Low Illumination Image Enhancement Algorithms Based On Wavelet Transform And Retinex Theory

At present,images are widely used in industry and life,bringing convenience to industrial production and life records.However,in some poor lighting conditions like night,the quality of images obtained by sensors is often poor,which will bring inconvenience in industrial detection,monitoring security,and mobile device recording.Therefore,how to solve the problem of low illumination image under poor lighting conditions has become a hot issue in many industries.This thesis proposed two algorithms to improve the image quality and restore the original information.One of these two algorithms is a de-noising algorithm based on wavelet transform,and the other is a brightening algorithm based on Retinex theory.The specific content and innovation of the algorithm in this thesis are as follows:(1)In order to solve the noise problem of low illumination images,this thesis proposes an image de-noising algorithm based on wavelet transform.After determining wavelet base and the layers of wavelet pyramid,two improvements are proposed,one is improved Visu Shrink threshold and the other is improved soft threshold function.The experimental results show that the enhanced image has good structure retention in the process of multi-layer wavelet decomposition and reconstruction de-noising when used the improved Visu Shrink threshold;The improved soft threshold function with excessive smoothing function is smoother in the low wavelet coefficient part and it can eliminate the impact of breakpoints;The two improved wavelet de-noising improvements applied at the same time can effectively remove noise and preserve the texture details.And compared with other filtering algorithms,the algorithm this thesis proposed has better performance in subjective and objective evaluation.(2)A brightening algorithm with texture enhancement effect based on Retinex theory is proposed to address the issues of low brightness,low contrast,and blurry detail information in color low illumination images.The algorithm chooses to process the image in the HSV color space to protect the color information of the image.The algorithm in this thesis has been improved on the illumination component L and observation component S of Retinex theory.To solve the problems of low image brightness and contrast,this thesis combines the classical Gaussian convolutional illumination estimation of Retinex theory with its inverse image through gamma transform to obtain an improved illumination component L.To solve the problem of blurring image detail information,this thesis uses the difference map between the blurred image and the original image as an auxiliary image for texture enhancement,and then fuses the auxiliary image as a weight with the original observation component to form the final improved observation component S.The experimental results show that the improved Retinex algorithm can effectively restore the lighting information of images,effectively solving the problems of image overexposure and whitening that traditional Retinex theoretical algorithms may encounter.The algorithm also has good results in local contrast and edge preservation.