This paper presents a unified deep learning framework for breast cancer detection that generalizes effectively across mammography and histopathology.Using fine-tuned CNN architectures evaluated under a consistent cross-modal protocol, the method achieves stable, high accuracy on both imaging types, demonstrating robustness to domain shifts and heterogeneous clinical conditions.Another key contribution is the integration of model-agnostic (LIME, SHAP) and model-specific (Grad-CAM) explainability techniques, enabling a balanced trade-off between performance and interpretability.This hybrid XAI strategy provides clinically meaningful visual and feature-level insights, supporting transparent, reliable, and multi-modal diagnostic decision-making.

This paper presents a novel hybrid cryptosystem for image encryption that combines the lightweight Ascon authenticated cypher with neural networks and chaotic systems. The proposed Chaotic-Neural Ascon Image Encryption (CNAIE) system employs Mish activation functions in neural diffusion and reinforcement learning through Q-learning for key scheduling adaptability. Our approach addresses the urgent need for lightweight and secure encryption methods for Internet of Things (IoT) devices with minimal computational overhead. Experimental results on several test images demonstrate the proficiency of the cryptosystem with near-optimal encryption entropy ($\approx 7.99$) and negligible adjacent pixel correlation (<0.01) compared to plaintext images ($\gt0.90$). The uniform histogram distribution and randomised pixel relations within encrypted images confirm the resilience against statistical attacks. Security analysis confirms the algorithm’s sensitivity to minor key alterations, where changing a single bit causes drastically different outputs. Performance tests demonstrate the system’s feasibility in resource-constrained IoT environments with NIST-compliant security features.

This paper advances a novel encryption scheme using the integration of transformer neural networks and chaotic systems to enhance security for resource-constrained IoT devices. A novel lightweight realization of the ASCON authenticated encryption scheme with dynamically generated S-boxes inspired by Swish and GELU neural activation functions stimulated by a transformer architecture is proposed. The scheme employs a Piecewise Linear Chaotic Map (PWLCM) with the Lyapunov exponent of 1.25 to generate pseudorandom sequences for key derivation. Experimental evidence gives high-quality cryptographic characteristics with entropies of 7.997 (close to theoretical maximum), correlation coefficients reduced to near-zero (0.004-0.02), and best diffusion properties (NPCR $\sim 99.6 \%$, UACI $\sim 31.4 \%$). Both the Swish and GELU-based S-boxes have near-ideal nonlinearity ($\sim 3.99$) as well as bit change ratios ($\sim 0.5$). Performance analysis shows modest resource requirements (0.105 MB additional memory) at a rate of $\sim \mathbf{0. 1 1 ~ M B} / \mathbf{s}$, thereby rendering the system viable for security-concerned IoT systems where security demands overtake processing demands. Relative comparison with AES-like and reduced chaotic algorithms validates that our approach maintains comparable security statistics while presenting greater immunity against potential future attacks due to its neural-based mechanism. The proposed system is a paradigm shift in adaptive cryptographic protection for future IoT systems.

Surveillance, forensic, and biometric systems are continually evolving with modern technology. Despite these advancements, their effectiveness in crime deterrence remains debatable, especially given the low-resolution nature of footage from distant cameras. Gait-based gender identification is a crucial application in surveillance. This paper investigates the use of minimal human skeleton points to derive discriminative features for gender recognition. We propose a new descriptor using the 7 Hu’s moments, applied to two regions: shoulders-hips and hips-knees, extracted from only 5 frames per sequence. Experiments were conducted using three databases (CASIA-B and two Kinect databases) and evaluated with SVM and KNN classifiers. Comparative analysis with established research highlights the effectiveness of our approach. Results confirm the reliability of shoulders-hips and hips-knees regions as biometric features for gender recognition across various factors. The highest accuracy achieved was 95.19% using the SVM classifier with the Kinect v2 dataset.

Fog Computing is a term made by Cisco that insinuates extending cloud computing to the edge of a network. Fog computing supports the operation of Fog/cloud, storage, and networking services between end devices and conveyed processing data centers. While depending on the fog network will enhance the performance by eliminating the upper layer between IoT devices and Cloud servers by making users and devices closer to the servers. But as users become closer to the servers and data centers, attackers also become closer too, this will make the fog layer subjected more to attack and the data centers will be dangerous. Now fog computing faces new-fangled security and assurance defies other than those procured from cloud computing. In this paperwork, we will list the types of attacks that affect the Fog network and we discuss the available solution, moreover, we will illustrate our new method to protect this layer from the attacks which will introduce a new security layer between IoT devices and fog network to detect and prevent the attacks from reaching the fog layer, in addition, we will protect the fog layer from blocking when an attack is presented.

