Neural Harmonic Decoder for Image Region Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image enhancement and restoration methods using deep learning-based neural networks struggle to harmoniously adjust local and global features in images, leading to inconsistencies between different regions of an image.
Innovation Solution
A processor-implemented method that generates a local feature representation and a global feature representation based on an input image, region information, and style information, and uses a neural harmonic decoder to determine an adjustment parameter set that harmoniously adjusts the image by considering both local and global features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep learning-based neural networks are used for image enhancement, then image quality can be improved, but inconsistencies between different regions of the image occur due to inability to harmoniously adjust local and global features
Solution Approach 1:
The patent divides the image processing into local and global feature extraction. Local features are extracted from specific regions using a local feature extractor, while global features are extracted from the entire image using a global feature extractor. This segmentation allows independent optimization of local and global characteristics before harmonious integration.
Solution Approach 2:
The patent introduces a style transfer module as an intermediary that harmoniously integrates local and global features. This module acts as a mediator that combines the locally extracted features with globally extracted features to produce a harmonious representation, resolving the inconsistency between regional variations and overall image coherence.
2Measurement precision
If local feature extraction is performed for specific regions, then regional details can be enhanced, but the overall harmony and consistency with global image structure may be compromised
Solution Approach 1:
The patent applies local quality by extracting features from specific local regions of the image using a dedicated local feature extractor. This allows different regions to have their unique characteristics preserved and enhanced according to their specific properties, while still being integrated into the global structure through the style transfer module.
Solution Approach 2:
The patent merges local and global features through the style transfer module, which harmoniously integrates the locally extracted regional details with the globally extracted overall structure. This combining ensures that regional enhancements maintain consistency with the global image composition.
3Stability of the object's composition
If global feature extraction is performed for the entire image, then overall image structure can be preserved, but local regional details and specific style adjustments may be lost
Solution Approach 1:
The patent implements multi-functionality by using the global feature extractor to capture overall image structure while the local feature extractor captures regional details. Both extractors work together in the style transfer module, allowing the system to simultaneously preserve global structure and enhance local details through a unified processing framework.
Data Source
AI summary
A processor-implemented method with image enhancement includes, based on an input image, region information about a local region of the input image, and style information about a target style to be applied to the local region, generating a local feature representation, generating a global feature representation based on the input image, based on the local feature representation, the global feature representation, and the region information, determining an adjustment parameter set, and, based on the adjustment parameter set, generating a retouch result by adjusting the input image.