Research On No-reference Quality Assessment Method For Underwater Image

As an intuitive carrier of underwater vision,underwater images play an important significance role in exploration and development of underwater world.However,due to the limitation of lighting conditions and the influence of light selective absorption and scattering in water,the captured original underwater images usually suffer from various degradations,such as low contrast,blurry details,poor charity,uneven bright spots,color distortion,complex noise,etc.These degradations greatly reduce the amount of underwater scene information culminating into various degraded versions of underwater image quality.Moreover,the degradation of underwater image quality not only seriously affects the interpretation of the image content,but also greatly reduces the accuracy of high level visual applications.Consequently,how to effectively evaluate the quality of underwater images play an irreplaceable and significant role in the subsequent underwater image processing tasks,such as underwater image enhancement and restoration,underwater target detection and recognition,image reconstruction,shape recovery.Based on the previous research,through the characteristics of underwater imaging and human visual system research,this paper focuses on in-depth study on underwater image quality assessment technology.The main research content of this thesis can be summarized as follows:(1)A new no-reference underwater image quality measure based on human vision system(HVS)is proposed,dubbed CCS,which has stronger correlation with human subjective perception.It is taken into account the characteristics of underwater imaging and the HVS is susceptible to the changes of the visual properties such as color,contrast,and edge structure.CCS chooses the color feature with better correlation with human visual perception,the contrast feature based on the human brain visual cortex,and the sharpness feature reflecting the plenty of information to construct an underwater image quality assessment model.In addition,due to the lack of underwater image quality databases,a small subjective quality dataset of underwater images is established to verify the performance of the proposed CCS.Experimental results demonstrate that the proposed CCS metric has a higher correlation with subjective evaluations,which can effectively and accurately evaluate the underwater image quality.(2)A novel reference-free quality assessment metric for underwater image is designed,which is based on the support vector regression(SVR)of underwater image quality evaluation metric,dubbed the CSN.Considering the physical property of the water media,the indeterminacy of the authenticity of the underwater image color,and the importance of high-level feature extraction for underwater image analysis such as image segmentation and recognition,CSN metric extracts a total of 28 features for contrast measure,sharpness measure and naturalness measure,respectively.In addition,in view of the fact that HVS is a nonlinear and complex imaging system.The statistical 28 features fed into the trained support vector regression(SVR)model to perform a regression that maps these features to the image quality.Experimental results demonstrate that the proposed CSN has a better correlation performance with subjective evaluation when compared to the other underwater image quality measures.Moreover,CSN can also provide an effective evaluation of image enhancement and restoration methods.In addition,it is promising with respect to both computational complexity and practical reliability for underwater real-time applications.