Flicker Banding Image Generation for Paired Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Obtaining a large amount of paired pieces of training data for image processing is difficult due to cumbersome operations and the need for stable camera positioning, especially in dynamic scenarios when using high-speed shutters to capture alternating light and dark strips.
Innovation Solution
An image generation method that simulates flicker banding by performing a time-domain integral operation on a light source brightness function to generate a stripe image, which is then fused with a normal scenario image to create paired training data, allowing for the generation of a large number of non-repeated flicker banding images without requiring physical photography.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a camera device is fixed to photograph images with different shutter speeds in the same scenario, then paired training data can be obtained, but the operation becomes cumbersome and it is difficult to obtain a large amount of training data in dynamic scenarios
Solution Approach 1:
The patent uses image fusion technology to virtually copy and combine features from multiple images taken under different conditions into a single fused image. This allows the system to generate paired training data without requiring physical recapture of scenes, thereby reducing operational complexity while maintaining data quality for algorithm training
Solution Approach 2:
The system performs preliminary actions by capturing multiple images with different shutter speeds in advance, then uses image fusion to create paired training data from these pre-captured images. This eliminates the need for cumbersome real-time coordinated capture operations while ensuring high-quality paired data generation
2Quantity of substance
If multiple images are captured to obtain paired training data, then sufficient training data can be acquired, but it requires repeated camera captures which is difficult in dynamic scenarios
Solution Approach 1:
The patent merges multiple images captured under different conditions into a single fused image that contains combined features from all source images. This merging process allows the system to generate sufficient paired training data from a single capture session, dramatically improving data acquisition efficiency while maintaining large data quantities for comprehensive algorithm training
3Speed
If high-speed shutter is used with alternating current-powered lighting, then photography can be performed, but alternating light and dark strips appear on the image
Solution Approach 1:
The patent converts the harmful flicker banding effect caused by high-speed shutter photography into a beneficial feature for training data generation. By using image fusion to combine images taken under different lighting conditions, the system generates paired training data that includes flicker banding characteristics, allowing algorithms to learn to handle and remove such artifacts effectively
Data Source
AI summary
An image generation method and an electronic device, and relates to the field of image processing technologies. The method includes: obtaining a first image, the first image including M×N pixel points, where M and N are positive integers; determining a light source brightness function; performing, based on a preset parameter, a time-domain integral operation on the light source brightness function for M times, to obtain M brightness values; generating a second image based on the M brightness values, the second image being a stripe image having M×N pixel points, where brightness values of pixel points in M rows in the second image respectively correspond to the M brightness values; and fusing the first image and the second image into a flicker banding image.