ABSTRACT
Content-based image retrieval is an automated method that involves extraction of inherent content of the image. In this method, feature vectors generated using a single pixel provides less information as compared to a group of pixels. Correlation among the pixels of a neighborhood represents more useful information about the structure of texture. Local binary patterns are proved to be effective in deriving texture information in an effective manner for any practical system. Local extremas are calculated in increased neighborhood of a pixel in the image. Color and texture are integrated with directional local extrema patterns to enhance performance. Different types of image databases such as Corel-1k, Corel-5k, Corel-10k, ImageNet-25k and MIT VisTex are used to test the effectiveness of the proposed methods. Euclidean distance, D1 distance, Manhattan distance and Canberra distance are used for comparison purposes.
