Based On The K-nearest Neighbor Fast Region Merging Image Segmentation Algorithms And Application

Image segmentation is one of classic problem in image processing.With the increase demand for image processing,image segmentation draws more and more researchers' attention.However, all the existing image segmentation algorithms have shortcomings,and need compromise between efficiency and effect.Furthermore,it is very hard to find an algorithm which fits all kinds of images.However,this is just our goal in this paper.The contents in this paper include:1.We focus on local relationship between neighbor pixels in image with the goal that all similar pixels should be segmented in the same region.A new region similarity measure function is proposed,which can make use of pixel intensity,edge feature,texture and so forth in a unit form.2.A fast region merging method based on the region similarity is proposed for solving the image segmentation problem.The image segmentation problem here is treated as a region merging procedure.To solve it,an initial oversegmentation is performed on the image and a k-Nearest Neighbor(k-NN)Graph whose vertexes denote regions is built.And then the region similarity can be assigned to the edge as its weight.In k-NN graph,each vertex chooses exactly k nearest neighbors to connect.With it,the computation complexity of merging process can be reduced to 0(τK log₂(K));here,τdenotes the number of nearest neighbor updates required at each merging while K denotes the number of the initial regions.3.Based on the intrinsic images division algorithm,design a classifier to divide an image into two parts,one includes the illuminance,and the other includes the color and shape.Then delete the non-uniform lighting from the image.With our proposed segmentation algorithm,the non-uniform illuminance image segmentation problem is solved.4.Implementation of the all proposed algorithms is introduced,and some experiment results are given to prove our method's robustness and efficiency.Besides,we propose an automated algorithm to detect exudates based on our image segmentation algorithm.The running time of the detect algorithm is quite low,and an average specialty of 95.42%with the average sensitivity of 91.08%is obtained in 8 images with exudates from STARE database.This method is likely helpful in a clinical environment.