Azeddine Beghdadi

Videos and images undergo various distortions and artefacts during acquisition, processing, compression, transmission, and storage. These distortions and artefacts may impact the performance of high-level Computer Vision (CV) tasks. Therefore, identifying distortions/artefacts is crucial for developing and improving CV technologies. This article provides an overview of the distortions and artefacts that affect the quality of image and video signals. It also offers a framework highlighting the role of databases, especially in this era where data represents the fuel of current technologies. Various methods related to CV and image processing, focusing on distortion detection and classification in images/videos, are discussed. A wide range of applications and techniques developed in computer vision and associated databases are also analysed. Our study finds that while Machine Learning (ML) based approaches provide promising results for single distortion, multiple distortion classification remains challenging. Moreover, some distortions are still missing in the existing datasets, indicating room for constructing more databases with various distortions necessary to improve the ML-based models used in various CV-based applications. This article benefits researchers working in CV applications where visual data quality plays an important role. It provides insights for designing efficient ML-based solutions for several real-world CV applications.

This paper addresses the problem of no reference visual quality assessment in point clouds, useful for extended reality communication service such as remote surgery and education. Accurate, computationally efficient metrics for point cloud visual quality are needed, but the literature lacks a unified view of the signal attributes associated with psycho-visual features of point clouds.We propose a method to estimate point cloud visual quality by jointly assessing geometry and color noise in areas critical to visual perception. Our approach introduces a novel descriptor that identifies visual degradation through variations in local color and normal components. This descriptor, along with its saliency-weighted variant, leverages Non-Euclidean Laplacian Filtering (NEUF). The NEUF algorithm extracts descriptors using advanced non-Euclidean filtering techniques and employs regression learning to predict end-users’ perceived subjective quality by optimally selecting the most relevant features. We show by simulations that the NEUF approach mixing non Euclidean filtering with data-driven learning is effective on various point cloud categories, including human, non-human, natural, and synthetic. The NEUF-based method for point cloud quality estimation outperforms existing methods, offering a no-reference quality estimator that can be used for the design of cutting-edge extended reality services.

The high-impact scenario of UAV live uplink streaming is gaining significant interest in diverse applications, such as ambient monitoring, disaster rescue, and smart surveillance. This paper addresses the problem of uplink streaming by a fleet of camera-equipped UAVs, with one UAV acting as the sink, collecting and transmitting videos from the others. We demonstrate that performing video stabilization at the source UAVs or the sink enhances video quality and reduces required communication throughput, leading to bandwidth savings. We analyze the UAV live uplink streaming architecture to identify the most effective stabilization point within the network in a distributed manner. Using a reinforcement learning framework, we develop a method to dynamically optimize the stabilization gain-cost trade-off, pinpointing the optimal node for stabilization tasks. Through targeted numerical simulations under different system conditions we identify when and where stabilization should be applied to maximize efficiency. Our results show that video stabilization improves system performance in terms of media quality, battery life, and bandwidth usage.

Scene understanding plays an important role in several high-level computer vision applications, such as autonomous vehicles, intelligent video surveillance, or robotics. However, too few solutions have been proposed for indoor/outdoor scene classification to ensure scene context adaptability for computer vision frameworks. We propose the first Lightweight Hybrid Graph Convolutional Neural Network (LH-GCNN)-CNN framework as an add-on to object detection models. The proposed approach uses the output of the CNN object detection model to predict the observed scene type by generating a coherent GCNN representing the semantic and geometric content of the observed scene. This new method, applied to natural scenes, achieves an efficiency of over 90\% for scene classification in a COCO-derived dataset containing a large number of different scenes, while requiring fewer parameters than traditional CNN methods. For the benefit of the scientific community, we will make the source code publicly available: https://github.com/Aymanbegh/Hybrid-GCNN-CNN.

The recent development of deep learning methods applied to vision has enabled their increasing integration into real-world applications to perform complex Computer Vision (CV) tasks. However, image acquisition conditions have a major impact on the performance of high-level image processing. A possible solution to overcome these limitations is to artificially augment the training databases or to design deep learning models that are robust to signal distortions. We opt here for the first solution by enriching the database with complex and realistic distortions which were ignored until now in the existing databases. To this end, we built a new versatile database derived from the well-known MS-COCO database to which we applied local and global photo-realistic distortions. These new local distortions are generated by considering the scene context of the images that guarantees a high level of photo-realism. Distortions are generated by exploiting the depth information of the objects in the scene as well as their semantics. This guarantees a high level of photo-realism and allows to explore real scenarios ignored in conventional databases dedicated to various CV applications. Our versatile database offers an efficient solution to improve the robustness of various CV tasks such as Object Detection (OD), scene segmentation, and distortion-type classification methods. The image database, scene classification index, and distortion generation codes are publicly available \footnote{\url{https://github.com/Aymanbegh/CD-COCO}}