| Image segmentation is an important research in the low level of computer vision,which lays the foundation for image analysis, understanding, and code. However, mostthresholding methods only consider the gray-scale property of image without investigatingthe spatial relationship of pixels, as a result, segmentation results are discontinuous andinaccurate. In recent years, many algorithms are proposed based on fuzzy theories for theirsuperiority in processing vagueness incidents, but they involve lots of index.Based on the above analysis, the paper takes the spatial relationship and the distancesof within-class and between-class into consideration, designs algorithms with highefficiency. The main work of our paper is described as follows.Firstly, this paper has summarized the procedures of fuzzy theories applied to theimage segmentation algorithms: firstly, change the digital image into the form of fuzzymathematical definition using the fuzzy theory; secondly, modify the membership valuesgenerated by the membership function and design a criteria to obtain the thresholdingvalue of fuzzy image; last, change the fuzzy image into digital image.Secondly, this paper advocates a weighted thresholding algorithm based on the fuzzytheory and takes account of gray-scale property and spatial information, which utilizes thesimilarity relationship of pixels and its neighbor pixels. It considers highly the effects ofspatial relationship among pixels on classifying.Thirdly, an automatic global thresholding algorithm is simultaneously proposed, whichadds the spatial geometry distances between pixel and the given thresholding value in thedesign of membership function. Furthermore, a new criteria for thresholding selection isalso given with the mean and variance of different classes in mind, which can takedistances of between-class and within-class into consideration.Last, we test the validity of our experiments, compare with other recent proposedmethods and also analysis the experimental results. |
PDF Full Text Request