AERASCIS: An efficient and robust approach for satellite color image segmentation

Satellite color images carry a vast amount of information which needs an efficient image segmentation method to analyze. Because of its simplicity and low complexity, K-Means algorithm is frequently adopted for color image segmentation. But, usually the results of K-Means algorithm suffers from noises and hence over segmentation. This is due to the reasons that K-Means works on the basis of random K initialization and “Euclidean Distance Metric” as default. Also, in the case of satellite color image segmentation local contrast management is an important issue which is not paid attention in the traditional K-Means algorithm. So, in this paper, these problems are taken into consideration and a robust method has been proposed to tackle the same. First of all, HSV color space is chosen for color based transformation and calculations. Here, a Binary Search Based CLAHE is introduced for local contrast management. An entropy based technique is developed for determining the total number clusters and detection of initial centers of the clusters. “Cosine Distance Metric” is employed for distance based calculations involved in K-Means algorithm. The performance of the proposed approach is found robust with respect to noise and over-segmentation is removed up to a satisfactory level.