Research On Image Enhancement Algorithm In Dust Weather

Image is one of the important means for modern human to obtain information,and also an important carrier for people to understand the objective world.Due to the absorption and scattering of light by sand particles suspended in the air during the collection of dust images,the acquired images are seriously degraded.Degenerated dust images often look brown,yellow or even orange,with color decay,color imbalance,low contrast,and loud noise.In order to obtain more image information,it is necessary to conduct color correction and contrast enhancement processing on the dust image to eliminate the color distortion of the dust image and enhance the clarity.In this regard,in recent decades,the research on dust image processing algorithms has been widely concerned,for dust image color correction,the traditional algorithms and assumptions and improved algorithms will have the problem of insufficient or excessive correction.In this paper,two methods of dust image enhancement are proposed to effectively solve the problems of low definition and abnormal color of dust image.In this paper,the degradation and distortion of the above dust images are studied.The main work is as follows:(1)Design a fast color balance and multi-path fusion algorithm for dust image enhancement.Firstly,color channel compensation and piecewise affine transformation were used to balance the color of dust image to remove color bias.Then,for the problems of underexposure and enhanced contrast,a multi-path fusion method is used to enhance the visibility,in which saturation and Laplacian contrast are measured as weighted graphs and constructed as an image pyramid.In order to reduce the block effect and artifact of the reconstructed image,a multi-scale strategy is introduced for each path.(2)Design a color transfer algorithm for dust image enhancement.The first step is to obtain the eigenvalues of the reference image and the target image by SIFT algorithm,and then the extracted eigenvalues of the reference image and the target image are processed by K-means clustering.Then,similarity detection is conducted for the processed data of the two groups according to the cosine distance between the reference image library and the target image.The most familiar image with the target image in structure is retrieved from the reference image library,and the most similar image and the target image are used for color transfer processing.Finally,the results of color transfer were equalized by compression histogram.(3)The two sand image enhancement methods proposed in this paper are compared with several other sand image processing algorithms with high reference rate,and the evaluation is made from subjective and objective methods.Finally,through the subjective observation and the scientific evaluation of objective data,the superiority of the sand image enhancement algorithm proposed in this paper is proved,and the algorithm can efficiently deal with the color distortion and blur of the sand image.