An Image Rain Removal Algorithm Based On The Depth Of Field And The Sparse Coding

Rainfall weather often results in video quality degradation,which may cause image distortion.Accumulation of rain streaks may generate atmospheric veiling effects,which is similar to mist or fog,and nearby rain streaks exhibit strong specular highlights and occlude background scenes.heavy rain may cause severe visibility degradation and affect the imaging quality of outdoor visual system seriously,which is not conducive to image analysis.In order to improve the quality of video images in intelligent traffic and security precaution,video image enhancement has become a hotspot in the field of machine vision.Therefore,image rain removal is selected as the research object.In recent years,with the rapid development in video and image rain removal,there have been many excellent algorithms.Based on the investigation of a large number of domestic and foreign literature,we describe current research of single image rain removal in detail,and introduce the commonly used algorithms.This paper focuses on the algorithm of rain removal based on image decomposition.Aim at the problem of the background detail loss and the rain residual,which leads to the research focus of this paper:an image rain removal algorithm based on the depth of field and sparse coding.There exist problems of rain residual and loss of contour edge of low-frequency part,the image is decomposed by the combination of bilateral filtering and short-time Fourier transform,so that the contour of the low frequency part of the image is better preserved.For the background mismatch of the same gradient with rain in the high frequency component,the introduction of depth of field can conduct secondary classification according to the texture and gradient direction of the high frequency,which can effectively improve the accuracy of classification and retain the background with the same gradient of rain.The algorithm mainly includes four parts:image decomposition,dictionary learning,atomic clustering based on Principal Component Analysis and Support Vector Machine,the correction of depth of field.Firstly,the low-frequency components are decomposed by bilateral filtering and Fourier transform.This method preserves the high contour and edge.Secondly,the input image is classified according to its texture,and the high frequency components are divided into blocks based on each category,then a dictionary of each image category is obtained for dictionary learning.Thirdly,the dictionary is classified by Principal Component Analysis and Support Vector Machine.According to the gradient information,it is divided into two categories:rainy dictionaries and non-rain dictionaries.Orthogonal Matching Pursuit is used to obtain non-rain components of high frequency components.Finally,the significant feature of the rain image is extracted by using the depth of field to further remove the residua]rain in the low frequency component and retain the background with the same gradient of rain.Subjective visual effects and objective indicators are used to evaluate the proposed algorithm.The experimental results show that the subjective effect is improved obviously and the objective indexes are also improved.The proposed image rain removal algorithm based on the depth of field and the sparse coding not only can remove rain,but also preserve the detailed information of the image.