Research On Low-light Image Enhancement Algorithm Based On Retinex

As an important carrier for people to obtain information,images have played an important role in various fields.In particular,the application of images and computer vision technology to the field of industrial production has greatly improved the production efficiency.Obtaining high-quality images is not only conducive to the observation of the human eyes,but also facilitates the computer's further application of the images.However,in a variety of complex lighting environments,the quality of the images obtained is not high.For example,in low-light environment at night or backlight condition,the brightness and contrast of an image obtained are low,and the color is distorted,and it is difficult to directly obtain information in the image.Therefore,the enhancement of such images is of great significance.In this thesis,the properties of two types of low-light images under low-light and back-light conditions are introduced respectively.Common low-light image enhancement algorithms are analyzed.The variational-retinex and multi-scale retinex are improved.The main research contents as follows:First,a relative total variational Retinex algorithm is proposed.In the HSV color space,the luminance component V is processed using the relative total variation Retinex algorithm to obtain the irradiation component and the reflection component.The illumination component be improved brightness and contrast by gamma correction,The reflection component is de-noised by filtering,and then,enhancement details.In order to make image have more natural of the color,the saturation is adaptively stretched using the difference between the before and after the enhancement of brightness components,and finally the image is subjected to white balance processing to obtain an enhanced image.Secondly,in order to avoid over-enhancement of regions with higher brightness in the image,gamma correction has been improved.Image segmentation is used to divide the image into high luminance regions and low-light regions,and adjustment factors are introduced to the gamma correction parameters to weaken the high luminance regions.The enhancement effect thus avoids over-enhancement,and an adaptive anti-sharpening mask detail enhancement model is constructed with using the local variance of the image.The experimental results show that the improved variational Retinex algorithm can effectively improve the brightness and contrast of the low-lightimage,enhance the detail information,and overcome the halo artifacts that the traditional Retinex algorithm,has a better enhancement effect.The image fusion technique is used to improve the traditional multi-scale Retinex algorithm,and proposed a color recovery algorithm.first,using the single-scale Retinex algorithm with different parameters,enhancement the low-light image,and then enhancement the contrast.The enhancement image is taken as the fusion input image.using the chroma image,feature image and exposure image of the fusion input image generate the fusion weight maps,decomposes the fusion weight maps and input maps into image pyramids with multiple resolutions,after fusion,reconstruction,and finally using the proposed color recovery algorithm for color adjustment,obtain the final enhanced image.Simulation experiments verify the feasibility and enhancement of the algorithm.