ABSTRACT Consumer Internet of Things (CIoT) has revolutionised agriculture by integrating intelligent computing, artificial intelligence and big data technologies in crop monitoring. This paper explores the application of intelligent computing and deep learning methodologies in crop monitoring within the CIoT framework. In CIoT‐based crop monitoring, a vision sensor collects real‐time data from crop leaf images. The image dataset is processed using state‐of‐the‐art deep learning models and intelligent computing algorithms. This integration enables the early detection of crop diseases by leveraging computer vision and deep learning. Intelligent computing systems provide accurate disease classification, real‐time alerts, and actionable recommendations for optimised crop management practises. This advanced system empowers farmers to make data‐driven decisions, such as irrigation optimization, targeted pesticide application and nutrient supplementation, to maximise crop productivity and minimise losses. A benchmark dataset of leaf images is used, and a deep learning based model is presented for classifying healthy and diseased leaves. Experimental results demonstrate an accuracy rate of 0.98, with detailed validation, including dataset size and model parameters. Key benefits of intelligent computing in CIoT‐based crop monitoring include enhanced resource efficiency, reduced environmental impact, and improved sustainability. The paper also addresses the challenges of implementing AI and big data technologies, such as data privacy, security, interoperability and resource management in agricultural settings.

ABSTRACT Medical image analysis is a critical component of modern healthcare, enabling accurate disease diagnosis and effective patient treatment. However, the process is fraught with challenges, including inter‐ and intra‐observer variability, time constraints, and data‐related issues such as privacy, heterogeneity and accessibility. Within this framework, Federated Learning (FL) has emerged as a promising solution, allowing collaborative model training across distributed healthcare entities without sharing sensitive patient data. This study provides a comprehensive Systematization of Knowledge (SoK) review of FL and its extension, Federated Unlearning (FU), within the context of medical image analysis. FL enables privacy‐preserving, decentralised model training, while FU addresses the ‘Right To Be Forgotten’, ensuring compliance with data protection regulations like GDPR and HIPAA. We explore the opportunities and challenges of FL and FU, detailing their methodologies, frameworks, datasets, and evaluation metrics. The review highlights the potential of FL and FU to enhance diagnostic accuracy, improve patient care, and foster trust in AI‐driven healthcare systems. We also identify research gaps and propose future directions for advancing FL and FU in medical imaging, emphasising the need for interdisciplinary collaboration and the development of dedicated frameworks. Thus, this study aims to bridge the gap between theoretical advancements and practical applications, paving the way for more robust and privacy‐compliant AI models in healthcare.

Abstract Human emotional states encompass both basic and compound facial expressions. However, current works primarily focus on basic expressions, consequently neglecting the broad spectrum of human emotions encountered in practical scenarios. Compound facial expressions involve the simultaneous manifestation of multiple emotions on an individual's face. This phenomenon reflects the complexity and richness of human states, where facial features dynamically convey a combination of feelings. This study embarks on a pioneering exploration of Compound Facial Expression Recognition (CFER), with a distinctive emphasis on leveraging the Label Distribution Learning (LDL) paradigm. This strategic application of LDL aims to address the ambiguity and complexity inherent in compound expressions, marking a significant departure from the dominant Single Label Learning (SLL) and Multi‐Label Learning (MLL) paradigms. Within this framework, we rigorously investigate the potential of LDL for a critical challenge in Facial Expression Recognition (FER): recognizing compound facial expressions in uncontrolled environments. We utilize the recently introduced RAF‐CE dataset, meticulously designed for compound expression assessment. By conducting a comprehensive comparative analysis pitting LDL against conventional SLL and MLL approaches on RAF‐CE, we aim to definitively establish LDL's superiority in handling this complex task. Furthermore, we assess the generalizability of LDL models trained on RAF‐CE by evaluating their performance on the EmotioNet and RAF‐DB Compound datasets. This demonstrates their effectiveness without domain adaptation. To solidify these findings, we conduct a comprehensive comparative analysis of 12 cutting‐edge LDL algorithms on RAF‐CE, S‐BU3DFE, and S‐JAFFE datasets, providing valuable insights into the most effective LDL techniques for FER in‐the‐wild.

