Global–Local Progressive Integration and Semantic-Aligned Quality Transfer for No-Reference Image Quality Assessment

Accurate image quality assessment without reference signals presents a fundamental challenge in low-level visual tasks. In this article, we propose a global–local progressive integration model with three key contributions. (1) We develop a dual-feature extraction framework combining vision Transformer (ViT)-based global feature extractor and convolutional neural networks (CNNs)-based local feature extractor to capture image distortions at different granularities. (2) We propose a progressive feature integration scheme with multi-scale kernels to align global–local features, followed by channel-wise self-attention and spatial interaction for multi-grained representations. (3) We propose a semantic-aligned quality transfer (SAQT) method that extends the training data by assigning subjective quality scores to diverse image content. Experimental results demonstrate that our model yields 5.04% and 5.40% improvements in SROCC over the second-best state-of-the-art methods (i.e., DEIQT and CICI) for cross-authentic and cross-synthetic dataset generalization tests, respectively. Furthermore, the proposed SAQT method further yields 2.26% and 13.23% performance gains in evaluations on single-synthetic and cross-synthetic datasets. Code and dataset are available at: https://github.com/SYSU-Video/GlintIQA .