Hau-San Wong

Adaptive Image Processing: A Computational Intelligence Perspective, Second Edition

Illustrating essential aspects of adaptive image processing from a computational intelligence viewpoint, the second edition of Adaptive Image Processing: A Computational Intelligence Perspective provides an authoritative and detailed account of computational intelligence (CI) methods and algorithms for adaptive image processing in regularization, edge detection, and early vision. With three new chapters and updated information throughout, the new edition of this popular reference includes substantial new material that focuses on applications of advanced CI techniques in image processing applications. It introduces new concepts and frameworks that demonstrate how neural networks, support vector machines, fuzzy logic, and evolutionary algorithms can be used to address new challenges in image processing, including low-level image processing, visual content analysis, feature extraction, and pattern recognition. New to the Second Edition: A new chapter on a family of unsupervised algorithms with a basis in self-organization yet somewhat free from many of the constraints typical of other well-known self-organizing architectures New material on recent challenges in image content analysis and classification, including small sample problems and fuzzy user perception A new technique in visual query processing and visualization in 2D space New experiments and updates on perceptual error-based restoration Emphasizing developments in state-of-the-art CI techniques, such as content-based image retrieval, this book continues to provide educators, students, researchers, engineers, and technical managers in visual information processing with the up-to-date understanding required to address contemporary challenges in image content processing and analysis.

A low-complexity contextual Hebbian detector for blind multiuser detection

This paper proposes a blind multiuser detector for CDMA systems based on a contextual Hebbian paradigm. Conventional blind detectors employ second-order statistics in their formulation, leading to first-order filter update procedure. These approaches restrict the convergence rate and tracking capability of the detectors. Hebbian learning has shown potential in handling blind source separation problems. Nevertheless, it experiences order ambiguity of the extracted sources. This often leads to undesirable local convergence and consequently erroneous symbol demodulation. This paper presents a new contextual Hebbian paradigm that encapsulates domain information of the multiple-access interference to achieve blind detection. An adaptive detector is developed to address the issues of source indeterminacy and slow convergence. Experimental results show that the detector provides good performance in terms of fast convergence rate, optimal steady-state SINR profile, and low BER when compared with other detectors.