CNN Super-Resolution for Sharp Screen Content Without Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional super-resolution techniques suffer from artifacts such as overshoot, undershoot, and aliasing, particularly in screen and gaming content, due to corruption of high-frequency information during downsampling, and existing methods either introduce excessive computational workload or result in blurry outputs.
Innovation Solution
A scalable neural network approach using nearest neighbor downsampling, self-similarity loss, and PReLU activation to enhance image quality, reducing computational resources and preserving sharpness in restored images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional super-resolution techniques are used, then image resolution is increased, but artifacts such as overshoot, undershoot, and aliasing are introduced
Solution Approach 1:
The patent segments the image processing into distinct components: a convolutional neural network processes the illuminance component to eliminate artifacts, while a hardware scaler handles the chrominance component. This segmentation allows each component to be optimized for its specific function, with the CNN focused on artifact removal in the luminance channel.
Solution Approach 2:
The patent applies different processing methods to different parts of the image: the illuminance component (luminance) undergoes convolutional neural network processing to remove artifacts, while the chrominance component (color information) is processed by a hardware scaler. This local quality approach recognizes that different image components have different requirements for artifact suppression.
2Manufacturing precision
If deep learning based super-resolution is used, then image quality is improved, but computational workload increases excessively
Solution Approach 1:
The patent divides the computational workload by segmenting the image into illuminance and chrominance components. Only the illuminance component, which contains the edge information requiring artifact suppression, is processed by the computationally intensive convolutional neural network. The chrominance component is processed by a simpler hardware scaler, significantly reducing overall computational requirements.
Solution Approach 2:
The patent applies full convolutional neural network processing only to the illuminance component rather than the entire image. This partial action approach focuses computational resources where they are most needed (at edges and boundaries) while avoiding unnecessary computation on chrominance data, reducing overall computational workload.
3Ease of manufacture
If downsampling is used for training, then training data is generated, but high-frequency information is corrupted
Solution Approach 1:
The patent separates the illuminance and chrominance components during the downsampling process. By processing only the illuminance component through the convolutional neural network, the method preserves high-frequency edge information in the luminance channel while generating training data, avoiding the information loss that occurs when downsampling the entire image.
Data Source
AI summary
An example apparatus for super resolution imaging includes a convolutional neural network to receive a low resolution frame and generate a high resolution illuminance component frame. The apparatus also includes a hardware scaler to receive the low resolution frame and generate a second high resolution chrominance component frame. The apparatus further includes a combiner to combine the high resolution illuminance component frame and the high resolution chrominance component frame to generate a high resolution frame.