Single-frame high-dynamic imaging method based on convolutional neural network

Abstract

The invention discloses a single-frame high-dynamic imaging method based on a convolutional neural network. The method comprises the following steps: constructing an enhancement module; constructing an initial convolutional neural network with an enhancement module according to the enhancement module; training the initial convolutional neural network to obtain a trained convolutional neural network; and processing the original single-frame low-dynamic image through the trained convolutional neural network to generate a high-dynamic image. According to the method, the high-dynamic image can bedirectly generated from the single-frame low-dynamic image, the problem that multiple frames of exposure images cannot be obtained at the same time by a moving video in multi-exposure fusion is solved, parameter adjustment is not needed, and the method is convenient to use.

Description

technical field [0001] The invention belongs to the field of digital image processing, and in particular relates to a single-frame high dynamic imaging method based on a convolutional neural network. Background technique [0002] Scenes in nature have a large dynamic range, and there are often bright and dark areas at the same time. The dynamic range of an ordinary digital camera is only two to three orders of magnitude, and the brightness range of the obtained digital image will be compressed. Details and textures in dark and dark areas are lost, colors are distorted, and the layered brightness effects in the original scene cannot be displayed, making it impossible to accurately capture all the information in the real situation. Professional high-dynamic cameras are expensive, and their effects are limited by hardware conditions. Therefore, high-dynamic imaging methods are required to process images. [0003] High-Dynamic Range imaging (HDRI for short) is a photography tec...

AI Technical Summary

Problems solved by technology

One of the existing methods is to modify the image by cascading two encoder-decoders. The first encoder-decoder is used to generate high-dynamic range images (High-Dynamic Range, referred to as HDR), and the second encoder-decoder The decoder is used to map the HDR image, and correct the overexposure and underexposure areas in the LDR image through reciprocating HDR conversion, but the model will lose image information when downsampling, and artifacts will be generated when upsampling, and the model is complex. heavy calculation

Another existing method is to sample two different sizes of convolution kernels and down-sampling to extract the high-frequency, mid-frequency, and low-frequency details in the LDR image respectively and fuse these three detail feature maps to generate an HDR image. , but the calculation amount and running time of the three branches of the method are quite different, and the image details need to be improved

Images