Novel texture-based image retrieval approach using variogram function

This paper presents a novel texture based image retrieval approach using the variogram function, which is often used in the field of geostatistics. Our method achieves the desired retrieval efficiency using a four-stage hierarchy: In the first stage, we randomly select some subimages where variogram functions are computed. And, the image is classified into regular texture category or irregular texture category based on its empirical variogram; In the second stage, we use change distances of empirical variogram function in the various directions to describe regular texture features, and use strict single step variogram values in various directions to describe irregular texture features; In the third stage, we utilize histogram to compute texture spectrum of the image and cluster the texture spectrum to create a personal texture table; In the last stage, the feature vector of is generated based on the personal texture table, and the image queries are based on those feature vectors. We have implemented an image retrieval system containing 4000 images based on the proposed algorithm. Experimental results demonstrate that our approach is promising.