The WKNN-MPFCM Algorithm For HD Color Image Segmentation

Fuzzy C-Means(FCM)Clustering algorithm is a kind of unsupervised Clustering algorithm based on division.It is also one of the more common image segmentation algorithms.It carries out image segmentation by searching for Fuzzy membership level between 0 and 1.The goal of FCM is to find the clustering center in the feature space to minimize the objective function.But it has its limits.We want to achieve better image segmentation effect by improving its shortcomings.In this paper,a weighted k-nearest neighbor-Modified Penalty Fuzzy C-means(WKNN-MPFCM)Clustering algorithm for HD color image segmentation is proposed.The algorithm solves the problem that a small part of the pixels is wrongly segmented due to Fuzzy C-Means(FCM)Clustering fails to consider surrounding spatial factors when processing image segmentation.The weighted k-nearest neighbor(WKNN)algorithm corrects the membership value of the current pixel by considering the relationship between the pixels in the surrounding d ?d window.The Penalty Fuzzy C-means(MPFCM)Clustering algorithm is developed by modifying the objective function of the standard FCM algorithm with a penalty term that takes into account the influence of the neighboring d ?d pixels on the center pixels.So the small areas which are misclassified is corrected by combining WKNN and MPFCM.This paper also compares the influence of different features on FCM and WKNN-MPFCM algorithm.By comparing the colorless features,the colorless features RGB and the colorless features Lab,the experiment shows that different features have a certain impact on FCM and WKNN-MPFCM algorithm,especially the color feature Lab.At the same time,the effects of different weights of Lab features on images with big color difference in the background or target are compared.The experiment shows that the Lab features with different weights have a certain impact on the FCM algorithm.Finally,the membership matrix of WKNN-MPFCM with different features is weighted to obtain better image segmentation results.The real-time performance of FCM algorithm is not very ideal.Therefore,I also applied the mini-batch method to the FCM and WKNN-MPFCM algorithm to accelerate the convergence speed of the FCM and WKNN-MPFCM algorithm,improve the efficiency of the algorithm,and thus save time.