A study of feature determine-based compressed image retrieval

Abstract

The distance measure for features is of critical importance for all kind of classification methods in image retrieval algorithm. In this paper, we propose to introduce the Mahalanobis distance and additional Whirling correlation test to formulate a combined content similarity detection with discrete walsh transform (DWT). We show that, with the same content based image retrieval algorithm, adding Mahalanobis distance to the existing operational process can improve the performance of content based image retrieval significantly. Finally, In a series of experiments we show the improved noise robustness of image retrieval by the proposed modifications in contrast to the traditional approach.