Research On Underwater Optical Image Sharpening Technology

The underwater imaging system is an essential guarantee for the autonomous search and detection of underwater objects.More importantly,image sharpening technology is an effective measure to enhance underwater images and highlight object areas.This thesis relies on the national science and technology support program project "Autonomous search and detection technology for deep-sea underwater objects," which focuses on image acquisition,image classification,image enhancement,object detection,and tracking of underwater object scenes,etc.Meanwhile,this thesis’ s primary focus is on clarifying and classifying underwater images.Additionally,this thesis studies the imaging characteristics and degradation characteristics of underwater images to study the special underwater image sharpening method,and research underwater image classification methods according to the imaging characteristics and degradation characteristics of each underwater image.For underwater images that are mainly faced with color distortion,this thesis proposes an underwater image enhancement method based on the integration of dual-color models,which takes fully into account the visual perceptual characteristics of underwater images in the RGB and CIELAB color models,and combines the optical properties of underwater imaging and the theory of grey-scale word assumptions.The proposed method implements color correction and contrast enhancement in the RGB and CIELAB color models,receptively.Experimental results demonstrate that the proposed method can effectively remove the color distortion and enhance the contrast and texture details of underwater images with good robustness.For underwater images that mainly face the problems of blurred vision,this thesis proposes a global search-and-stretch underwater image enhancement method,which combines the attenuation characteristics of underwater images and the global search-and-stretch contrast enhancement theory.The proposed method performs piecewise linear correction for the severely attenuated color channel while using a global search-and-stretch method to improve the contrast between the foreground and background sub-images.Experimental results demonstrate that the proposed method can effectively remove the fog and blur from underwater images and improve the sharpness and contrast of underwater images.For underwater images that mainly face low light illumination problems,this thesis proposes an underwater image enhancement method based on global and local contrast fusion,which fully combines the advantages of global and local contrast enhancement by the histogram with different rules and the theory of multi-scale fusion.The proposed method integrates the complementary advantages of the global contrast-enhanced image,the local contrast-enhanced image,and the texture detail-enhanced image.Experimental results demonstrate that the proposed method can effectively improve the brightness,global contrast,and local contrast of underwater images.For the classification issue of underwater images,this thesis proposes an underwater image classification method based on multi-color model embedding learning according to the applicability of different underwater image enhancement methods,which combines the differences in visual perceptual characteristics such as chromaticity,luminance,saturation,and contrast in RGB,CIELAB,and HSV color models of underwater images.The proposed method uses the visual perceptual characteristics of images in different color models and the deep network structure to improve the classification model’s effectiveness and its robustness.This method can extract representational features well without a complicated feature extraction process.In this study,the author collected 1520 color distorted underwater images,1480 haze blurred underwater images,and 1495 low light level underwater images.Meanwhile,the classification and clarification methods were validated using the collected 4495 underwater images.Experimental results demonstrate that the proposed classification and clarification methods have good classification and enhancement performance for different classes of underwater images.