【作者】 李平； 章毅；

【Author】 Ping Li, Yi Zhang (Computational Intelligence Lab, University of Electronic Science and Technology of China, ChengDu 610054, China)

【机构】 电子科技大学计算智能实验室；

【摘要】 计算机多媒体技术的迅猛发展同时也向图像处理理论和相关技术提出了挑战。图像处理理论在人工智能、神经网络以及模糊逻辑等领域引起了广泛关注,相关的理论也不断取得新的突破,对于社会发展起到了非常重要的作用。更多还原

【Abstract】 The rapid progress of the computer multimedia technology challenges the theory and technology of image process. Image process has been widely focused on by the relative fields such as artificial intelligence, neural networks, and fuzzy logic, etc. Image segmentation is not only the key step of the subsequent image process and analysis, but also the basis of understanding the image contents. So the segmentation results directly affect the analysis results. In the past several decades, the algorithms based on statistics and neural networks have been developed in a certain extent. However, image segmentation still exists some problems, for instance, the brightness of pixels can not be well continuous because of illumination, the contrast between targets and background is not sufficiently distinctive, mostly existing algorithms partition the images only according to certain property, which are hard to apply to the other images, maybe to the same picture different regions have to be applied by different algorithms, and these algorithms is short of efficiency. Complex networks research show some lights on the statistical mechanism and the universal relationships among various real networks, at the same time, it confers powerful methods to analyze and solve problems in the other fields. In the past years, plenty of work on characterization of complex networks has been done to survey the measurements of network topology. With the help of the complex networks, image segmentation can be made by taking into account not only the raw image but also its enhanced representation as a complex networks. PCNN (Pulse Coupled Neural Networks) for image process has been proposed for a period of time, as it is characterized by its low time complexity and high efficiency. The most important properties of PCNN used in the image segmentation are that its kindred neurons will produce synchronized pulses. Any neuron’s firing must trigger its neighbors’ firing, thereby forming a firing cluster which just corresponds to a certain target region. So PCNN can be used for image segmentation. Considering the different characteristics between complex networks and PCNN, converting images into complex networks includes: assign all the image’s characteristics to the weights of the network; reduce the redundant information from image by setting up corresponding complex network model; detecting community in the network with PCNN, the community is just in correspondence with targets in the image.更多还原