Approximate Matching in ACSM Dissimilarity Measure

Abstract

The paper introduces a new patch-based dissimilarity measure for image comparison employing an approximation strategy. It extends the Average Common Sub-matrix measure computing the exact dissimilarity among images. In the exact method, dissimilarity between two images is obtained by considering the average area of the biggest square sub-matrices in common between the images, by exact match of the extracted sub-matrices pixel by pixel. As an extension, the proposed dissimilarity measure computes an approximate match between the sub-matrices, which is obtained by omitting a controlled number of pixels at a given column offset inside the sub-matrices. The proposed dissimilarity measure is extensively compared with other well-known approximate methods for image comparison in the state-of-the-art. Experiments demonstrate the superiority of the proposed approximate measure in terms of execution time with respect to the exact method, and in terms of retrieval precision with respect to the other state-of-the-art methods.