A method and device for generating high-resolution ceramic decal pattern

Abstract

The invention belongs to the field of information processing technology and the field of artificial intelligence, and discloses a method and device for generating a high-resolution ceramic decal pattern, including: (1) Constructing a semantic sketch of a ceramic decal pattern for approximate modeling by an artificial neural network Generate a model; (2) Acquire a semantic sketch generation model of ceramic decal patterns that can generate semantic sketches of ceramic decal patterns consistent with user intentions through artificial intelligence-based training methods based on the labeled data of ceramic decal patterns; (3) Construct A ceramic decal pattern generation model based on artificial neural network approximate modeling; (4) According to the labeled data of the ceramic decal pattern and the semantic sketch of the ceramic decal pattern, a training method based on artificial intelligence is used to obtain high-resolution ceramic decals. A ceramic decal pattern generation model for decal patterns. The method of the present invention divides the generation process of the high-resolution ceramic decal pattern consistent with the user's intention into two stages. First, the semantic sketch of the ceramic decal pattern consistent with the user's intention is generated according to the image semantics contained in the user's intention, and then according to the Paper Pattern Semantic Sketch generates high-resolution ceramic decals consistent with user intent.

Description

technical field [0001] The invention belongs to the fields of information processing technology and artificial intelligence, and in particular relates to a method and device for generating high-resolution ceramic decal patterns. Background technique [0002] Since the invention of the Generative Adversarial Network in 2014, a large number of scientific research results have appeared in the field of scientific research. Although people can generate images in various formats according to the needs of users, the resolution of the generated images is relatively low, such as 64*64, 128*128, the current maximum resolution usually selected is 256*256, for example, the patent application CN107016406A discloses a four-layer generation network using a three-layer deconvolution network layer plus a fully connected layer to achieve a pixel distribution rate of 256*256 The program for generating images of diseases and insect pests, and the patent application CN107220600A discloses using ...

AI Technical Summary

Problems solved by technology

[0004] Approximate real pictures are only distributed as a fairly compact manifold (Manifold) in the high-dimensional image space. The training process of the generative confrontation network is to enable the artificial neural network to sample data from this popular in the vast high-dimensional image space. Therefore, when the resolution of the generated image is larger, the dimension of the image space is higher. Although the artificial neural network can sample the image data in the image space, it is difficult to sample the image data in this fashion, which makes the high-resolution image generated by the current technology Pixel resolved images are blurry and unrealistic images

Images