Video Upscaling with Overlapping Frame Groups for Image Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using machine learning models struggle to maintain high image quality when upscaling moving images due to the inability to fully utilize information from previous and subsequent frames, leading to reduced quality in output frames.
Innovation Solution
A method that acquires and processes multiple input frame groups with overlapping time frames, using machine learning models to generate output frames by concatenating or averaging frames that fully utilize information from multiple time points, thereby enhancing image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a machine learning model processes moving images using conventional methods, then the processing can be performed with standard computational resources, but the image quality in output frames is reduced due to inability to fully utilize temporal information from previous and subsequent frames
Solution Approach 1:
The patent applies preliminary action by acquiring and preparing multiple input frame groups with overlapping time frames before processing. The system pre-organizes frame groups where each group contains frames at different times, ensuring that when the machine learning model processes each frame, the necessary temporal context is already prepared and available for optimal processing.
Solution Approach 2:
The patent introduces another dimension by transitioning from processing single frames to processing multiple frame groups with overlapping time frames. This dimensional expansion in the temporal domain allows the model to simultaneously consider multiple time points (previous and subsequent frames) during processing, thereby fully utilizing temporal information and improving image quality.
2Manufacturing precision
If multiple input frame groups with overlapping time frames are processed, then image quality is improved by fully utilizing temporal information, but the processing complexity and computational requirements increase
Solution Approach 1:
The patent applies segmentation by dividing the moving image into multiple discrete input frame groups with overlapping time frames. Each frame group is processed independently through the machine learning model, allowing the system to manage complexity through modular processing units while still benefiting from the temporal context provided by overlapping frames across different groups.
3Manufacturing precision
If conventional single-frame processing is used, then processing speed is maintained, but image quality deteriorates due to incomplete use of temporal information
Solution Approach 1:
The patent merges multiple input frames with overlapping time frames into consolidated frame groups that are processed together. By combining temporal information from multiple frames within each group and leveraging the overlapping structure across groups, the system achieves high image quality while maintaining efficient processing through parallelization of independent frame group processing.
Data Source
AI summary
An image processing method includes acquiring first and second input frame groups from a moving image, first and second output frame groups through a machine learning model, and an output moving image frame based on the first and second output frames. Each of the first and second input frames includes first and second frames. A time of each first frame included in one of the first and second input frames is different from any of times included in the other of the first and second input frames. A time of each second frame included in the one of the first and second input frames overlaps a time of one second frame included in the other of the first and second input frames.