No-Reference Image Quality Assessment Based on Human Visual System and Dual-Branch Multi-Level Residual Network

This paper presents a No-Reference Image Quality Assessment (NR-IQA) method inspired by human visual perception. It integrates a dual-branch multi-level residual network that combines two complementary streams: one processing HSV images to capture content features aligned with human perception, and another utilizing contrast-sensitive weighted gradient (CSG) images to extract structural and texture features. This dual-branch architecture provides a comprehensive assessment of image quality from both content and structural perspectives. An advanced feature fusion strategy is introduced, employing a dedicated weight module to assign varying importance to content and structural features, ensuring an accurate final quality score. Experimental results on benchmark datasets, including LIVE, CSIQ, TID2013, LIVEC, and KonIQ-10k, showcase the superiority of our method over existing state-of-the-art NR-IQA techniques in key metrics like PLCC and SROCC. Our work holds significant practical value in advancing image quality assessment and has potential applications in multimedia compression, image restoration, enhancement, and related image processing technologies.