Customized K-Means Clustering Based Color Image Segmentation Measuring PRI

Abstract

Image segmentation has been considered as one of the most important tool for image processing in many application of computer vision, remote sensing, medical science and so on. The principle target of image segmentation is the way toward partitioning an image into various fragments that cover the whole image. However, it is very challenging task to segment or partitioning the images exactly. This paper proposes a customized k-means clustering based image segmentation method. This customized form of this technique implies k-means lustering algorithm, de-noising factor associated with velocity field of each pixel and edge distinguishing using Canny edge detector. For the evaluation, we conduct the experiment on the Berkeley segmentation dataset (BSDS300). Probabilistic rand index (PRI) is utilized for the similarity measure of segmented images and their predetermined ground truths. The experimental results reveal that the proposed method is highly capable of partitioning the images and in some cases, the outcomes of this method is better than the state-of-the-art methods.