spl times/N video sequence. We further discover that the scheme produces better results by weighting the original motion vectors adaptively. The proposed approach can lead to significant computational savings compared to the conventional spatial (pixel) domain approach. The proposed approach is useful for video severs that provide quality of service in real time for heterogeneous clients.

For direct manipulation in the compressed domain, we develop a DCT domain convolution theorem which besides exploiting the sparseness of the DCT domain representation also exploits the orthogonality and symmetry in the DCT domain representation. These properties lead to efficient compressed domain based processing methods unlike their spatial domain counterparts, where such properties are not available. This theorem can be used in a variety of image and video editing functions when the image and video data are available only as a JPEG or MPEG bitstream. We illustrate the use of these DCT domain convolution theorems in a typical video editing application such as video bluescreen editing.

Conventional processing of JPEG or MPEG compressed image or video data involves first decompressing the data and applying the desired processing function, and then the processed data are recompressed for the purposes of transmission or storage. We propose an alternate processing pipeline which involves direct manipulation of the JPEG or MPEG compressed domain representation to achieve the desired spatial domain processing. For direct manipulation in the compressed domain, we develop a discrete cosine transform (DCT)-domain convolution theorem which besides exploiting the sparseness of the DCT-domain representation also exploits the orthogonality and symmetry in the DCT-domain representation. These properties lead to efficient compressed domain-based processing methods unlike their spatial domain counterparts, where such properties are not available. This theorem can be used in a variety of image and video editing functions when the image and video data are available only as a JPEG or MPEG bitstream. We illustrate the use of the DCT-domain convolution theorem in a typical video editing application such as video bluescreen editing.

Digital video is becoming widely available only in the compressed form. In applications such as video browsing or picture-in-picture or transcoding for a lower bit rate, there is a need to downscale the video before transmission. In such instances, the conventional approach of generating a downscaled video bitstream at the video server is to first decompress the video, perform the downscaling operation in the pixel domain and then recompress it for efficient delivery. This process is computationally expensive due to the motion-estimation process needed during the recompression phase. We propose an alternative compressed domain based approach that computes the motion vectors for the downscaled (N/2/spl times/N/2) video sequence directly from the original motion vectors residing in the N/spl times/N video sequence. Furthermore, the downscaling of the DCT information is also done in the DCT domain. The proposed approach can lead to significant computational savings compared to the conventional spatial domain approach.

Off-line cursive script recognition has got increasing attention during the last three decades since it is of interest in several areas such as banking and postal service. An off-line cursive handwritten word recognition system is described in this paper and is used for legal amount interpretation in personal checks. The proposed recognition system uses a set of geometric and topologic features to characterize each word. By considering the spatial distribution of these features in a word image, the proposed system maps each word into two strings of finite symbols. A local associative indexing scheme is then used on these strings to organize a vocabulary. When presented with an unknown word, the system uses the same indexing scheme to retrieve a set of candidate words likely to match the input word. A verification process is then carried out to find the best match among the candidate set. The performance of the proposed system has been tested with a legal amount image database from real bankchecks. The results obtained indicate that the proposed system is able to recognize legal amounts with great accuracy.

Establishing the identify of a signature by automatic search through a database of signatures is of interest in several areas. This paper describes a system for this purpose. The proposed identification system uses a set of geometric and topologic features to characterize each signature. By considering the spatial distribution of these features, the system maps each signature into two strings of finite symbols. A local associative indexing scheme is then used on these strings to organize the collection of signatures of known identity. When presented with a signature of unknown identity, the system uses the same indexing scheme to retrieve a candidate set of signatures. A verification process is then carried out to find the best match from the candidate set. The performance of the proposed system has been tested with a moderate database. The results obtained indicate that the proposed system is able to identify signatures with great accuracy even when a part of a signature is missing.

