Training Dataset Image Enhancement Using GAN Quality Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning training datasets often contain low-quality images that hinder effective learning, necessitating a need for systems and methods to enhance image quality automatically.
Innovation Solution
A system and method that utilizes image quality assessment methods like DNN features, GANs, and Super-Resolution ML to identify and enhance low-quality images, replacing them in the training set, and if necessary, generating additional images to ensure sufficiency for neural network training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If large datasets are collected using automatic tools with large audiences assisting in labelling efforts, then the quantity of training data increases, but the quality of images becomes insufficient for correct machine learning
Solution Approach 1:
The patent replaces manual image quality assessment and enhancement mechanisms with automated machine learning systems. Specifically, it uses convolutional neural networks (CNNs) to automatically assess image quality and generative adversarial networks (GANs) to automatically enhance low-quality images, substituting the need for manual human review and processing while maintaining or improving quality standards.
Solution Approach 2:
The patent introduces an intermediary automated quality assessment system that acts as a mediator between the collected training data and the machine learning training process. This intermediary system evaluates image quality using CNN-based metrics and filters or enhances images before they are used for training, ensuring that only adequate images proceed to the training phase.
2Manufacturing precision
If image quality enhancement methods are applied to low-quality images, then image quality improves, but processing time and computational resources increase
Solution Approach 1:
The patent applies partial enhancement by selectively processing only those images that fall below quality thresholds, rather than enhancing all images in the dataset. The system assesses each image's quality metric and applies enhancement only when necessary, avoiding unnecessary processing time for already adequate images.
Solution Approach 2:
The patent performs preliminary quality assessment of images before they are used for training. By identifying low-quality images in advance through automated CNN-based evaluation, the system can pre-process and enhance these images before the main training phase, preventing quality issues from arising during critical training operations.
3Manufacturing precision
If manual image quality assessment and enhancement is performed, then image quality can be improved, but the complexity of the system increases and automation is reduced
Solution Approach 1:
The patent replaces manual image quality assessment and enhancement operations with automated machine learning systems. CNNs automatically evaluate image quality based on learned features, and GANs automatically generate enhanced versions of low-quality images, completely eliminating the need for manual human intervention in the quality control workflow.
Solution Approach 2:
The patent implements a self-service quality control system where the machine learning models automatically assess and enhance their own training data without external human intervention. The CNN-based quality assessor and GAN-based enhancer work autonomously to maintain training data quality, making the system self-sufficient in its quality control needs.
Data Source
AI summary
Systems and methods for automatically enhancing the quality of images in the training set of a neural network NN. A method includes gaining access to a training set including a plurality of images. Using at least one image quality assessment method, at least one image is identified from a plurality of images in the training set, which matches a low-quality criterion as at least one low-quality image. At least one image enhancement method is used for enhancing the at least one low-quality image to obtain at least one enhanced image. The at least one low-quality image is replaced with the corresponding at least one enhanced image in the training set.