Blind Image Quality Evaluator with Scale Robustness

Abstract

Image quality assessment (IQA) algorithms usually need define some visual-related features and build their relationship with image visual quality. Regardless of whether the features are extracted in spatial or frequency domain, these features are often sensitive to image resolution, resulting in that existing no-reference (NR) IQA algorithms are only efficient for images with similar resolution. In this paper, we propose a more generalized blind image quality evaluator with scale robustness (BIQESR) to assess image quality by locating the robust feature points in a multi-scale space. Compared with the state-of-the-art NR IQA methods, BIQESR has higher performance and stability for images with significantly different resolutions.