The Research On The Technique Of Image Processing Based On Fuzzy Arithmetic

Image Segmentation is the hot-topic of image processing. Especially, it is a key technique in image understanding, target recognition and track, etc. It will make a strong impact on performance of vision system. Image filtering is an important pre-processing step and its output effects the resolution of segmentation. Due to ignorance of information loss in image process, it is difficult to describe premise knowledge accurately and mathematically. Moreover, because of unknown factors in the process of image processing, there are likelihood and uncertainty between target and background. Fuzzy information processing technique developed from fuzzy set theory, have advantage to process uncertainty event and describe uncertainty knowledge. In this dissertation, image filtering and image segmentation using different image information based on fuzzy information process method is studied, and a series of new ideas and approaches are presented. The dissertation is organized as follows: Started with image filtering, a novel mixed filter is presented based on local statistic threshold after fully analysis literature and actuality. Firstly, this mixed filter classify all pixels into gauss-noise pixels and salt and pepper-noise pixels using a variable local threshold. Secondly, it use average filter to de-noise the gauss-noise pixels and use median filter to de-noise the salt and pepper-noise pixels. It has the advantage of average filter and median filter. A novel rule is added to preserve the details of image. It is easily to design because of the simple structure. Fuzzy divergence segmentation is a novel method proposed recently. It has a perspicuous concept and has a simple structure of mechanism. But it is difficult to initialize parameters when a image has a multi-peaks histogram. A novel method of image segmentation is presented in this dissertation combined with the principle of histogram smoothing and peaks detecting automatically. Firstly, this method smoothed the histogram. Secondly, it detects the peaks, the valley and the amount of the valley using the method of peaks detecting automatically. Coarse segmentation is achieved. Finally, the optimal threshold is obtained if we use the detected results to initialize the parameters. The refined segmentation is accomplished.