Springer Science and Business Media LLC

There has been less attention towards the research on face recognition with partial occlusion. Facial accessories such as masks, sunglasses, and caps, etc., cause partial occlusion which results in a significant performance drop of the face recognition system. In this paper, a novel compressed string matching algorithm based on run-length encoding (CSM-RL) is proposed to solve the partial occlusion problem. In this, the face image is represented by a string sequence that is then compressed using run-length encoding. The proposed CSM-RL algorithm performs string matching between query face and gallery face string sequences by computing the edit distance between string sequences, finally, classifies query face based on the minimum edit distance. The proposed method does not require a classifier and has less time complexity, thus it is more suitable for real-world face recognition applications. The proposed method performs better than the state-of-the-art methods even limited sample images per person are available in the gallery. Extensive experimental results on benchmark face datasets such as AR and Extended Yale-B prove that the proposed algorithm exhibits significant performance improvement both in terms of speed and recognition accuracy for the recognition of partially occluded faces.

Video provides a comprehensive visual record of environment activity over time. Thus, video data is an attractive source of information for the creation of virtual worlds which require some real-world fidelity. This paper describes the use of multiple streams of video data for the creation of immersive virtual environments. We outline our multiple perspective interactive video (MPI-Video) architecture which provides the infrastructure for the processing and analysis of multiple streams of video data. Our MPI-Video system performs automated analysis of the raw video and constructs a model of the environment and object activity within this environment. This model provides a comprehensive representation of the world monitored by the cameras which, in turn, can be used in the construction of a virtual world. In addition, using the information produced and maintained by the MPI-Video system, our immersive video system generates virtual video sequences. These are sequences of the dynamic environment from an arbitrary view point generated using the real camera data. Such sequences allow a user to navigate through the environment and provide a sense of immersion in the scene. We discuss results from our MPI-Video prototype, outline algorithms for the construction of virtual views and provide examples of a variety of such immersive video sequences.

This paper explores the issue of user embodiment within collaborative virtual environments. By user embodiment we mean the provision of users with appropriate body images so as to represent them to others and also to themselves. By collaborative virtual environments we mean multi-user virtual reality systems which explicitly support cooperative work (although we argue that the results of our exploration may also be applied to other kinds of collaborative system). The main part of the paper identifies a list of embodiment design issues grouped by the general themes of personal representation, conveying activity, embodiment in heterogeneous systems, embodiment of agents, and ethical issues. These issues are illustrated with examples from our own DIVE and MASSIVE collaborative virtual environments. The paper also uses this set of issues as an analytical framework for comparing a number of other communication technologies.

Face verification models based on centralized training on large face datasets have achieved excellent performance on various test benchmarks. However, due to the increasingly sophisticated privacy protection law, centrally collecting large amount of face images becomes more difficult. We consider learning a face verification model in the federated setting, where each client has access to the face images of only one class and class embeddings cannot be shared to other clients because of data privacy. In this paper, we propose Federated face verification (FedFV), in which server transfers some equivalent class embeddings to clients so that the clients’ class embeddings can be separated far away from each other. We show that our proposed method FedFV outperforms the existing approaches in several face verification benchmarks.

Original medical images are often inadequate for clinical diagnosis. Certain prior information can be used as an important basis for disease diagnosis and prevention. In this study, an oriented grouping-constrained spectral clustering method, OGCSC, is proposed to deal with medical image segmentation problems. OGCSC propagates the group information from the affinity matrix and subdivides the group information into two constraints. By adopting the normalized framework, OGCSC can be transformed into normalized spectral clustering. The solution of OGSCSC can be viewed as a generalized eigenvalue problem that can be solved using eigenvalue decomposition techniques. The significance of our work is that the use of group information and constraints information to analyse image data can greatly enhance the results achieved using the clustering segmentation method. The empirical experimental results reveal that the proposed method achieves robust and effective performance for medical image segmentation.