The increasing interest in eXplainable Artificial Intelligence (XAI) is driven by the need to understand complex deep learning models, especially in critical fields like skin cancer classification. Existing research has provided various methods to interpret and clarify the decision-making processes of complex AI systems. Among the most widely used XAI methods for deep learning models are post-hoc saliency maps, which highlight the features that contribute most to a particular prediction. However, the trustworthiness of these explanations remains a concern, particularly when interpretations can be subjective. This raises a critical question: how can we effectively evaluate the quality of explanations generated by XAI methods for deep learning predictions and how it is possible to select the appropriate saliency map method? Consequently, this issue has caused the need to develop methods for evaluating XAI. These methods aim to not only interpret model decisions but also to compare different explanation methods through qualitative and quantitative measures. This study contributes to quantitatively evaluate and compare the performances of four saliency mapping-based XAI methods, namely LIME, SHAP, Attention Maps, and Grad-CAM. In our proposal, the outputs of XAI methods are used to create occluded images using feature importance scores. The masked images will then be fed to the end-to-end classifier. To measure the performance, the main metrics that we used to assess the XAI method faithfulness are the correlation between the classifiers prediction and the features importance score before and after occlusion. The obtained results show that SHAP outperforms the other three methods and is thus more faithful. These results may help indicate that SHAP is the most suitable XAI method that can explain skin lesion classification through InceptionResnetV2 model.

Human fall detection is a crucial topic to study, since there are a lot of cases of person’s fall at hospitals, homes and retirement homes. In fact, falls are very costly, especially for elderly people and persons with special needs, since they may cause death or serious injuries that require instance medical intervention. In order to prevent further repercussions after this type of accidents, modern automated fall detection methods are presented as a type of effective alerting systems that are widely used for emerging healthcare applications. In this study, we present a multi-view-based fall detection method that runs in real time, using only CPU and consequently it can be deployed in hospitals, retirement homes and cribs without any financial problems related to expensive hardware. Indeed, a light weight human pose estimator has been adopted in order to detect human body key-points from two different views in order to solve the problematic of image depth ambiguity. Then, we extract few explainable features, based on the automatically detected key-points, while being associated to confirmed descriptors of posture and balance biomechanics. The extracted features are thereafter fed into a machine learning classifier in order to predict whether there is a fall or not. The proposed method has been tested on a challenging public dataset, and the preliminary obtained results show its effectiveness compared to other relevant state-of-the-art methods.

Facial Expression Recognition (FER) is a crucial aspect in various domains, given its significance in understanding human emotions. However, designing efficient FER systems entails addressing challenges in feature extraction and selection. While previous studies have primarily focused on static feature selection methods, these approaches often struggle with spontaneous expressions due to the unique facial characteristics of each individual. To address this challenge, we implemented a Facial Morphology-Guided Feature Selection Method that combines texture features using Local Binary Pattern histograms and geometric features employing linear and eccentricity features. Subsequently, we employ Recursive Feature Elimination (RFE) and Binarized Genetic Algorithm (BGA) algorithms for feature selection, combining their outputs to identify the optimal subset of features tailored for each face. Experimental validation using the CK+ and DISFA datasets demonstrates the effectiveness of our approach in enhancing facial expression recognition accuracy.

Rich heterogeneous data provided by social networks can be very big, which imposes considerable challenges for object extraction and image classification. Therefore, the objective of this work is to propose an unsupervised object cosegmentation method that could be notably efficient to improve image classification performance. The main goal of cosegmentation is to extract the salient common objects within each image. To this end, we propose to minimize an energy function based on the Markov Random Field using the saliency detection, while considering linear dependence of generated foreground histograms of the input image collection. In fact, the saliency detection is processed in two steps. In each image, we detect salient objects, by considering appearance similarity and spatial distributions of image pixels. Then, fuzzy quantification is used to correct the belonging of pixels to the foreground objects. Finally, an iterative optimization permits to enhance the final segmentation results. The proposed method has been validated as a preprocessing step for image classification. Indeed, to enhance cosegmentation-based classification performance, we have applied a semi-supervised object classification based on ensemble projection. Qualitative and quantitative evaluations of the proposed cosegmentation and classification techniques on the iCoseg, CDS and Oxford Flowers 17 datasets demonstrate the effectiveness of the proposed framework.

