A Training-Based Blind Underwater Image Quality Evaluation Metric

Due to the unique image formation principle and the limitations of imaging devices, underwater images usually suffer from low contrast, color degradation, and blurring effects, which seriously hinder the interpretation of the image content. Additionally, several underwater image enhancement and restoration (UIER) algorithms have been proposed to improve the quality of underwater images. However, their performances vary greatly under different underwater scenarios. Therefore, establishing an effective quality evaluation metric plays an irreplaceable role in assessing the underwater image quality as well as evaluating the performances of UIER algorithms. In this paper, we propose a novel training-based blind underwater image quality elevation metric. Technically, the proposed metric extracts and fuses three groups of features covering naturalness, color, and contrast. Moreover, observing that the severity of color cast is highly related to the chroma of the image, we develop an underwater color factor feature to accurately estimate the severity of color distortion. Experimental results demonstrate the superior performance of the proposed method and its ability to evaluate the performance of UIER algorithms.