Research On Novel Two-Phase Single Image Removal Algorithms

Since the 1990 s,the processing power of computers has been increasing day by day,and the popularity of tablet,mobile phone cameras,digital cameras,scanners,and other imagecapturing equipment has made computer vision systems widely used in all fields of the society,it ranges from satellite,aviation,and navigation to transportation,medical,and security,digital video,images and other visual data play an important role in many fields.However,in actual life,the image data acquired by various channels will inevitably be affected by bad weather during the shooting and collection process,which will lead to the situation that the acquired image data will reduce the quality to a certain extent.The presence of rain,snow,fog,etc.causes occlusion or reflection of the image,which makes the target image details lost or visibility and sharpness greatly reduced,which seriously affects the extraction of image features,thereby affecting the effectiveness of related visual algorithms,then interfere with the normal use of application systems such as aerial detection,outdoor monitoring,and intelligent navigation.Therefore,the sharpness restoration and enhancement algorithm for degraded images obtained in bad weather has become one of the important topics that domestic and foreign scholars have invested in research.Common severe weather includes fog,snow,rain,etc.for images acquired under different types of weather,there have been a lot of researches on image defog,snow,and rain.The physical characteristics such as fog and snow are relatively simple,and there are many relatively mature studies,but the rain weather is more frequent,and the shape,scale and density of the rain streaks are diverse,so it is more challenging and more necessary for research.This paper is devoted to researching effective image quality enhancement algorithms to improve the quality of degraded single images acquired in rainy weather,mainly based on the location of rain pixels,and put forward an effective repair model to repair the background area damaged by the rain and remove the effect of the rain streaks occlusion,finally results in a clean image.Specifically,in this paper,the overlapping sampling image patch is used as the basic unit,and the prior features of rain pixels and non-rain pixels are used as clues.A twostage strategy of first detection and later repair is adopted to efficiently and quickly solve the problem of single image de-raining.The research algorithms and innovation contributions of this paper are summarized as follows:a)The first step in removing rain is to mark the location of rain pixels as accurately as possible.Therefore,this paper first uses the rain brightness prior and the rain direction prior to propose a more accurate rain detection strategy to establish almost mark all rain pixels while filtering out the non-rain background pixels that are mislabeled,and mark rain pixel locations more accurately.b)This repair algorithm is based on the expansion and utilization of the similarity of image patches.In the process of similar patch matching,the two-step patch matching algorithm of coarse and fine granularity is combined to achieve high-precision matching to a certain extent.c)After locating the rain streaks in the image,the next step is to use rain position information to repair pixels damaged by rain.This paper proposes two repair models: 1.Use the priori of the similarity matrix composed of similar patches with low rank attributes to build a repair model based on the low rank matrix completion algorithm.2.Use the prior information of the position of similar patches to build a repair model based on minimum value filtering.Comparative experiments show that the single image de-raining algorithm proposed in this paper can eliminate obvious rain streaks,and to a large extent ensure that the image details are not destroyed,whether it is subjective observation or objective evaluation criteria,there are relatively good results.And more importantly,the algorithm of this paper also shows certain advantages in the processing of heavy rain images.