Image change detection in remote sensing is crucial for monitoring environmental changes at different temporal and spatial scales. The primary goal is to identify altered pixels in multi-temporal images accurately. However, challenges such as response latency and limited large-scale validation persist. In this study, we propose an accurate automated change detection method called ”ResUNet” based on multi-spectral NDVI imagery. ResUNet combines UNet and residual networks, while employing deep learning-based features for precise change detection. We evaluate the proposed method on low-resolution data from three geographical regions: Colombia, California, and Duluth. Each region comprises 145,161 patches, ensuring comprehensive coverage for experimentation. We validate our method in three distinct areas, achieving an accuracy of 99.50% and an F1-score of 99.41%.

Convolutional Neural Networks (CNNs) have been highly successful in computer vision tasks, including leaf disease detection. However, their lack of interpretability limits our understanding of their decision-making process and undermines trust in their predictions. In this study, we aim to unravel the black box of CNNs for leaf disease detection through a comprehensive model analysis and feature importance study. We review various techniques in explainable artificial intelligence and propose a methodology that combines model analysis and feature importance methods. We analyze the model’s internal representations and activations in order to understand how different layers process information. Additionally, we employ feature importance methods, like SHapley Additive exPlanations (SHAP), to quantify the influence of individual features on predictions. By assigning importance scores to each pixel or feature, we identify the most discriminative regions or characteristics used by the CNN for disease detection. These insights provide a deeper understanding of learned representations, the decision-making process, and the key visual cues used by CNNs. Our findings enhance interpretability, foster trust in automated leaf disease detection systems, and facilitate the development of more explainable and reliable models in precision agriculture.

Accurate segmentation of the Corpus Callosum (CC) plays a crucial role in studying brain connectivity and understanding neurological disorders. However, limited availability of annotated data poses a significant challenge for developing robust segmentation models. In order to deal with this issue, we propose in the study an effective approach that combines one-shot learning and a modified multiclass U-Net architecture. The proposed approach represents the first attempt in this context, to the best of our knowledge. We begin by generating additional Ground Truth (GT) data using one-shot learning, effectively expanding the limited annotated dataset. This approach leverages the inherent generalization capability of one-shot learning to predict segmentation for unlabeled data, which are then validated and refined by domain experts. The refined segmentation serves as new GT data, enhancing the training process. To further improve parcellation accuracy, we modify the U-Net architecture to handle the complex task of multiclass CC parcellation. The modified multiclass U-Net effectively captures the intricate features and spatial dependencies within the CC, enabling precise parcellation into distinct sub-regions. The framework has been tested and evaluated on two challenging datasets that are publicly available. The obtained results are promising and show the performance of the proposed solution against geometric methods from the state of the art.

In this paper, a multi-stage deep learning model is proposed for brain glioma tumor detection and segmentation from MRI scans. The model consists of two stages: object detection using YOLOv7 with EfficientNet-B0 backbone, and active contour snake model for boundary refinement and segmentation. The proposed method also includes a customized CNN with feature selection and GRU layer for accurate class label prediction. The proposed model has been trained on the BraTS 2020 dataset and has achieved state-of-the-art performance in terms of accuracy and effectiveness. This proposed method can potentially assist radiologists and clinicians in detecting and segmenting brain tumors in medical images, leading to better diagnosis and treatment planning for patients.

Precision agriculture plays a crucial role in optimizing crop yield, reducing environmental impact, and ensuring sustainable agricultural practices. Early detection and accurate diagnosis of leaf diseases are essential for preventing significant losses in crop production and maintaining food security. However, the inherent challenge of class imbalance in leaf disease datasets poses a significant obstacle for machine learning algorithms. In this paper, we explore and compare different techniques for handling class imbalance in leaf disease detection to improve the accuracy and reliability of machine learning models in the context of precision agriculture. We investigated the performance of different methods for leaf disease detection using the challenging New Plant Diseases Dataset (NPDD), which consists of image-based plant leaves. Our experiments reveal promising results, particularly with the hierarchical approach, achieving an accuracy of 97.17%. The outcomes of our study contribute to the growing body of knowledge in precision agriculture by providing a comprehensive analysis of techniques for handling class imbalance in leaf disease detection. Furthermore, our findings serve as a valuable resource for researchers and practitioners in the field, offering guidance on selecting and implementing the most effective approaches to tackle class imbalance challenges and improving the overall performance and reliability of machine learning models in the domain of precision agriculture.

