Style migration method for target edge sharpening

Abstract

The invention relates to the field of image processing, in particular to a style migration method for target edge sharpening, which comprises the following steps of: establishing a deep neural mattingnetwork; performing feature map extraction, style image reconstruction and content image reconstruction on an original style image and an original content image; optimizing a loss function generatedby the style image and the content image; calculating an Euclidean distance between the original style image and an extracted content image feature map; calculating a mean square error between the original style image and an extracted style image feature map; determining a total loss function of the style image, the original style image and the content image; and iteratively updating the error calculated by the total loss function, and outputting a style image. According to the invention, structural constraint and texture synthesis of style images and parameter sharing among images with similar styles are realized.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a style transfer method for object edge definition. Background technique [0002] Image artistic style rendering is an important research direction in the field of computer vision. Image art stylization has a large number of applications in the film industry, animation production, game rendering and other fields. Artistic image style means that the semantic content of an image is represented by the style of another image. Traditional style transfer methods are mostly manual modeling, requiring professional experience and complex mathematical formulas, and have poor semantic content preservation and spatial constraints. Due to their powerful image representation capabilities, deep neural networks have quickly become a popular tool for image stylization, driving the development of many neural style transfer methods in recent years. Therefore, an efficient and accurate image styli...

AI Technical Summary

Problems solved by technology

[0003] In view of the destruction of the image layout and the blurring of the boundaries between the foreground, background and other objects, the style color of the generated image will be mixed with colors that are not in the style image, the semantic content and spatial layout of the image are limited, and the generated image The problem of distortion will be generated. The present invention proposes a style transfer method for object edge clarity, which realizes structural constraints of style images, texture synthesis, and parameter sharing between images with similar styles.

Images