No‐reference image quality assessment via a dual‐branch residual network
Peng Ji,Chang Liu,Hao Chen
DOI: https://doi.org/10.1049/ipr2.13057
IF: 2.3
2024-02-23
IET Image Processing
Abstract:This manuscript starts from the perspective of the structural and color information of distorted images and explores a No‐Reference Image Quality Assessment (NR‐IQA) method.We conducted experiments on four publicly IQA datasets, and the proposed method exhibits strong performance and generalization capabilities in predicting image quality compared to peer methods. The No‐Reference Image Quality Assessment (NR‐IQA) method can predict quality scores of distorted images without the reference image. However, due to the variability in both color and structure of images, existing NR‐IQA methods struggle to accurately predict quality scores of distorted images. Therefore, an NR‐IQA method based on a Dual‐Branch Residual Network (DBRIQA) for evaluating the quality scores of color‐distorted images is proposed. First, guided filtering is applied to the hue images in the HSV space to extract distortion information from the image's color. Then, due to significant differences in distorted images, enhancements are made to the traditional residual blocks, forming a feature extraction module that captures multi‐scale features from the image. In order to capture the global relationships in the distorted image, a Global‐Level Attention Block (GLAB) is introduced, facilitating the interaction of information among the extracted features. Experiments were conducted across four publicly available IQA datasets, including LIVE, CSIQ, TID2008, and TID2013. The experimental results demonstrate that the proposed method exhibits strong performance and generalization capabilities in predicting image quality compared to peer methods.
computer science, artificial intelligence,engineering, electrical & electronic,imaging science & photographic technology