Evaluation of Mean Shift Algorithm as Applied to Image Segmentation

Abstract

This thesis work implements three segmentation algorithms - mean shift algorithm, CMeer clustering, and K-means clustering. Three color spaces, RGB, HSV, and Luv, and two texture features, Gabor and Blobworld, are used with each of the segmentation algorithms. Thus twenty-one experiments are designed. The performance of these experiments are first evaluated based on their consistency with human segmented images. Secondly, these segmentation techniques are used in conjunction with a content based image retrieval (CBIR) algorithm and are evaluated based on their retrieval precision. A possible correlation between the performance of these experiments under the two evaluation methods is also looked into.