Evaluating the average common submatrix measure for the similarity of real-world images

An image dissimilarity measure, based on the concept of average common sub-matrix, is presented. It considers the pixel matrices representing the images and the sub-matrices in common between them to quantify the dissimilarity. Here we briefly revisit the concepts underlying the ACSM measure and the algorithm to compute it, together with its computational complexity. Then we introduce a comparison of the measure with other state-of-the-art informationtheoretic measures on a real-world image data set, demonstrating the superiority of the ACSM measure with respect to the other measures in terms of retrieval precision.