Research On Underwater Image Quality Enhancement And Stereo Matching Algorithm

The 21 st century is the century of the ocean.The ocean,rich in resources,provides new space for human survival and development.The prerequisite for the exploitation and utilization of marine resources is the effective perception of the marine environment.Underwater binocular stereo vision technology,which can perceive the underwater environment,has become a hot topic for scholars all over the world due to its rich information,high accuracy,low cost and simple structure.High-precision matching algorithm is the premise of obtaining reliable 3d information in binocular stereo vision technology,and high-precision stereo matching relies on high-quality images.Underwater binocular stereo vision technology faces two problems: First,underwater imaging environment is bad: image color distortion,contrast reduction,detail loss,which can not meet the requirements of binocular stereo matching for high-quality images.Second,similar to the conventional stereo matching algorithm,underwater stereo matching algorithm has a low matching accuracy in the areas of image occlusion,discontinuity and texturefree,and there is a large space for improvement.Aiming at the above two problems,this thesis firstly studies the underwater image quality enhancement,and then studies the underwater stereo matching based on the image quality enhancement.Specific research contents are as follows:1)To solve the problem that the underwater image quality enhancement method based on image restoration cannot effectively estimate the transmission,an image degradation model is considered and an image restoration method based on light channel prior is proposed.Firstly,the transmission map is estimated by the bright channel prior,and the maximum chromatic aberration image is used to correct the transmittance map,so as to improve the stability of the transmission map.Secondly,an atmospheric light estimation method is proposed for underwater images,and the transmission map and the estimated atmospheric light are used to restore the image of water degradation quality,so as to improve the quality of underwater images.Finally,in order to improve the image color deviation correction ability of the method,histogram equalization of the restored image with limited gray scale range is carried out to balance the colors of different channels of the image.2)In order to improve the contrast of underwater images and balance the color of underwater images,the image degradation model is considered,and an image quality enhancement method based on light channel prior is proposed.Firstly,to solve the problem of image color distortion,an adaptive proportional fusion color balance method is proposed to obtain a color balanced image.Secondly,in order to solve the problem of edge information loss,the fractional-order differential method is used to enhance the edge information of the image and obtain the edge enhanced image.Thirdly,in order to prevent the red channel image from being over-enhanced,the proportional fusion method is used to obtain the red channel suppressed image.Finally,by using the local triple-fusion method,the color balanced image,edge enhanced image and red channel suppressed image are fused to improve the image quality of underwater degraded image.3)In order to improve the accuracy of underwater stereo matching,on the basis of local expansion movement algorithm,an improved PatchMatch algorithm based on pixel category is proposed.To build a more reasonable energy function,the Geman-McClure function is used to process outliers of matching cost(energy function data item),and the smooth term of energy function is constructed by considering the correlation information in the color.A new cross window model is proposed to extract more local windows for calculating matching cost and sharing labels.In order to make the initial disparity tag value more reasonable and effective,the constraint information is used to initialize the tag during initialization.In the process of iteration,candidate tag sets are generated through various mechanisms to update disparity tags by using various pixel category information,and different optimization strategies are adopted for different type pixels in disparity optimization.4)Aiming at the problem that the improved PatchMatch stereo matching algorithm based on pixel category is not perfect in textureless/texturefree regions,on the basis of cross-scale cost aggregation algorithm,a local matching algorithm based on multi-scale disparity fusion and multi-type supporting windows is proposed.In order to improve the accuracy of initial disparity maps in textureless/texturefree regions,a method of adaptive combination of different matching costs by guided filtering is proposed.In view of the current local algorithm cannot effectively use global information,using more pixels classification information construct multi-type support region,thus put forward a multi-scale disparity image fusion and disparity optimization strategy which can consider the local and global information comprehensive.In the comparison experiment of image quality enhancement algorithm,the performance of the algorithm is evaluated by using public images and images taken in the laboratory and real seawater environment.In the comparison experiment of stereo matching algorithm,due to the lack of standard disparity measurement map in underwater environment,the matching algorithm is firstly compared and evaluated with the general matching test set image,and then the performance of the matching algorithm is evaluated with the binocular images taken in the laboratory and real seawater environment.Experimental results show that the proposed algorithms in this dissertation can effectively enhance the image quality of underwater images and improve the stereo matching accuracy of underwater binocular stereo vision.