An Image Harmonization Synthesis Method Based on Color Constancy

Abstract

The invention discloses an image harmony synthesis method based on color constancy, which includes: combining a coarse adjustment network model based on deep learning and a fine adjustment network model introducing an attention mechanism into an overall confrontation generation network, and the training data is a pair of Public image datasets of real images and synthetic images; take the synthetic image to be adjusted as the original input image, obtain grayscale information and input it into the coarse adjustment network model, output the influence weight of the chromaticity of the image light source and estimate the color information of the light source accordingly, Adjust the foreground area based on color balance to get a rough and harmonious synthesis result; use the fine adjustment network model to convert the coarsely adjusted synthetic image and the corresponding mask map to obtain a color harmonious synthesis result by converting similar domain features; through Gaussian The loose method combines the color low-frequency features of the fine-tuned synthetic image and the gradient information of the original input image to jointly optimize and generate a harmonious synthetic image.

Description

technical field [0001] The invention relates to the technical field of image editing, in particular to an image harmony synthesis method based on color constancy. Background technique [0002] Image processing technology has gradually been widely used in various fields, and the most common form of expression is in the application of culture and art. In order to further improve the quality of image works, images with poor visual effects are often "recreated" in the later stage. In daily life, users rely on related software such as Photoshop, Meitu Xiuxiu, etc. to create images, and when acquiring images in scenes such as movie advertisements, they will consider using "green screen" and other technical means to facilitate post-processing. When the target object of interest is fused with the known new background image, in order to reduce the burden of manual processing, people have proposed related image matting and synthesis techniques to meet the needs of various image editin...

AI Technical Summary

Problems solved by technology

[0008] In order to reduce the sense of incongruity when the target object is fused with the new background image, especially for the problem of inharmonious foreground and background colors, it is possible to automatically achieve a batch of harmonious and natural synthesis results. The invention provides an image based on color constancy correction Harmonized Synthetic Method

Images