The segmentation of off-line cursive handwriting is an important step in cursive handwriting recognition. In this paper, a new approach is described for this task. The suggested approach uses a set of heuristic rules to determine possible letter boundaries in the image of a cursive word. The heuristic rules are based on associations that exist between certain geometric and topologic features and the English language characters. A segmentation system incorporating the proposed approach has been built to perform segmentation on postal address images. The system includes several preprocessing steps to extract cursive handwritten words from a postal envelope and normalization steps to allow variations in pen thickness and tilt in writing. The experimental results obtained thus far show that the proposed approach is capable of accurately locating the letter boundaries in cursive words.

No abstract available.

Tree classifiers represent a popular non-parametric classification methodology that has been successfully used in many pattern recognition and learning tasks. However, "is feature-value/spl ges/thrsh" type of tests used in tree classifiers are often found sensitive to noise and minor variations in the data. This has led to the use of soft thresholding in decision trees. Following the decision tree to feedforward neural network mapping of the entropy net, three neural implementation schemes for tree classifiers, that allow soft thresholding, are presented in this paper. Results of several experiments using well-known data sets are described to compare the performance of the proposed implementations with respect to decision trees with hard thresholding.>

A neural learning methodology is presented that is capable of providing rules from learned weights. An application of this methodology to systematic lupus erythematosus is demonstrated. It is shown that the proposed approach can disregard irrelevant features in the data and can generate different criteria combinations indicating the presence of systemic lupus erythematosus in a patient.>

Relative motion carries valuable information for motion understanding and form perception. The authors present a scheme for detecting relative motion in image sequences of moving point patterns. The proposed scheme uses the complexity of image plane trajectories to determine the presence of relative motion. Experimental results for two moving point pattern sequences are presented to demonstrate the proposed approach.>

A class of hierarchical feedforward layered neural networks, called entropy nets, has been proposed to overcome the difficulty of the credit assignment problem and to introduce the self-configurable capability in the design and training of artificial neural networks. These networks are based on mapping binary decision trees into a three-layer structure. The authors report the results of an experimental study comparing the classification performance of the decision tree and the entropy-net classifiers. In addition to the consistently better performance of the entropy net with respect to the decision-tree classifier, it is observed that the entropy net is able to achieve its highest performance level in far fewer iterations than commonly required in the nonhierarchical feedforward networks trained using backpropagation. The reason for this lies in the coarse learning achieved during the binary tree development

Occlusion and poor feature point detection are two of the main difficulties in the use of multiple frames for establishing correspondence of feature points. A formulation of the correspondence problem as an optimization problem is used to handle these difficulties. Modifications to an existing iterative optimization procedure for solving the formulation of the correspondence problem are discussed. Experimental results are presented to show the merits of the formulation.>

How the mapping of decision trees into a multilayer neural network structure can be exploited for the systematic design of a class of layered neural networks, called entropy nets (which have far fewer connections), is shown. Several important issues such as the automatic tree generation, incorporation of the incremental learning, and the generalization of knowledge acquired during the tree design phase are discussed. A two-step methodology for designing entropy networks is presented. The methodology specifies the number of neurons needed in each layer, along with the desired output, thereby leading to a faster progressive training procedure that allows each layer to be trained separately. Two examples are presented to show the success of neural network design through decision-tree mapping.>

Identifying the same physical point in more than one image, the correspondence problem, is vital in motion analysis. Most research for establishing correspondence uses only two frames of a sequence to solve this problem. By using a sequence of frames, it is possible to exploit the fact that due to inertia the motion of an object cannot change instantaneously. By using smoothness of motion, it is possible to solve the correspondence problem for arbitrary motion of several nonrigid objects in a scene. We formulate the correspondence problem as an optimization problem and propose an iterative algorithm to find trajectories of points in a monocular image sequence. A modified form of this algorithm is useful in case of occlusion also. We demonstrate the efficacy of this approach considering synthetic, laboratory, and real scenes.

A nonparametric algorithm is presented for the hierarchical partitioning of the feature space. The algorithm is based on the concept of average mutual information, and is suitable for multifeature multicategory pattern recognition problems. The algorithm generates an efficient partitioning tree for specified probability of error by maximizing the amount of average mutual information gain at each partitioning step. A confidence bound expression is presented for the resulting classifier. Three examples, including one of handprinted numeral recognition, are presented to demonstrate the effectiveness of the algorithm.