| Vision is the most important way that humans perceive the world,and it is also the main approach for people to perceive and recognize the outside world.With the rapid development of information technology,there are more and more images that people are exposed to,and images have become an indispensable method of information transmission.However,in real-life images,due to bad weather conditions,different ambient light intensity changes,poor light source conditions,and insufficient self-exposure of the photosensitive device,the collected images are often not satisfactory in terms of contrast and hue.In a wide variety of images,low-light images are a prominent type of image.The main features are dark brightness,low contrast,small dynamic range,and large noise.These prominent problems seriously affect people's perception of image.Therefore,contrast enhancement and de-noising of low-light images have a high research significance.Based on the deep understanding and analysis of low-light images,this paper first introduces the image contrast enhancement algorithm and image de-noising algorithm,and summarizes the most cutting-edge research results in this field in recent years.In view of the deficiencies of the existing image enhancement and image de-noising algorithms,we have proposed related improved algorithms and achieved good results.This article starts from the following two aspects:1.An image enhancement algorithm based on perceptual optimization has been proposed.This work is based on the Generalized Equalization Model(GEM),which for the first time integrates contrast enhancement and color constancy into a unified framework.First,we fully consider the human visual system(HVS)characteristics by using the JND model and Weber's law to extract high contrast and human-sensitive pixels in the image.Then using the extracted pixels to reconstruct a histogram which is called perceptual histogram.We apply this method to histogram-based GEM to effectively prevent over-enhancement and tone distortion.Then we use the JND Transform,which based on the HVS response model,to extract the human visual perceptual response to different pixels of the image.We call this the JND map.According to Weber's law,we enhance the extracted JND map to achieve the effect of image detail enhancement.Finally,we obtain the final enhanced grayscale image by the inverse JND transform algorithm,and then use the color restoration algorithm to obtain the final result.Experimental results show that the image obtained by this method has a good enhancement effect,such as color reproduction and detail enhancement,and is more in line with human visual characteristics.2.A low-light image enhancement and full-channel de-noising algorithm based on human visual perception has been proposed.This work was based on the previous work.First,it reconstructed a perceptual histogram by using the JND model and Weber's law,in this work,the method is commonly used in various histogram-based image enhancement algorithms,which can effectively prevent over-enhancement and tone distortion.Then we extract the JND Map by the JND Transform algorithm,and according to Weber's Law,we enhance the extracted JND map.At the same time,it can be known from the visual masking effect that the human eye is more sensitive to the noise in the smooth region,and on the contrary,the texture region has a strong masking effect,resulting in the noise being not easily perceived by the human eye.Therefore,we use the human visual masking effect to do luma channel de-noising.Finally,we obtain the final enhanced and de-noised grayscale image by using the inverse JND transform.And we use the guided filter to do the chroma channel de-noising under the guide of de-noised luma channel and obtain the final result.The experimental results show that by introducing a suitable human visual model,the image is more in line with human visual characteristics,and the expected enhancement and de-noising effects are achieved. |
PDF Full Text Request