Research On The Evaluation Of Underwater Image Quality And The Establishment Of Benchmark Database

Image processing technology is one of the most studied topics in the field of underwater image applications.Today,image quality assessment(IQA)plays a key role to advance the progress of underwater image processing technology.Unfortunately,there are a short-developed history and few studies about underwater image quality evaluation.Most researchers employed subjective assessment and objective no-reference assessment to evaluate the underwater image enhancement and restoration methods.A good objective underwater image quality indicator aims to assess the results in accordance with the human vision system(HVS)and color space characteristics,as well as achieves a maximum linear fit with the subjective evaluation results.In view of the shortage of underwater no-reference(NR)quality evaluation index proposed in recent years,in this paper,we study the characteristics of human keen perception of color,contrast and clarity,and propose a comprehensive no-reference underwater image quality assessment(UIQA)index based on HVS.UIQA is a linear combination of color saturation measurement component,contrast measurement component,and clarity measurement component.Experiments show that UIQA is suitable for underwater images in different scenes.For underwater image real-time processing,it can be regards as an accurate indicator between the distorted image and the enhancement result of the similar scene after processing to reflect the real visual perception of human beings.In addition,due to the lack of underwater benchmark dataset,the classic fullreference(FR)image quality evaluation index can hardly be applied to the evaluation of underwater image enhancement and restoration algorithms.In order to address these problems,we design a synthesized underwater image formation(SUIF)algorithm based on underwater imaging model.IN SUIF approach,we generate a synthesized underwater image(SUI)from a clear ground-truth image depending on a real underwater image.The proposed large-scale SUI dataset contains real atmospheric reference images and a simulated underwater image of the same scene.The SUI dataset can provide a novel criteria for the evaluation of underwater image enhancement and restoration algorithms,and promotes the development of underwater image vision.The main advantage of the proposed dataset is the presence of the clean image that can be regards as a ground-truth for full reference underwater image quality evaluation.