Nowadays, gender recognition systems are very important in several fields such as security, human machine interaction, surveillance and targeted advertising. However, many factors, such as makeup and disguise, can affect recognition and extend the processing time. Our research revolves around this issue. This is a comparative experimental study of the significance of each part of the face (eyes, mouth, nose) in the gender facial recognition via convolutional neural networks (CNN). As a first step our goal is to find the most crucial part of the face in order to determine the most important part in the gender recognition. The used method was tested on the UTKFace dataset and the preliminary results confirm that the eyes contain the most discriminating information regarding gender identification. We achieve a classification accuracy of 92% for eyes, 91% for mouth and 89% for nose. Then we propose a second study on the degree of importance of the eyes for both genders by training the system using only eyes. We achieve a classification accuracy of 99% for eyes of men and 99% for eyes of women.

Facial Expression Recognition (FER) is an active area of research in computer vision with a plethora of applications that have invested several techniques to improve recognition performance. We notice that most of these applications are oriented much more towards posed and environment-controlled emotions. However, FER in the wild remains an area deserving more attention. To address this issue, we investigate the use of deep learning-based methods, which have proven their effectiveness in several recent studies in FER. First, we challenge the studied methods within the context of in-the-wild using the Static Facial Expressions in-the-Wild (SFEW) benchmark dataset to assess their performance in real-world conditions. Then a method based on deep-learned features using effective CNN models, is proposed to handle the challenge involved in the comparative study. The suggested method performs a transfer learning based on freezing weights technique. Indeed, some shallow layers have been frozen, features in the deeper ones have been exploited to conceive the best configuration for the facial expression predictor model, and classification layers of models have been removed and replaced by our own SVM classifier. The proposed method achieved remarkable performance with the VGG19 and ResNet101 pre-trained models and outperformed other state-of-the-art deep learning methods for in-the-wild FER.

In this study, an effective method used to examine the changes of two optical images captured by Landsat satellite is presented. The proposed method is based on two main parts: A preprocessing step where Shearlet Transform is applied to get a smoother rendering followed by a classification process using Convolutional Neural Network (CNN) to change detection.The proposed method out performed the state-of-the-art methods with an accuracy out of 99,32%.

Face recognition has attracted the attention of many researchers during the last years due to its many applications in various fields. However, this task faces several challenges related to many changes that can affect the human face. In particular, make-up faces represent a major challenge for facial verification. To deal with this issue, we propose in this work an efficient salient patch-based method for verifying faces under makeup variation. Firstly, we use Mutli-Task Cascaded Convolutional Neural Networks (MTCNN) in order to jointly detect and align the face. Then, we have adapted the O-NET network in order to robustly detect five landmarks by training it on makeup faces. The Histogram of Oriented Gradients (HOG) descriptor and the Local Binary Patterns (LBP) are then used to represent the face by concatenating their histogram features in few salient regions around the detected landmarks. Finally, we estimate the similarity measure between the extracted features in order to compare the two faces while determining whether they are for the same person or not. The performance of the proposed method has been validated on the challenging YMU (YouTube Makeup dataset ) and MIFS (Makeup Induced Face Spoofing) datasets, and the obtained results proved the superiority of the proposed method against relevant multi-patch-based methods from the state of the art.

Retinal blood vessel segmentation in fundus images has become essential for various applications of computer-aided anomaly analysis. In this work, we propose an automated segmentation method based on mathematical morphology combined with entropy information, what allows an accurate classification of each pixel depending on its neighbors’ comportment. The main contribution resides in the joint integration, for the first time within the context of retinal blood vessel segmentation to the best of our knowledge, of the entropy information within the hysteresis thresholding. The first step of the method consists of classifying the image pixels using morphological operators. Then, we extract the entropy information followed by a hysteresis thresholding in order to isolate the retinal blood vessels from the background, while ensuring the smoothness and the spatial coherence of the kept pixels. Finally, morphological operators are applied to refine the segmentation results. Qualitative and quantitative tests were performed on the standard DRIVE and STARE datasets and the obtained results proved the effectiveness of the proposed method.

