Research On Low Light Image Enhancement Methods

Under low illumination,the camera cannot get enough light source,which seriously affects the viewing effect of the image due to underexposure.The low brightness and contrast of low-light images are a huge obstacle to subsequent computer vision tasks.In low illumination enhancement,many factors such as brightness,contrast,artifact and noise need to be considered simultaneously,which makes the problem challenging.Although people have conducted extensive research on low-light problems,there are still dark areas,artifacts,and loss of details in the enhancement results.Considering the advantages and disadvantages of existing methods,this paper innovated the low-light enhancement method and proposed our own solutions.The work of this paper can be divided into the following aspects:(1)A low-light image enhancement method based on physical model is proposed.In a low-light environment,the reflected light on the surface of the object is weak,and the captured image is under-exposed.After statistics and analysis of low-light images,the inverted image of the under-exposed image is similar to the hazy image.Aiming at this phenomenon,a low-light image enhancement method based on physical model is proposed.First,invert the low-light image to get a hazy image.Then the hazy image uses the pyramidal dense residual block network and the K-means classification method based on the dark channel prior to calculate the transmission map and atmospheric light intensity.Finally,the parameters obtained by the solution are brought into the low-light imaging model to solve the low-light enhanced image.The experimental results were evaluated subjectively,and the information entropy and average gradient were used to evaluate objectively.(2)A low-light image enhancement method based on multi-light estimation is proposed.Then the Laplacian pyramid-based multi-scale image fusion technology is used to adaptively fuse the well-exposed areas in the multiple exposure-corrected images to obtain the final low-light enhanced image.Among them,gamma correction and inversion are performed on the input image,thereby obtaining images with multiple exposures(such as underexposure,overexposure,and partial area overexposure and underexposure).The image obtained by gamma correction is used for the light adjustment of the underexposed areas in the low-light image,and the inversion is used for the light adjustment of the overexposed areas.This paper use a variety of methods to conduct experiments on multiple public data sets,and compare the experimental results from both qualitative and quantitative aspects.Experimental results show that this method can effectively eliminate the influence of low light and improve the quality of image.(3)A progressive recursive image enhancement network is proposed,which includes an attention layer and a recursive unit composed of residual blocks and recurrent layers.The low-light image first enters the attention layer for global feature extraction,then uses the recursive unit for local feature extraction,and finally outputs the enhanced image.In this paper,the attention mechanism is used to assign weights to pixels in the image,so that the network is more sensitive to low-light areas.The recurrent layer is introduced into the recursive unit to transfer depth features across stages.In addition,the recursive operation using a single residual block significantly reduces the number of parameters while also ensuring the performance of the network.Although the network structure is simple,the network can produce better low-light enhancement effects for different lighting conditions.Experiments on multiple datasets reveal the advantages of this method from a qualitative and quantitative perspective.