A quality assessment algorithm for no-reference images based on transfer learning

• DOI: 10.7717/peerj-cs.2654
• Publication Date: 2025-01-31T08:26:07Z
• Authors: Yang Yang; Chang Liu; Hui Wu; Dingguo Yu
• Keywords: Transfer of learning; Feature (linguistics)

Image quality assessment (IQA) plays a critical role in automatically detecting and correcting defects images, thereby enhancing the overall performance of image processing transmission systems. While research on reference-based IQA is well-established, studies no-reference remain underdeveloped. In this article, we propose novel algorithm based transfer learning (IQA-NRTL). This leverages deep convolutional neural network (CNN) due to its ability effectively capture multi-scale semantic information features, which are essential for representing complex visual perception images. These features extracted through module. Subsequently, an adaptive fusion integrates these fully connected regression correlates fused with global perform final assessment. Experimental results authentically distorted datasets (KonIQ-10k, BIQ2021), synthetically (LIVE, TID2013), artificial intelligence (AI)-generated content dataset (AGIQA-1K) show that proposed IQA-NRTL significantly improves compared mainstream algorithms, depending variations complexity.