Improved Retinex-Net vignetting image correction method

The invention belongs to the technical field of digital image processing, and particularly relates to an improved Retinex-Net vignetting image correction method, which comprises the following steps: step 1, image preprocessing: carrying out batch processing on the size and format of an acquired image set; step 2, inputting a pair of vignetting and contrast images into the trained improved Retinex-Net model; according to the method, the Retinex theory and the convolutional neural network are combined, the vignetting image is decomposed into the illumination component image and the reflection component image by using the constraint condition, the illumination component image is enhanced, the receptive field is expanded through the cavity convolutional network, the reflection component image is input into the dense residual network to be denoised, and the image quality is improved. And the two processed components are fused and reconstructed, so that a corrected recovered image is obtained. Compared with other c