The aim of this work is to present an adaptive serious game based on virtual reality (VR) for functional rehabilitation of the hand after stroke. The game focuses on simulating the palmar grasping exercise commonly used in clinical settings. The system’s design follows a user-centered approach, involving close collaboration with functional rehabilitation specialists and stroke patients. It uses the Leap motion controller to enable patient interaction in the virtual environment, which was created using the Unity 3D game engine. The system relies on hand gestures involving opening and closing movements to interact with virtual objects. It incorporates parameters to objectively measure participants’ performance throughout the game session. These metrics are used to personalize the game’s difficulty to each patient’s motor skills. To do this, we implemented an approach that dynamically adjusts the difficulty of the exergame according to the patient’s performance during the game session. To achieve this, we used an unsupervised machine learning technique known as clustering, in particular using the K-means algorithm. By applying this technique, we were able to classify patients’ performance into distinct groups, enabling us to assess their skill level and adapt the difficulty of the game accordingly. To evaluate the system’s effectiveness and reliability, we conducted a subjective evaluation involving 11 stroke patients. The standardized System Usability Scale (SUS) questionnaire was used to assess the system’s ease of use, while the Intrinsic Motivation Inventory (IMI) was used to evaluate the participants’ subjective experience with the system. Evaluations showed that our proposed system is usable and acceptable on a C-level scale, with a good adjective score, and the patients perceived a high intrinsic motivation.

Recently, mobile TV has been launched in several countries. While mobile TV integrates television contents into mobile phones, the most personal of communication devices, it becomes interesting to know how this feature will be used throughout the day and in varying contexts of everyday life. This paper presents empirical results on the use of mobile TV with different delivery mechanisms and both quantitative and qualitative results on how end-users prefer to use mobile TV contents in different situations. The data is based on ongoing empirical research in Finland in 2006 and 2007. The mobile TV services under study included both news and entertainment contents, and were tested in 3G, DVB-H and Wi-Fi networks using different delivery paradigms: broadcast, on-demand and download. To explore the use of different delivery methods and content consumption, we have developed a mobile TV service protoype, called Podracing. The analysis shows that users appreciated up-to-date information and information-rich media forms and contents especially for mobile news delivery. There was high demand for only the latest news on mobiles. The real-time property was considered important. Most of the users looked at the headlines or followed the news several times a day – much more often than the traditional TV and news prime times would allow.

A novel objective evaluation method for image fusion based on target quality factor is proposed from the point of view of target recognition. Three-component indicators named preservation degree of special features of the target, target edge quality and interference edge suppression ratio are defined. These component indicators can evaluate the quality of the target from three different aspects. Preservation degree of special features of the target quantitatively describes the preservation degree of special features in the fusion image, which are very important for target recognition in the special source images. Target edge quality is used to evaluate the integrity and quality of the boundary information contained in the fusion image. Interference edge suppression ratio quantitatively describes the boundary information whether the fusion image will produce confusion. Target quality factor is obtained by weighted averaging of these component indicators. The experimental results show that the target quality factor can evaluate the image fusion for target recognition quantitatively and reasonably, and the evaluation results are in accordance with the visual effect of the fusion images.

Multi-view stereo (MVS) methods based on deep learning have developed rapidly in recent years, but inaccuracies in reconstruction due to the general effect of feature extraction and poor correlation between cost volumes are still present, opening possibilities for improvement in reconstruction accuracy and completeness. We therefore develop a hierarchical MVS network model with cost volume separation and fusion to mitigate these problems. First, to obtain a more complete and accurate feature information from the input images, a U-shape feature extraction module was designed that outputs feature information simultaneously according to a hierarchical structure composed of three different scales. Then, to enhance the learning ability of the network structure for features, we introduced attention mechanisms to the extracted features that focus on and learn the highlighted features. Finally, in the cost volume regularization stage, a cost volume separation and fusion module was designed in the structure of a hierarchical cascade. This module separates the information within the small-scale cost volume, passes it to the lower level cost volume for fusion, and performs a coarse-to-fine depth map estimation. This model results in substantial improvements in reconstruction accuracy and completeness. The results of extensive experiments on the DTU dataset show that our method performs better than Cascade-MVSNet by about 10.2% in accuracy error (acc.), 7.6% in completeness error (comp.), and 9.0% in overall error (overall), with similar performance in the reconstruction completeness, showing the validity of our module.