Research On Underwater Image Sharpening Based On Multi-scale Fusion

As an important way for human to detect the underwater world,underwater optical images have been widely used in various fields,such as inspection of the underwater buildings at wharfs.The underwater robot is used to carry high-definition cameras to inspect the underwater wharf.The structural image information of the wharf underwater engineering can be obtained by means of pre-designed trajectory or real-time remote control.Simple and convenient,and less affected by wind and waves.However,due to the influence of the complex underwater environment on the absorption and scattering of light,the problems of low resolution,blurred image and color distortion are common in the underwater images taken,which seriously limits the practical application and development of this technology.Therefore,it is very necessary to obtain more abundant and real underwater image information and solve the problem of underwater image degradation.Based on the quality reduction of underwater images,this paper proposes a multi-scale fusion underwater image processing method which accords with human visual effect and high contrast.The main research contents of this paper are as follows:Firstly,the attenuation of light absorption and scattering in water is analyzed,and the degradation model of underwater imaging is established.Two kinds of image quality evaluation methods are introduced in detail,and the classical underwater image sharpening algorithm is reproduced through simulation experiment,and its advantages and disadvantages are analyzed.Secondly,the attenuation of the selective absorption of light in the process of underwater propagation leads to the color deviation of the underwater target image.An improved histogram matching algorithm is used to correct the color.By fitting the R and B channels to G channels and stacking the histogram,the underwater pre-processing image after color correction is obtained.The scattering attenuation of light in the process of underwater propagation leads to image contrast reduction and detail blurring.The V-channel of the color correction image is processed by the limited contrast adaptive histogram equalization enhancement algorithm and Gamma correction algorithm to obtain the preprocessed image with color correction and contrast improvement.Then,three weight features such as significance graph and chrominance graph were selected to calculate the normalized weight graph of the preprocessed image respectively.Combined with the idea of multi-scale pyramid fusion algorithm,Gaussian and Laplacian pyramid decomposition and reconstruction are carried out to complete the enhancement processing based on image fusion algorithm.The effectiveness of the proposed algorithm is proved by comparative experiments.Finally,the color-corrected image and the pre-processed image with improved contrast were used as the input images for fusion processing,and the weight graphs and normalized images of the input images were calculated respectively.Then,the Gaussian and Laplace pyramid algorithms were used to decompose and reconstruct the input images respectively,and the multi-scale image fusion was completed.Due to the loss of image detail and texture due to the generation of forward scattering,the improved edge detection Sobel operator is used to extract the edge texture information of contrast enhanced image,and the enhanced image is added according to the gain and fusion algorithm to obtain the final detail enhanced image.And through the design of a comparative experiment,combined with the subjective vision of human eyes and objective quality evaluation indicators to evaluate the effectiveness and universality of the algorithm in this paper.