Weak light image enhancement method and device based on conditional adversarial network

Abstract

The invention relates to a weak light image enhancement method and device based on a conditional adversarial network, computer equipment and a storage medium. The method comprises the following steps: using a plurality of convolutional neural networks of a residual structure connected in series as a generator and a WGAN-GP as a discriminator, performing down-sampling on a weak light image through a stepping convolutional network, processing a down-sampled feature map through a residual network, performing up-sampling on a residual correction feature map through a transposition network, fusing the weak light image sample original image and the up-sampling feature image through a jump connection structure and then activating and outputting, and obtaining a generator generation image; judging the generator generation graph through a discriminator network to obtain a discriminator output matrix, and generating an effective weak light enhancement model under the constraint action of a preset loss function. According to the model, detail information of an original image can be reserved, the brightness of the image is improved, meanwhile, gain noise interference cannot be generated, and the generated image is more natural and high in quality.

Description

technical field [0001] The present application relates to the technical field of image enhancement, in particular to a conditional adversarial network-based low-light image enhancement method, device, computer equipment and storage medium. Background technique [0002] In a low-light environment, the image quality of the camera is poor, manifested in high noise, color distortion, low brightness and contrast, and image enhancement processing is usually required. [0003] Traditional low-light image enhancement methods are mainly divided into two categories. Type 1 methods build on the histogram equalization technique and add additional priors and constraints, the purpose of which is to expand the range to improve image contrast. For example, the difference grayscale histogram method can improve the contrast of the image to a certain extent by enlarging the grayscale difference between adjacent pixels, but it does not perform well in color processing. The second type of algo...

AI Technical Summary

Problems solved by technology

LLNet uses deep autoencoders for low-light image denoising, but its network results are too simple and the results are not good

Other CNN-based methods, such as LLCNN, cannot enhance image contrast and denoise simultaneously

Retinex-Net's deep network combines image decomposition and lighting mapping, and utilizes denoising tools to deal with reflection components, but the enhanced image color is severely distorted

Therefore, there is the problem of poor effect in prior art

Images