Generative Adversarial Network Super-Resolution for Design Surface Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current generative models, such as GANs and autoregressive models, struggle to produce high-resolution digital images suitable for printing on design surfaces efficiently, requiring complex tools and skilled personnel, and result in lengthy processing times due to the complexity of replicating natural materials like marble or wood.
Innovation Solution
A generative system utilizing a generative adversarial network (GAN) with a generator and discriminator, combined with a super-resolution and refinement network, trained on high-resolution scans of natural surfaces, capable of generating and refining images to high resolution, allowing for efficient creation of diverse digital images comparable to manually edited or scanned images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If generative adversarial networks and autoregressive models are used to generate digital images, then image generation speed is improved, but image resolution is reduced
Solution Approach 1:
The patent divides the image generation process into two distinct stages: a first generative adversarial network generates initial images at lower resolution for speed, then a super-resolution network processes these images to achieve high resolution. This segmentation allows each network to optimize for its specific resolution requirement, resolving the contradiction between generation speed and image quality.
Solution Approach 2:
The patent introduces an intermediate resolution dimension by using a super-resolution network that takes low-resolution images from the GAN and transforms them into high-resolution images. This additional processing dimension allows the system to achieve both fast generation (from GAN) and high resolution (from super-resolution network).
2Manufacturing precision
If high-resolution scanning of natural materials is performed, then image quality is improved, but processing time is increased
Solution Approach 1:
Instead of physically scanning each natural material surface, the patent creates a training dataset from scanned images and trains a generative model to copy and generate similar images. This allows the system to produce high-quality images without the time-consuming physical scanning process for each new material sample.
Solution Approach 2:
The patent performs preliminary scanning and training during an initial phase, creating a comprehensive training dataset and trained model. Once trained, the system can rapidly generate high-quality images for new materials without repeating the time-consuming scanning and training process, thus reducing processing time for individual image generation.
3Manufacturing precision
If complex tools and skilled personnel are used to create digital images, then image quality is improved, but device complexity is increased
Solution Approach 1:
The patent implements an automated generative model system that performs image generation autonomously without requiring skilled personnel. The trained GAN and super-resolution network automatically create high-quality images from training data, eliminating the need for manual intervention and reducing operational complexity.
Solution Approach 2:
The patent replaces manual image creation processes (requiring skilled personnel and complex tools) with an automated computational system. The generative model uses neural networks and deep learning algorithms to automatically generate images, substituting mechanical/manual operations with electronic/computational processes.
Data Source
AI summary
A generative system for the creation of digital images for printing on design surfaces comprises a training dataset comprising a plurality of sample images for printing on design surfaces, a generative adversarial network comprising a generator and a discriminator, wherein the generator receives noise at input and is trained to generate at output starting from the noise a new artificially generated image adapted to be used for printing on design surfaces, and wherein the discriminator receives at input the new artificially generated image and is trained to compare and distinguish the new image generated by the sample images of the training dataset.