Super Resolution Image Restoration Tone Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques, such as those using Total Variation (TV) and Bilateral Total Variation (BTV) for super resolution, tend to flatten feature values in texture areas, leading to a loss of tone detail in these regions during image restoration.
Innovation Solution
An image processing device and method that calculates weights based on pixel gradient and direction to reduce regularization constraints in texture areas, allowing for the preservation of feature values by adjusting the regularization term calculation and incorporating reconstruction constraints to restore high-resolution images effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If regularization processing is applied to restore super resolution images, then image restoration quality is improved, but feature values in texture areas are flattened leading to loss of tone detail
Solution Approach 1:
The patent applies different regularization strengths to different image regions by calculating weights based on local gradient characteristics. Texture areas are identified through gradient analysis and assigned lower regularization weights to preserve their feature values, while non-texture areas receive higher weights for effective restoration. This local differentiation resolves the contradiction by allowing selective application of regularization processing.
Solution Approach 2:
The patent dynamically adjusts the regularization parameter (weight) based on local image characteristics, specifically the gradient magnitude and direction. By changing the regularization parameter according to regional features, the method achieves both effective restoration in smooth areas and preservation of texture details in complex regions, thereby resolving the contradiction between restoration quality and information preservation.
2Measurement precision
If regularization constraint is increased to improve restoration accuracy, then restoration precision is improved, but texture area features are overly smoothed
Solution Approach 1:
The patent employs dynamic regularization where the constraint strength varies spatially across the image. The regularization weight is adjusted dynamically based on local gradient properties, being stronger in regions requiring restoration and weaker in regions requiring feature preservation. This dynamic approach resolves the contradiction by making the regularization constraint adaptive rather than static.
Solution Approach 2:
Different regularization constraints are applied to different local regions based on their characteristics. Texture areas identified through gradient analysis receive reduced regularization constraints to maintain feature values, while other areas receive full constraint for accurate restoration. This local quality differentiation resolves the contradiction between overall restoration accuracy and local feature preservation.
3Loss of information
If gradient-based weight calculation is used to preserve texture features, then tone detail preservation is improved, but processing complexity increases
Solution Approach 1:
The patent segments the image into different regions (texture areas vs. non-texture areas) based on gradient characteristics. This segmentation allows the system to apply different processing strategies to different regions, preserving tone details in texture areas while maintaining simplicity in other areas. The segmentation approach balances feature preservation with processing efficiency.
Solution Approach 2:
The image itself provides the information needed for weight calculation through its own gradient properties. The gradient magnitude and direction, which are inherent to the image content, are used to automatically determine regional characteristics and appropriate regularization weights without requiring external complex analysis. This self-service approach reduces processing complexity while maintaining effectiveness.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An objective of the present invention is to reduce tone degradation of a region wherein tone is to be preserved with respect to restoration of a super-resolution image. This image processing device comprises: a weighting computation means for determining a region wherein a feature quantity of an inputted image is stored on the basis of a slope of a feature quantity of a pixel of the inputted image and the direction of the slope, and computing a weighting for reducing a regularization constraint, which is a constraint based on regularization of an image process in the region wherein the feature quantity is stored; a regularization computation means for computing, using the weighting, a regularization constraint of a high-resolution image which is restored on the basis of the inputted image; a reconstruction constraint computation means for computing a reconstruction constraint of the high-resolution image, which is a constraint based on the reconstruction; and an image restoration means for restoring the high-resolution image from the inputted image, on the basis of the regularization constraint and the reconstruction constraint.