GAN Image Correction via Multi-Loss Neural Networks

Solution Overview

Problem

Existing image processing methods using deep learning and convolutional neural networks require manual correction of images using editing software, which is time-consuming and costly, especially for large datasets.

Innovation Solution

A method and apparatus for image processing using machine learning on a Generative Adversarial Network (GAN), which generates intermediate images through a fixed parameter algorithm and performs machine learning on multiple loss functions across multiple convolutional neural networks to correct input images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If manual image correction using editing software is used to generate teaching images for supervised learning, then image quality is improved, but time consumption and cost increase significantly

Engineering Contradiction:

Improveimage qualityVSAvoidtime consumption

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The system uses automated algorithms with fixed parameter values to generate intermediate images from input images, eliminating the need for manual photographer correction. The convolutional neural networks automatically learn and optimize image correction, making the system self-sufficient without human intervention for data preparation.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical process of manual image editing with software tools with an automated machine learning system. Convolutional neural networks perform the image correction task that previously required human photographers, substituting mechanical human operation with automated computational processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Manufacturing precision

If manual image correction using editing software is used to generate teaching images for supervised learning, then image quality is improved, but cost increases significantly

Engineering Contradiction:

Improveimage qualityVSAvoidcost

Core Design Contradiction:

Manufacturing precisionVSEase of manufacture

Solution Approach 1:

The system uses automated algorithms with fixed parameter values to generate intermediate images from input images, eliminating the need for manual photographer correction. The convolutional neural networks automatically learn and optimize image correction, making the system self-sufficient without human intervention for data preparation.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical process of manual image editing with software tools with an automated machine learning system. Convolutional neural networks perform the image correction task that previously required human photographers, substituting mechanical human operation with automated computational processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Manufacturing precision

If multiple convolutional neural networks and multiple loss functions are used for image correction, then image correction quality is improved, but device complexity increases

Engineering Contradiction:

Improveimage correction qualityVSAvoidsystem complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system divides the image correction task into multiple specialized convolutional neural networks, each responsible for specific aspects of image correction. Multiple loss functions are segmented to evaluate different quality metrics independently, allowing each component to be optimized separately while contributing to overall image correction quality.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The multiple convolutional neural networks and loss functions work together in a unified GAN framework, where the generator and discriminator cooperate to achieve comprehensive image correction. The system performs multiple functions (color correction, sharpness enhancement, noise reduction) through a integrated architecture that leverages the synergistic interaction between multiple components.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12277676B2Image processing method and apparatus based on machine learning

Publication Date: 2025.04.15 POSTECH ACADEMY INDUSTRY FOUNDATION

AI summary

An image processing method and apparatus based on machine learning are disclosed. The image processing method based on machine learning, according to the present invention, may comprise the steps of: generating a first corrected image by inputting an input image to a first convolution neural network; generating an intermediate image on the basis of the input image; performing machine learning on a first loss function of the first convolution neural network on the basis of the first corrected image and the intermediate image; and performing machine learning on a second loss function of the first convolution neural network on the basis of the first corrected image and a natural image.