Research On Image Enhancement Algorithm Based On Retinex Theory And Laplacian Pyramid Reconstruction

In the field of image processing,the Retinex algorithm is widely used for image enhancement,and it also performs well in image defogging and image restoration.Retinex algorithm is focus on estimating the illuminance image from the original image and separating reflections to remove the effects of light intensities.However,the enhancement is always ineffective because of the halo phenomenon when the edge intensity of the original image changes greatly,and the local white grey phenomenon after global enhancement.In order to deal with these issue,a Laplacian pyramid reconstruction is proposed.Gauss weighted bilateral filter is used to replace the Gauss kernel function of original algorithm to estimate the illumination image,so that the reflected image is separated.The improved estimation of illuminance image can solve the halo phenomenon effectively.The reflected image in the logarithmic domain is nonlinear stretched and the light and dark regions are stretched using Otsu threshold segmentation algorithm to produce local optimal images.Finally,Laplacian Pyramid reconstructs the original image and the optimal image from bright and dark area into a global optimal image to solve the grey problem.The experimental results show that the proposed algorithm solve the halo phenomenon and grey problem effectively,which are produced in the single scale Retinex algorithm.Compared with the unsharp masking algorithm and histogram equalization algorithm,the proposed algorithm can imporve the information entropy and average gradient in large extent.Besides,the proposed algorith has strong robustness for uneven illumination image,the single backlight image and the low light image.