K-Means Clustering Using Localized Histogram Analysis

The first step required for many applications and algorithms is the segmentation or clustering of an image. This intuitively easy problem is actually very difficult for a computer to automatically perform. While having an effective solution to this problem would be valuable for all types of image processing, it would be especially useful in medical imaging. One method that addresses this problem is the K-means clustering algorithm. This paper presents a simple case of applying this algorithm to images, and extends it by redefining the distance measure to consider texture information. Results are shown for both algorithms, with the improvements from the modified algorithm clearly visible.