Several methods of brain image registration have been proposed in order to overcome the requirement of clinicians. In this paper, we assess the performance of a hybrid method for brain image registration against the most used standard registration tools. Most traditional registration tools use different methods for mono- and multi-modal registration, whereas the hybrid registration method is providing both mono and multi-modal brain registration of PET, MRI and CT images. To determine the appropriate registration method, we used two challenging brain image datasets as well as two evaluation metrics. Results show that the hybrid method outperforms all other standard registration tools and has achieved promising accuracy for MRI/X brain image registration.

In this paper, we introduce an unsupervised method for the parcellation of the Corpus Callosum (CC) from MRI images. Since there are no visible landmarks within the structure that explicit its parcels, non-geometric CC parcellation is a challenging task especially that almost of proposed methods are geometric or data-based. In fact, in order to subdivide the CC from brain sagittal MRI scans, we adopt the probabilistic neural network as a clustering technique. Then, we use a cluster validity measure based on the maximum entropy (Vmep) to obtain the optimal number of classes. After that, we obtain the isolated CC that we parcel automatically using SLIC (Simple Linear Iterative Clustering) as superpixel segmentation technique. The obtained results on two challenging public datasets prove the performance of the proposed method against geometric methods from the state of the art. Indeed, as best as we know, it is the first work that investigates the validation of a CC parcellation method on ground-truth datasets using many objective metrics.

This work proposes a framework for simultaneously segmenting foreground objects in a collection of images having heterogeneous contents. Rather than resorting to image co‐segmentation to segment similar objects in multiple images, which requires the use of categorised images, the authors’ idea disseminates segmentation information within images. In this way, it becomes easier to detect foreground objects in all of them simultaneously, mainly under the hypothesis of using similar or different images. General information is aggregated, on foregrounds as well as on backgrounds, from a set of images for joint segmentation of category‐independent objects. The key idea is to estimate the linear dependence of the foreground histograms of the input images to optimise a Markov random field‐based energy function. Iterative optimisation of each image permits after that the enhancement of the final segmentation results. Extensive experiments demonstrate that the proposed method (PM) enables full‐object segmentation of foreground objects within a collection of images composed of different classes. Indeed, the validation of the accuracy on five challenging datasets (iCoseg, Oxford Flowers, MicroSoft Research Cambridge (MSRC), Caltech101 and Berkeley) shows that the PM achieves satisfactory results as compared with state‐of‐the‐art methods. Besides, it has the challenging ability to efficiently deal with uncategorised objects.

This research presents an automatic method to detect and evaluate the shoreline changes from Landsat satellite images. In fact, a method, that we called Lukas-Kanade Adapted for Coastal Changes (LKA2C), has been developed to calculate and detect the changes around the study region. Mainly the proposed method is based on SURF algorithm to detect the study region from the satellite image. Then, Canny edge detector was used on NDWI images to detect the shorelines. Finally, the pyramidal Lukas-Kanade optical flow algorithm was adapted to detect and to calculate the rates of changes. Realized experiments on real satellite images of the island of Djerba in Tunisia proved the effectiveness of the proposed method.

Content-based image retrieval (CBIR) is a primordial task to provide the most similar images especially in the context of medical imaging for diagnosis aid. In this paper, we propose a CBIR method for a large-scale mammogram datasets. In fact, to extract region of interest (ROI) signatures, four moment descriptors were defined after computing the curvelet coefficients for each level of the ROI. Then, an unsupervised technique based on locality sensitive hashing was adopted for indexing the extracted signatures. The main contribution of the suggested method resides in the variance-based filtering within the retrieval phase in order to extract the suitable buckets in the shortest time, while optimizing the memory requirement. After that, an accurate searching in Hamming space is performed in order to identify the similar ROIs to the query case. Realized experiments on the challenging Digital Database for Screening Mammography (DDSM) dataset proved the performance of the proposed method for the retrieval of the most relevant mammograms in a large-scale dataset. It achieves a mean retrieval precision rate of 97.1% over a total of 11218 mammogram ROIs.

