Image retrieval system using block-based statistical features

This paper presents a color image retrieval system based on the statistical features of block partitioned image. A human perception based HSV quantization has been utilized for color histogram generation. First, an image is divided into several non-overlapping blocks. Then, the first and second moments of each block are extracted at the first stage. In order to reduce the feature vector dimension, statistical moments are then applied over extracted block feature vectors. The overall FV size of the proposed feature extraction technique is 18, which is independent of the block size chosen. The dissimilarity between two images is measured using Euclidean distance. In this paper, WANG image test database has been used to demonstrate the retrieval accuracy of the proposed system.