A human face image stylization method based on spatial constraint

Abstract

The invention discloses a human face image stylization method based on spatial constraint, and the method comprises the following steps: firstly, carrying out the normalization of an input style imageand an original image, extracting a multi-layer depth feature map of a human face through a feature encoder, and extracting human face spatial feature points through a human face modeling network; Then, at each layer of the depth feature map, executing a feature block matching method by utilizing spatial constraint and Platts transform of face spatial feature points to obtain mapping from the style image to the original image; And carrying out feature fusion on the mapped style image and the original image to obtain a new depth feature map, and finally obtaining a stylized original image through a pre-trained feature decoder. The invention belongs to the field of computer vision, fully utilizes the space corresponding relation of a human face, designs a rapid feature block matching method, and can effectively improve the speed and accuracy of human face image stylization.

Description

technical field [0001] The invention belongs to the field of computer vision and image processing, in particular to a method for stylizing human face images based on space constraints. Background technique [0002] Face image style transfer is an image processing method. Preprocessing, analysis and encoding, feature fusion and decoding of the input image can apply the given style features to the original face image to obtain a stylized face image. The previous style transfer methods need to consume a lot of computing resources to solve the numerical optimization problem between the style image and the reference image. In recent years, some fast style transfer methods have appeared. These methods are mostly based on a large number of images of the same type for long-term training. , to extract a specific style model and then render it on the reference image. This type of method can only perform indiscriminate transfer of the overall color and texture features of the image, an...

AI Technical Summary

Problems solved by technology

The previous style transfer methods need to consume a lot of computing resources to solve the numerical optimization problem between the style image and the reference image. In recent years, some fast style transfer methods have emerged. These methods are mostly based on a large number of images of the same type for long-term training. , extract a specific style model and render it on the reference image. This type of method can only transfer the overall color and texture features of the image indiscriminately, and does not have the ability to transfer styles at the semantic level.

So far, no fast, accurate, semantic-level arbitrary style transfer method has emerged

Images