Image processing method and device, electronic equipment and computer readable storage medium

Abstract

The image processing method comprises the steps of determining a color component of each point of a preset channel based on a color component of the original image in the preset channel in a preset color space; determining a color component of each point of remaining channels based on color components of the original image in the remaining channels except the preset channel in the preset color space; determining optimization parameters of a native image in the preset channel through a first neural network model; determining optimization data of the preset channel based on the optimization parameters and the color component of each point of the preset channel; determining optimization parameters of the native image in all channels of the preset color space through a second neural network model; and determining a final full-color image based on all channel optimization parameters of the preset color space and color components of each point on all channels in the preset color space. The advantages of an interpolation algorithm and a neural network can be combined to obtain a good image.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular to an image processing method, device, electronic equipment, and computer-readable storage medium. Background technique [0002] The existing demosaicing algorithms mainly include two types, one of which is to use interpolation, through a large number of mathematical analysis and the exploration of correlation principles, trying to restore the lost channel information from the perspective of logical analysis and mathematical models, but considering Due to the diversity and complexity of real scenes, it is difficult to use a single algorithm model to solve all interpolation problems, and the introduction of a large number of mathematical formulas will lead to an increase in the complexity of algorithm design. Although the interpolation effect can be optimized, However, the computing performance that needs to be consumed is also increasing, which is not conducive...

AI Technical Summary

Problems solved by technology

[0002] The existing demosaicing algorithms mainly include two types, one of which is to use interpolation, through a large number of mathematical analysis and the exploration of correlation principles, trying to restore the lost channel information from the perspective of logical analysis and mathematical models, but considering Due to the diversity and complexity of real scenes, it is difficult to use a single algorithm model to solve all interpolation problems, and the introduction of a large number of mathematical formulas will lead to an increase in the complexity of algorithm design. Although the interpolation effect can be optimized, However, the computing performance that needs to be consumed is also increasing, which is not conducive to the hardware implementation of the algorithm; the other is to realize it through machine learning. Compared with the traditional algorithm, the calculation effect of this method is also improved to varying degrees, but it is complicated. The neural network model also consumes a lot of computing resources, especially the end-to-end model training structure, which often improves the performance by continuously increasing the number of layers of the neural network, which affects the practicability of the algorithm.

In the prior art, both methods are used alone, and it is difficult to give full play to the advantages of both

Images