Image Segmentation Based On Fuzzy Theory And Spatial Information

Image segmentation is the basis of image analysis and computer vision application, andalso is a challenging and very difficult task in the field of digital image processing. Imagesegmentation has permeated through all walks of life and has important application in thesefields such as satellite remote sensing image recognition, spilled oil monitoring, productdetection, fingerprint recognition, optical characteristic recognition, notes identification, radarimage monitoring, medical image diagnosis, license plate information extraction and sealidentification. Up to now, scholars at home and abroad have proposed hundreds of imagesegmentation methods, however, there is no method that can be applied to all imagesegmentation and achieve satisfactory segmentation results.In this paper, firstly the development status of image segmentation methods is described,and traditional methods of image segmentation are studied further. Then, some improvedimage segmentation methods are proposed after problems encountered in practical applicationand disadvantages are analyzed. The main work of this paper is as follows:1. This chapter examines the non-local spatially adaptive threshold method, first pointedout the shortcomings of Otsu method, and then divided according to the proposed objectivesand background gray statistic for judging standard, gray histogram image multiple timessegmentation model in order to obtain optimal threshold, that is non-local adaptive thresholdmethod. The method of image using non-local information acquisition method and iterativeprocess for obtaining a threshold value matrix to determine its membership relations throughrelationships with neighboring pixel threshold matrix elements, the final image segmentation.Finally, morphological filtering process images to obtain more satisfactory test results.2. This paper examines image segmentation method based on generalized fuzzy entropy,and introduced segmentation methods based on fuzzy entropy proposed by De Luca andTermini at first. On the basis of affirmation of classical algorithms, its disadvantages arepointed out. Afterwards, an image segmentation algorithm based on generalized fuzzy entropyand other generalized fuzzy entropy segmentation algorithm based on complement operatorthat is the combination of generalized compliment and generalized fuzzy entropy areintroduced to address the problems aforementioned. This article improved the values of step length in generalized fuzzy entropy, then computed the image quality indicatorevaluation value. Put the new algorithm and the previously proposed generalized fuzzyentropy algorithm on comparison. It proved that parameter m is the key element of the imagesegmentation based on generalized fuzzy entropy and the choice of appropriate m is essentialfor segmentation results.3. The study of fuzzy C-means clustering algorithm based on the shadow set of thespatial information is made. Firstly, fuzzy C-means clustering algorithm based on the shadowset that is very effective in processing the border region of the image is introduced. Butbecause of its complete discard of image spatial information, this algorithm has lowprocessing capability in processing image with noises. Therefore, this paper proposed fuzzyC-means clustering algorithm based on the shadow set of spatial information, and comparedthese three cases of local space, non-local space, the combination of local and non-local space.Experiments show that proposed algorithm has better effect and superior performance thantraditional fuzzy C-means clustering algorithm based on the shadow set.