Deep Learning Image Upscaling with Region-Specific Sharpness Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods using Deep Learning (DL) models cannot perform upscaling with different sharpness for each region of an image.
Innovation Solution
A training method that involves acquiring low-resolution and high-resolution training images, generating interpolated and sharpness-adjusted training images, and training a machine learning model to upscale images with varying sharpness across regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a DL model is used for image upscaling, then image resolution is improved, but uniform sharpness across all regions is lost
Solution Approach 1:
The image is divided into multiple regions based on luminance characteristics (high-luminance regions and low-luminance regions). Different processing approaches are applied to each region: high-luminance regions use one type of sharpness processing while low-luminance regions use another type, allowing region-specific sharpness control in the upscaled image
Solution Approach 2:
Different sharpness processing methods are applied to different regions of the image based on their luminance characteristics. High-luminance regions receive different treatment compared to low-luminance regions, enabling each region to have optimized sharpness properties suitable for its characteristics
2Productivity
If traditional interpolation is used for upscaling, then processing speed is maintained, but image quality and sharpness are degraded
Solution Approach 1:
A trained DL model serves as an intermediary between the input low-resolution image and the final upscaled output. The model learns optimal upscaling patterns from training data and applies them during inference, achieving high-quality results without requiring complex real-time processing
Solution Approach 2:
The DL model is pre-trained on a dataset of image pairs before deployment. This preliminary training phase allows the model to learn effective upscaling patterns, so that during actual use, high-quality upscaling can be achieved through inference rather than complex real-time computation
3Measurement precision
If region-specific sharpness processing is implemented, then image quality is improved, but processing complexity increases
Solution Approach 1:
The image is automatically divided into regions based on luminance thresholds, with high-luminance regions processed differently from low-luminance regions. This segmentation approach enables region-specific processing without requiring manual intervention or complex processing logic
Solution Approach 2:
Different processing parameters (sharpness settings) are applied to different regions based on their luminance characteristics. The system automatically adjusts processing parameters according to the region type, achieving quality improvement through parameter variation rather than complex processing algorithms
Data Source
AI summary
A training method includes the steps of acquiring a first training image and a second training image corresponding to the first training image, having a higher resolution than the first training image, generating a third training image by enlarging the first training image by interpolation, generating a fourth training image with different sharpness for each region based on the second training image, the third training image, and at least one of a first region having a luminance value which is equal to or larger than a predetermined value and a second region having a luminance change rate which is equal to or larger than a predetermined rate in the first training image, and training a machine learning model based on the first training image and the fourth training image.