Meshing local watershed-based fuzzy clustering method

Abstract

The invention discloses a meshing local watershed-based fuzzy clustering method. According to the method, a superpixel thought and a watershed algorithm are combined, and improvement is carried out onthe basis of a fuzzy C-means clustering algorithm (Fuzzy c-means), so that the ability and efficiency of capturing image salient regions by using an image segmentation algorithm and the denoising ability are improved. The method comprises the following steps of: carrying out non-uniform meshing on an image according to a region grey level value variance; carrying out a watershed algorithm of a local optimum threshold value on each mesh so as to decrease local information loss caused by a global watershed; obtaining a salient water accumulation basin in each mesh; carrying out region fusion and equalizing the grey level of each marked region; and finally carrying out clustering by using a region area-considered FCM to obtain a final segmented image. The method is strong in robustness for noise, is capable of effectively eliminating interference regions and segmenting salient regions in images, and has relatively low time complexity at the same time.

Description

【Technical field】 [0001] The invention relates to the technical field of image processing, and relates to image segmentation. 【Background technique】 [0002] Image segmentation is a key preprocessing process in image recognition and computer vision, and remains a challenging research in image analysis. Its goal is to segment similar and adjacent pixels into corresponding structurally coherent elements in an image. According to the gray value, texture, color, etc. of the pixels in the image, the image is divided into several irrelevant regions. Each region has its similarity inside, and different regions have differences. Image segmentation can be considered as the process of dividing an image into several meaningful sub-regions based on homogeneity or heterogeneity criteria. [0003] Since the models to be segmented are different, the segmentation methods suitable for different models also have their own advantages and disadvantages. Nowadays, image segmentation methods m...

AI Technical Summary

Problems solved by technology

The watershed algorithm was first proposed by Beucher et al. Then in 1991, Vincent and Soille improved on the original basis and proposed a more classic watershed algorithm. This algorithm has fast calculation speed, accurate positioning, and relatively stable results, which can form A closed target contour, but there are still deficiencies, easy to fall into local optimal segmentation, over-segmentation, sensitive to noise

In order to solve the over-segmentation phenomenon of the watershed, gradient thresholding is often used, but the global watershed gradient threshold is easy to cause the loss of local information of the image

Images