Underwater optical image quality assessment via chrominance-texture fusion

Abstract

Underwater optical image degradation caused by light absorption, scattering, and color distortion makes accurate assessment of underwater optical image quality a challenging task. Most existing underwater image quality assessment (UIQA) methods focus on limited visual features, overlooking critical characteristics such as texture and saliency, leading to suboptimal performance in complex scenarios. This is particularly evident in cases with severe red color cast, where existing methods struggle to effectively evaluate image quality. To address these issues, we propose a novel and efficient UIQA method that improves prediction accuracy and robustness through multi-feature fusion. Our method extracts critical features from the luminance space, chrominance space, and saliency maps to capture the multidimensional image degradation information. We designed a texture feature extraction method based on the YCbCr color space and gray-level co-occurrence matrix (GLCM), effectively separating and measuring the color and texture of images, providing an accurate description of degradation characteristics. To address the common red tint problem in underwater image enhancement, we introduce a red cast feature extraction strategy that refines chrominance modeling and incorporates saliency map features, effectively reducing color distortion and improving the model’s predictive performance. Experiments conducted on two standard underwater image quality datasets, SAUD and UID, demonstrate that our method outperforms existing state-of-the-art UIQA models across multiple evaluation metrics, particularly in complex underwater scenes, exhibiting superior prediction accuracy, stability, and generalization capability.