A Haar Wavelet-Based Perceptual Similarity Index for Image Quality Assessment

In most practical situations, the compression or transmission of images andvideos creates distortions that will eventually be perceived by a humanobserver. Vice versa, image and video restoration techniques, such asinpainting or denoising, aim to enhance the quality of experience of humanviewers. Correctly assessing the similarity between an image and an undistortedreference image as subjectively experienced by a human viewer can thus lead tosignificant improvements in any transmission, compression, or restorationsystem. This paper introduces the Haar wavelet-based perceptual similarityindex (HaarPSI), a novel and computationally inexpensive similarity measure forfull reference image quality assessment. The HaarPSI utilizes the coefficientsobtained from a Haar wavelet decomposition to assess local similarities betweentwo images, as well as the relative importance of image areas. The consistencyof the HaarPSI with the human quality of experience was validated on four largebenchmark databases containing thousands of differently distorted images. Onthese databases, the HaarPSI achieves higher correlations with human opinionscores than state-of-the-art full reference similarity measures like thestructural similarity index (SSIM), the feature similarity index (FSIM), andthe visual saliency-based index (VSI). Along with the simple computationalstructure and the short execution time, these experimental results suggest ahigh applicability of the HaarPSI in real world tasks.