|
International Journal of Information Technology Vol. 12 No. 2 2006 (Special Issue) Guest Editorial Special Issue on Image & Cognitive Signal Processing Image and cognitive signal processing have long been studied by researchers from a variety of disciplines including computer science, electrical engineering, psychophysics, optics and so on. Due to widespread, real-world applications and the easy availability of computers with high performance, many novel theories have emerged and the corresponding algorithms have been developed as a result in recent years. The Special Issue comprises papers selected from International Conference on Intelligent Computing 2005, held at He Fei, China. These papers demonstrate recent advances in image processing, covering various topics including image segmentation, image retrieval and encoding, object detection and tracking, image matching, recognition and classification etc. Several papers concerned with neural modelling are included as well. - Image segmentation:Mean shift, recently re-discovered, is a kernel based algorithm for image processing. Hong et al. present an improved mean shift algorithm for image segmentation. With adjustable bandwidth, direct density searching and a global criterion for mode merging used, the proposed method proves to outperform the conventional mean shift algorithm. Chen explores another very popular topic--Active Contour Model. After analyzing the shortcoming of one of Active Contour Models--Gradient Vector Flow (GVF), the author proposes an implicit method to solve ordinary differential equations, achieving a larger stability area. In the work of Jin et al, segmentation problem is treated as decomposition of gray-scale image into a number of peaks. The theoretical analysis is first given in detail, then the companying algorithm is presented, which is followed by an application to the measurement of the quantum dots in the AFM photos.
- Image retrieval and image encoding: Two papers in the special issue are concerned with image retrieval. Ontology based image retrieval is one popular research topic. In the first paper, authored by Kong et al., a new image retrieval system is presented, which is characterized by a personalized ontology and spatial ontology. Addressing image encoding problem, the work of Yun et al. introduces an efficient algorithm for shape encoding. Based on Progressive Vertex Selection method, new selection conditions are proposed, which lead to a smaller number of selected vertices.
- Object detection and tracking: Object detection has many potential applications in the real world. The paper of Liu et al. considers triangular object detection (e.g. traffic signs) in color images. In their paper line segments are detected and merged from the edge map, upon which unknown triangular objects are labelled via an Markov Random Field. The method can locate triangular shapes of different size and orientation. Object tracking in image sequence has received a lot of interest in recent years, especially object tracking dealing with non-linear non-Gaussian problem and dealing with multiple targets. Wang et al. bring forward a robust tracking algorithm based on particle filter, in which state evolution depends on edge information and Hausdorff distance while histogram similarity is used to measurement process. Another paper of Liu et al. propose a novel multiple object tracking algorithm in aerial image sequences. The target detection is automated by a elective attention mechanism making use of information of color, size and speed, and detected objects are followed with multiple Active Contour Models.
- Matching, recognition and classification: The problem of point-sets matching is discussed by Zhao et al. They propose a new algorithm in generating a graphical model and the corresponding Junction Tree for point-sets matching. The algorithm is more stable and accurate despite existence of outliers. Li et al. aim at distinguishing object by shape variability. To this end, they introduce structure moment invariants by transforming the density in geometric moments into a new density. Experiments demonstrate higher recognition rate using this method than that using tradition one. The paper by Tian et al. present a non-parameter and multi-scale approach for classification of natural clutter in synthetic aperture radar imagery. Multi-scale auto-regressive model is employed to characterize statistics of different terrain types, and Bootstrap technique is used to calculate paramors involved. Classification is done via comparison between the current estimate and that of each type.
- Other topics: Wang et al. analyze the frequency characteristics of the Daubechies N wavelet. They put forward an adjacent frequency bands energy increments comparison method, which could eliminate the interference of the frequency band energy leakage and optimize feature extraction. Furthermore, with fuzzy processing the frequency band which the feature frequency component belongs to can be determined.
- Neural modelling: Three papers related to neural modelling are also included in the special issue. The first paper by Feng et al. improves the generalization ability of neural networks based on Fuzzy theory and screen effect. The effectiveness of the algorithm consists in shrinking or magnifying of the input vector and reducing of the difference between the training and test sets. The second paper of Zhou et al. discuses the interpolation mechanism of functional networks, designing a kind of four layer and a five layer networks and proposing a learning algorithm based on the method of the least square error. In the third paper, Mao et al. introduce a novel uniform neural model of logic operator on generalized interval of [a,b]. The paper gives basic definition and theoretical analysis of the model, and make experiments to validate it.
Acknowledgement The guest editors would like to thank the organizing committee of ICIC2005. Without their work, the Special issue is impossible. The reviewers are acknowledged who responded to a very demanding schedule. Finally but not the least, the guest editors wish to thank the authors for their contributed papers and their collaboration during repeated revision process.
Peihua Li, Guest Editor School of Computer Science and Technology, Hei Long Jiang University, Harbin, Hei Long Jiang Province, China, 150080 E-mail: peihualj@hotmail.com Bo Ma, Guest Editor Department of Computer Science, City University of Hong Kong, HKSAR Email: bma000@cs.cityu.edu.hk Peihua Li
|