New non-reference image quality evaluation method for underwater turbulence blurred images

Abstract

Turbulence is an important cause of image distortion in an underwater ocean environment, along with light scattering and color reduction. Turbulence blur is a universal phenomenon of underwater image degradation, and causes a loss of high frequency portion and image detail. We need a metric to evaluate underwater image quality and provide a reference for the restoration of turbulence-blurred images. It is difficult to obtain clear original images, and a subjective quality metric is time-consuming and impractical for real-time implementation, so we propose a new no-reference, objective quality evaluation algorithm, by employing CIELab color space features and mean subtracted contrast normalized (MSCN) statistical features to assess the quality of blurred images. The experimental results illustrate that our metric outperforms the other advanced image-quality metrics in the underwater turbulent-image dataset, and has a comparable performance in other datasets. Importantly, besides being highly in line with human perception, the proposed metric can effectively predict image quality with low computational complexity and meet the requirement of a real-time system.