Image quality assessment for a selective-processing noise-aided iterative enhancement algorithm

Abstract

This paper presents a revision of the existing universal image quality index (UIQ) metric in order to gauge the quality of images in a selective-processing iterative enhancement algorithm. The original UIQ is based on factors of loss of correlation, luminance and contrast similarity, and is, therefore, unsuitable in enhancement-related applications. While testing existing state-of-the-art metrics as image quality criterion in an iterative dynamic range compression algorithm, a lack of coherence was observed between the objective scores and that obtained from subjective evaluation study with twenty human subjects. We, therefore, propose to modify the existing UIQ with properties of a tone-mapped image. The proposed variant, named image quality metric for dynamic range compression, IQDRC, maintains the contributing effect of structural correlation and local contrast similarity, but observes an inverse relation with local luminance similarity. The proposed metric was observed to promisingly quantify the image quality and dynamic range compression of such images in close accordance with subjective scores for the target enhancement algorithm. Observations also suggest that IQDRC is indicative of image quality for various other dynamic range compression algorithms.