Abstract
This paper presents WhatAreYouLOOKing4 (WAY-LOOK4) system, a novel framework for content-based image retrieval (CBIR). Local descriptors are used to describe the visual contents of an image. Image signatures and similarity retrieval are based on the images' color and texture features. The main motivation of the system design is to use simple and efficient techniques to maintain reasonable computational and storage cost. The proposed technique has three system components: feature extraction, image database indexing and similarity retrieval. First, the use of circular sectors is proposed to represent local first order moment for the color feature. In addition, a local direction technique is used for texture feature extraction. Secondly, the hash indexing of the images' color properties is used to map the database images into classes. Hash indexing speeds up the search and enhances the system scalability for large image databases. Thirdly, for similarity retrieval, a degree of similarity is defined based on a weighted sum of the color and texture features. In addition, the similarity retrieval incorporates a minimum accepted degree of similarity provided by the user. The test of similarity is performed in two stages. In the first stage, the index is used to directly hit a class to which the query image may belong. In the second stage, a detailed sequential search is performed to retrieve the most similar images within that class. The simple design of the system and experimental selection of system parameters guarantee that the system maintains reasonable storage and computational cost. Our experiments demonstrate that the average precision of retrieved images is enhanced especially for higher accepted degrees of similarity.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have