Multi-Focal Image Pairing for Super-Resolution Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing super-resolution technologies struggle to produce high-resolution images with the same clarity and sharpness as low-resolution images, especially when trained on synthetic datasets, and high-resolution imaging requires significant storage and bandwidth, making it impractical for many applications.
Innovation Solution
An image data collection system captures images at different focal lengths to create high-resolution and low-resolution pairs, which are aligned and used to train a neural network model, utilizing a designed scene dataset to enhance image detail restoration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution image sensors are used to obtain high-resolution images, then image resolution is improved, but storage capacity and transmission bandwidth requirements increase significantly
Solution Approach 1:
The patent uses low-resolution images as copies of high-resolution images and processes them through a neural network model to generate high-resolution output. Instead of capturing and storing actual high-resolution images, the system creates synthetic high-resolution images from low-resolution inputs, thereby reducing storage and bandwidth requirements while maintaining image quality.
Solution Approach 2:
The patent replaces the mechanical/optical system of high-resolution image capture with a computational system. Instead of using expensive high-resolution sensors and optical equipment, the system uses low-resolution images processed through a neural network model to achieve high-resolution output, substituting physical hardware requirements with computational algorithms.
2Measurement precision
If deep learning neural network is used to perform super-resolution, then image detail restoration is improved, but training data requirements and computational complexity increase
Solution Approach 1:
The patent creates training data by processing existing low-resolution images through the neural network model to generate synthetic high-resolution image pairs. This copying approach allows the system to train on abundant low-resolution images and generate corresponding high-resolution pairs, eliminating the need to collect and store expensive high-resolution training data while maintaining training effectiveness.
Solution Approach 2:
The patent performs preliminary processing of low-resolution images through the neural network model before final output generation. By pre-processing and training on these transformed images, the system prepares training data in advance that reflects the desired high-resolution characteristics, reducing the need for complex real-time processing and extensive high-resolution training datasets.
3Ease of manufacture
If synthetic datasets are used for training super-resolution models, then ease of data generation is improved, but image quality and sharpness deteriorate
Solution Approach 1:
The patent replaces traditional synthetic data generation methods with a neural network-based approach. Instead of using simple mathematical models to generate synthetic images, the system uses a trained neural network model that processes real low-resolution images and generates high-resolution output, substituting simple computational generation with sophisticated learned transformations that preserve image quality and sharpness.
Solution Approach 2:
The patent uses real low-resolution images as the basis for generating training data, rather than creating entirely synthetic images. By copying and processing actual images through the neural network model, the system preserves the natural characteristics and details of real images while generating high-resolution output, thereby maintaining image quality and sharpness that synthetic data cannot achieve.
Data Source
AI summary
The embodiments of this application provide an image data collection system, an image model training method, and a device for improving image resolution. In this application, an image capturing device is used to capture images of an object at different focal lengths to obtain a first image and a second image respectively, and the first image and the second image are processed to obtain a first processed image with high resolution and a second processed image with low resolution, respectively. Image alignment is performed on these processed images to obtain a high-resolution and low-resolution image pair. Many high-resolution and low-resolution image pairs are collected as a training image dataset to train a model for upgrading low-resolution images to high-resolution images. The trained model can significantly improve the ability to restore image details.