Image texture synthesis method based on convolutional neural network feature map matching

Abstract

The invention relates to an image texture synthesis method based on convolutional neural network feature map matching, and the method comprises the steps: preparing a data set, and carrying out the preprocessing of the data set; establishing a model based on the VGG-19; testing by using the pre-training weight; processing the feature map by using a feature space nearest neighbor exchange matchingalgorithm, namely a swap algorithm; setting a loss function, and using the loss function to measure the similarity between the generated image and the original texture image; and updating the parameters through a back propagation algorithm to generate a final texture image.

Description

technical field [0001] The aspects involved in the present invention include computer vision, computer image processing, deep learning and other computer fields, and the present invention is more focused on the application of deep learning to image texture synthesis. Background technique [0002] Texture synthesis is a proposed method to solve problems such as seam aliasing in mapping. Texture synthesis algorithms can be divided into two categories, the first is Process Texture Synthesis (PTS), and the other is Texture Synthesis From Samples (TSFS). [0003] The procedural texture synthesis method is a simulation method, that is, texture synthesis is performed by imitating realistic textures, such as our common wood textures, hair and other textures. However, the procedural texture synthesis algorithm has a serious problem: different parameters need to be adjusted for different textures in nature, that is, parameters need to be adjusted every time a new texture is generated...

AI Technical Summary

Problems solved by technology

However, the procedural texture synthesis algorithm has a serious problem: different parameters need to be adjusted for different textures in nature, that is, parameters need to be adjusted every time a new texture is generated, and the parameter adjustment is uncertain. If If the adjusted parameters are not appropriate, the corresponding texture may not be generated

However, the texture image generated by their method is basically the same as the original image, which deviates from the original purpose of texture synthesis.

Images