Features Analysis for Content-Based Image Retrieval Based on Color Moments

In this study an efficient and accurate algorithm is proposed for Content-Based Image Retrieval (CBIR). The CBIR is performed in two steps: features extraction in images and similarity measurement for searching of similar images in image database. For efficient and effective CBIR system the features extraction must be fast and the searching must be accurate. In the proposed algorithm, the effective retrieval of the similar images from the database is based on the efficient extraction of the local statistical color moment features without using the spatial features of images. The basic idea in this algorithm is to convert the color RGB (Red Green and Blue) image into grayscale image to reduce the computations in feature extraction and to increase the efficiency. The grayscale image is divided into non-overlapping blocks of different sizes. The local statistical color moment features are extracted in all blocks. The features are combined into a feature vector. The similarity is measured by using Sum-of-Absolute Difference (SAD) to measure the similarity between query image and database images. In the experiment, the efficiency of feature extraction and accuracy of the image retrieval are measured for different block size methods using the proposed algorithm. The Corel database is used for testing. The results show that the proposed CBIR algorithm provides higher performance in terms of efficiency and accuracy.