A single image enhancement method based on full convolution neural network

Abstract

The invention discloses a single image enhancement method based on a full convolution neural network. Firstly by considering that the lost information of a single exposure image can be described by amulti-exposure sequence, a low-exposure image and a high-exposure image different from the input image exposure are generated by using the constructed predictive exposure network, so as to construct amulti-exposure sequence. Secondly, in order to avoid the problem of low robustness caused by manual feature extraction, the exposure fusion network is used to complete feature extraction, feature fusion and image reconstruction. Finally, by considering that the deconvolution layer in the predictive exposure network will cause the chessboard artifacts, the structure anisotropy loss related to human perception is used to train the predictive exposure network to alleviate the artifacts. The invention has the advantages that the whole contrast of the image can be improved, and certain informationof the underexposed and over-exposed regions of the image can be recovered.

Description

technical field [0001] The invention relates to a single image enhancement technology, in particular to a single image enhancement method based on a fully convolutional neural network. Background technique [0002] Due to the limitation of the dynamic range of the camera sensor, the details of the single-exposure image are lost. In order to improve the image quality, multi-image fusion technology is often used. However, the multi-image shooting process is to expose the same scene multiple times, and there is a certain time interval. For dynamic scenes, ghost images will be introduced. Therefore, multi-image fusion technology needs to additionally consider the detection and removal of ghost images. The single-image enhancement technology only needs to adjust the tone curve of the single-exposure image and will not introduce ghosting, so it is more practical, but the single-exposure image cannot represent the content of the entire scene, and the quality of the enhanced image w...

AI Technical Summary

Problems solved by technology

This method can improve the overall brightness of the image and restore the details of the saturated area, but considering the noise in the low and dark areas of the image, this method cannot restore the details of the low and dark areas of the image.

[0006] Although relevant research has achieved good single image enhancement effects, there are still some deficiencies in recovering the information of underexposed and overexposed areas of the image.

Images