Anti-noise FCM image segmentation method based on quadratic polynomial

Abstract

Image segmentation is an important and challenging task in computer vision and image understanding, in which the segmentation model is the key to improve the quality of image segmentation. In this paper, a new fuzzy c-means clustering(FCM) algorithm based on quadratic polynomials is proposed, which can better distinguish the weak edge region in the image and has certain noise resistance. Firstly, the algorithm proposes to define the segmentation center using a quadratic polynomial surface,and divide the set of data points according to the algebraic distance between data points and the segmentation surfaces. The existing model with constant as the segmentation center is a special case with quadratic polynomial surface as the segmentation center, so the new model has higher segmentation accuracy. Secondly, based on the quadratic polynomial surface as the segmentation center, a new fuzzy factor is designed, and the deviation value is used to represent the difference between the mean algebraic distance of neighborhood points and the algebraic distance of center pixel. By calculating the deviation value, the influence of neighborhood points on the center point can be better measured and the segmentation accuracy can be improved. Experimental results show that the new algorithm has better anti-noise ability as well. Thirdly, select a local window on the edge of the global segmentation result for local window segmentation, which is equivalent to using a clustering center that is more in line with local information to do segmentation in a local small window, optimize misclassified pixels, and obtain the final segmentation result. The experimental results show that in the final segmentation results of medical images with 5% noise, segmentation accuracy can reach above 96%, partition coefficient Vpc has been increased by 0.14. So the algorithm can get the membership matrix with less ambiguity, which means more reliable segmentation results, and can effectively eliminate noise effects, retain the image details.