A new subjective procedure for evaluation and development of texture similarity metrics

Abstract

In order to facilitate the development of objective texture similarity metrics and to evaluate their performance, one needs a large texture database accurately labeled with perceived similarities between images. We propose ViSiProG, a new Visual Similarity by Progressive Grouping procedure for conducting subjective experiments that organizes a texture database into clusters of visually similar images. The grouping is based on visual blending, and greatly simplifies pairwise labeling. ViSiProG collects subjective data in an efficient and effective manner, so that a relatively large database of textures can be accommodated. Experimental results and comparisons with structural texture similarity metrics demonstrate both the effectiveness of the proposed subjective testing procedure and the performance of the metrics.