No-reference underwater image quality assessment based on Multi-Scale and mutual information analysis

Abstract

Due to the attenuation and scattering of light in the water, there is a serious degradation for the quality of underwater imaging, severely hindering underwater exploration and research. Therefore, implementing quality assessment is crucial for the application of underwater visual tasks. To effectively assess the quality of underwater images, a novel no-reference underwater image quality assessment based on multi-scale and mutual information analysis (MMIQA) is proposed. Specifically, considering the issues of color cast and the importance of colorfulness in underwater images, chroma difference maps and chroma saturation maps were created based on chroma components. The statistical features of these maps were then extracted at multiple scales as chroma component features. Additionally, considering the importance of texture and structure, the multi-scale fractal dimension and high-frequency sub-band energy distribution features of the luminance component were extracted as statistical features of multi-scale underwater local texture and structure. Finally, considering the correlation between the chroma and luminance components of the image, the mutual information between chroma and luminance, as well as between luminance sub-band images, was extracted as a statistical measure of underwater mutual information distribution. Experimental results show that, compared to state-of-the-art methods, the proposed MMIQA has the highest correlation with actual quality scores.