Single picture reconstruction HDR method based on multistage dual feedback

Abstract

The invention discloses a single picture reconstruction HDR method based on multistage dual feedback. The method comprises the following steps: S1, inputting an LDR image; S2, reconstructing an inverse quantization image by using an inverse quantization network, and performing quantization operation on the inverse quantization image to obtain a dual quantization image; S3, performing ICRF mapping processing on the inverse quantization image to obtain a linear image, and using the same network learning CRF mapping as a dual feedback structure; S4, dividing the linear image into an overexposure area and other areas, inputting the overexposure area into the truncation information recovery network to obtain an HDR image, performing layered truncation on the HDR image, calculating truncation loss by using the truncation image, and feeding back the truncation loss to the truncation information recovery network; S5, integrally adjusting the HDR image by using a multi-scale receptive field feature extraction module and an attention mechanism. According to the method, a dual network is added for each process to form feedback, the reconstruction direction can be restrained, the reconstruction effect is improved, the model learning ability is improved, and the authenticity of the reconstructed image is enhanced.

Description

Technical field [0001] The present invention belongs to the field of image processing, and in particular, the present invention relates to a single picture reconstruction HDR method based on multi-stage pair feedback. Background technique [0002] The brightness of the light in the natural scene is very broad, and the bright and dark spans are very large, and many visual details are present. At present, it is limited to hardware equipment, causing people to obtain photos of the brightness of the part-wide brightness, which is common LDR images, and the LDR image basically reflects most of the information and details in the real scene, but there is still a lot of loss. In recent years, with the development of communication and display equipment, people have increasing the demand for more real HDR image data. In the movie camera, you can express more real scenes; in terms of medical diagnosis, HDR images can provide more details, let doctors judge the disease more accurate; in comp...

AI Technical Summary

Problems solved by technology

In the process of actually shooting and generating LDR images, there is a lack of dynamic range information. This patent simply adjusts the weight of the existing information to change the visual effect, but does not restore the lost information, resulting in reconstructed HDR. The image information is limited and the authenticity is not high

Images