Perceptual similarity metrics for retrieval of natural textures

Abstract

Abstract-We investigate perceptual similarity metrics for the content-based retrieval of natural textures. The goal is to find perceptually similar textures that may have significant differences on a point-by-point basis. The evaluation of such metrics typically requires extensive and cumbersome subjective tests. The focus of this paper is on the recovery of textures that are "identical" to the query texture, in the sense that they are pieces of the same texture. This is important in content-based image retrieval (CBIR), where one may want to find images that contain a particular texture, as well as in some near-threshold coding applications. The advantage of evaluating metric performance in the context of retrieving identical textures is that the ground truth is known, and therefore no subjective tests are required. We can thus compare the performance of different metrics on large sets of textures, and derive meaningful statistical results. We evaluate the performance of a recently proposed structural texture similarity metric on grayscale textures, and compare it to that of PSNR, as well as space domain and complex wavelet structural similarity metrics. Experimental results with a database of 748 distinct texture images, indicate that the new metric outperforms the other metrics in the retrieval of identical textures, according to a number of standard statistical measures.

Image analysis and compression: renewed focus on texture

“…46 and precision-recall plots 47 are also commonly used. Zujovic et al 32 performed extensive experiments for this type of application, using nearly 280,000 pairs of texture images, to assess the performance of their metric to the well known SSIM and CW-SSIM metrics, and demonstrated that it offers considerable advantages.…”

“…32. The conclusion was that STSIM-2 performs best, with 77.2% success in retrieving the correct document as the first returned result (precision at one), whereas for PSNR this happens only for 6% of the images.…”

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

“…For example, in [4], the focus was on the recovery of textures that are "identical" to the query texture, in the sense that they are pieces of the same texture. All that is needed in this case is to start with a database consisting of perceptually uniform textures, which we can then cut into (perhaps partially overlapping) pieces, in order to obtain the test database.…”

“…In the information retrieval community this known as the known-item search [5]. Common measures for this type of retrieval systems include precision at one (measures in how many cases the first retrieved document is relevant), mean reciprocal rank (measures how far away from the first retrieved document is the first relevant one), mean average precision and precision-recall plots [4].…”

“…However, our previous work indicates [3,4] that the two attributes should be considered separately. Thus, in the present study we focus only on grayscale textures.…”