Image style migration method based on multi-scale semantic matching

Abstract

The invention discloses an image style migration method based on multi-scale semantic matching, and the method comprises the steps: obtaining a content image training set and a style image training set, and forming a plurality of groups of content-style image pairs; extracting multi-scale depth features of the content image and the style image in the content-style image pair through a deep convolutional network; performing multi-scale semantic matching on the multi-scale depth features of the content image and the style image to obtain reconstructed depth features; synthesizing the reconstructed depth features into a reconstructed image after style migration through a decoder; iteratively updating parameters of the decoder until the decoder converges; and sequentially carrying out multi-scale depth feature extraction, multi-scale semantic matching and decoder synthesis after convergence updating on a content image to be processed and a style image to obtain an image after style migration. According to the method, the structural integrity and coherence of the input content image can be remarkably kept, and meanwhile, the style of the semantic part corresponding to the input style image is accurately migrated.

Description

technical field [0001] The invention belongs to the technical field of image processing, and specifically relates to an image style transfer method based on multi-scale semantic matching, which mainly completes the generation of images from photos to specific artistic styles, and can be used in fields such as digital entertainment. Background technique [0002] Drawing is an important form in the visual arts, captivating many people and producing a great deal of great work for thousands of years. Image style transfer refers to transferring the style of a style image (such as an artist's art painting) to a content image (such as a photo taken by a camera). But having an artist manually paint an image in a particular style can be time-consuming, and automating the process has broad applications in digital entertainment and more. Therefore, the research on image style transfer algorithm is of great significance. [0003] Early research, such as NPR (Non-Photorealistic Renderi...

AI Technical Summary

Problems solved by technology

[0004]However, most methods face a common problem: whether to use parametric methods to measure the correspondence between deep representations

The above existing methods all focus more on the transfer of style, but ignore the structure preservation of the content image during the transfer process, resulting in obvious defects in some types of content images such as portraits

Images