Screening mammography provides two views for each breast: Medio-Lateral Oblique (MLO) and Cranial-Caudal (CC) views. However, current content based image retrieval (CBIR) systems analyze each view independently, in spite of their complementarities. To further improve the retrieval performance, this paper introduces a two-view CBIR system that combines retrieval results of MLO and CC views. First, we computed the similarity scores between MLO (resp. CC) ROIs in the database and the MLO (resp. CC) query ROI. These ROIs are characterized using curvelet moments. Then, a new linear weighted sum scheme combines MLO and CC scores; it assigns weights for each view according to the distribution of the classes of its neighbors. The ROIs having the highest fused scores are displayed to the radiologist and used to compute the malignancy likelihood of the lesion. Experiments performed on mammograms from the Digital Database for Screening Mammography (DDSM) show the effectiveness of the proposed method.

In this study, the authors present an efficient method for approximate large‐scale image indexing and retrieval. The proposed method is mainly based on the visual content of the image regions. Indeed, regions are obtained by a fuzzy segmentation and they are described using high‐frequency sub‐band wavelets. Moreover, because of the difficulty in managing a huge amount of data, which is caused by the exponential growth of the processing time, approximate nearest neighbour algorithms are used to improve the retrieval speed. Therefore they adopted locality‐sensitive hashing (LSH) for region‐based indexing of images. In particular, since LSH performance depends fundamentally on the hash function partitioning the space, they exposed a new function, inspired from the E 8 lattice, that can efficiently be combined with the multi‐probe LSH and the query‐adaptive LSH. To justify the adopted theoretical choices and to highlight the efficiency of the proposed method, a set of experiments related to the region‐based image retrieval are carried out on the challenging ‘Wang’ data set.

Retrieval engines provide results according to user request. Nevertheless, reaching satisfaction can not be guaranteed with simple retrieval step. Therefore, it is necessary to communicate this dissatisfaction to the system through relevance feedback techniques. Indeed, with the growing number of image collections and by applying approximate nearest neighbor (ANN) algorithms to resolve the curse of dimensionality, the semantic gap may increase. For this reason, an additional step of relevance feedback is needful to add semantics to the next retrieval iterations. In this paper, a classification of the different relevance feedback techniques related to region-based image retrieval applications is elaborated. Moreover, a new technique of relevance feedback based on re-weighting regions of the query-image by selecting a set of negative examples is detailed. Furthermore, the general context to carry out this technique which is the large-scale heterogeneous image collections indexing and retrieval is presented. In fact, the main contribution of the proposed work is affording efficient results with the minimum number of relevance feedback iterations for high dimensional image databases. Experiments and assessments are carried out within an RBIR system for "Wang" data set in order to prove the effectiveness of the proposed approaches.

Voxel colouring is a popular method for reconstructing a three‐dimensional surface model from a set of a few calibrated images. However, the reconstruction quality is largely dependent on a thresholding procedure allowing the authors to decide, for each voxel, whether it is photo‐consistent or not. Nevertheless, in addition to the absence of any information on the neighbouring voxels during the photo‐consistency test, it is extremely difficult to define the appropriate thresholds, which should be precise and stable on all surface voxels. In this study, the authors propose an automated photo‐consistency test based on fuzzy hysteresis thresholding. The proposed method allows the incorporation of the spatial coherence during volume reconstruction, while avoiding ‘floating voxels’ and holes. Moreover, the ambiguity of choosing the thresholds is extremely minimised by defining a fuzzy degree of membership of each voxel into the class of consistent voxels. Also, there is no need for preset thresholds since the hysteresis ones are defined automatically and adaptively depending on the number of images that the voxel is projected onto. Preliminary results are very promising and demonstrate that the proposed method performs automatically precise and smooth volumetric scene reconstruction.

Summary form only given. In this paper we introduce a standard approach for medical images segmentation. In fact, for these images the segmentation consists in the extraction of an area of interest representing the organ subject of diagnosis. We distinguish two approaches depending on whether this area is composed of one region or of many regions. If it is composed of a single region, we introduce a growing region algorithm after a prestep based on fuzzy sets. Otherwise, we introduce an approach integrating a hierarchical system of segmentation in regions and a system of fuzzy classification. We illustrate our approach by applying it on real images in the frameworks of dermatology, neurology and mammography.