Portrait style transfer method based on semantic segmentation and deep convolutional neural network

Abstract

The invention discloses a portrait style transfer method based on semantic segmentation and a deep convolutional neural network. The method comprises the steps of firstly, selecting a portrait image needed to be converted and a target style portrait image; secondly, carrying out semantic segmentation on the two images to obtain a portrait area and a background area; thirdly, segmenting out the specific five sense organs from the portrait area; fourthly, defining a portrait style transfer loss function; fifthly, adopting the deep convolutional neural network VGG-19 as an image advanced style extraction basic model; sixthly, after a content constraint layer and a style constraint layer is set, defining the content constraint layer and the style constraint layer in the VGG-19 model; and finally, establishing a new model structure. The segmented semantic image and the original image are input to the new VGG-19 model; the advanced style features and content features of the image are extracted; the portrait style transfer loss function is used; by adopting a gradient descent method, the loss function is minimized through multiple iterations; and finally a style transfer result image is generated.

Description

technical field [0001] The invention relates to the field of deep learning, in particular to a portrait style transfer method based on semantic segmentation and deep convolutional neural network. Background technique [0002] With the rapid development of technology, in the field of deep learning research, the process of using CNN to fuse the semantic content of a picture with different styles is called neural style transfer (Neural Style Transfer). Gatys et al. The report article "image Style Transfer Using Convolutional NeuralNetworks" confirmed the amazing ability of convolutional neural network (CNN) in image style transfer: by separating and recombining image content and style, CNN can create works with artistic charm . Since then, there has been great interest in neural style transfer in academic research and industrial applications. Transferring artistic styles from artworks to everyday photos has become a very important computer vision task in both academia and indu...

AI Technical Summary

Problems solved by technology

[0003] The problems existing in the existing style transfer methods mainly include: the style transfer of images is very random, which leads to unsatisfactory results in many cases.

Especially for the style transfer of portraits, some mistakes sometimes occur, such as transferring the features of the eyes in the style image to the mouth, or transferring the features of the image background to the portrait, and the transfer effect is very unsatisfactory.

Images