Multi-Scale Sketch Colorization for Harmonious Color Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods automatically color sketch images without considering texture harmony, resulting in visually unharmonious color distributions.
Innovation Solution
A multi-scale feature extraction and encoding-decoding technique that combines sketch texture features with color guide information to generate colored images with harmonious color transitions and preserved texture details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If colors are simply filled into image regions based on user selection, then the coloring process is efficient and automatic, but the color distribution becomes visually unharmonious
Solution Approach 1:
The patent transforms discrete color selection into continuous color space optimization by adjusting color parameters (hue, saturation, value) to achieve harmony. The system modifies color parameters based on relationships between adjacent regions rather than simply filling selected colors, thereby resolving the contradiction between automation and visual quality.
Solution Approach 2:
The patent implements feedback mechanisms where the system evaluates color harmony based on relationships between adjacent image regions and iteratively adjusts color assignments. This feedback loop ensures that color distributions are optimized for visual harmony while maintaining automated processing, addressing the contradiction between efficiency and quality.
2Manufacturing precision
If multi-scale feature extraction and encoding-decoding techniques are used to preserve texture details, then the visual quality of colored images is enhanced, but the processing complexity increases
Solution Approach 1:
The patent segments the image processing into multiple scales (e.g., different resolution levels or feature layers) where each scale captures different texture details. By processing features at multiple scales separately and then combining them through encoding-decoding operations, the system preserves fine texture details while managing complexity through hierarchical organization of processing stages.
Solution Approach 2:
The patent adds the scale dimension to feature processing by extracting and processing features at multiple resolution levels. This dimensional expansion allows the system to capture both coarse and fine texture details simultaneously, enhancing visual quality while the structured encoding-decoding framework manages the increased processing complexity.
Data Source
AI summary
An image processing method, performed by a computer device, comprising: extracting sketch texture features at multiple scales from a sketch image; extracting image noise features at multiple scales from preset noise; determining color guide information corresponding to the sketch image; encoding, for each scale, a noise feature based on a sketch texture feature and the color guide information to obtain multi-scale image features; and performing multi-scale decoding on these image features to obtain a colored image comprising a sketch texture corresponding to the sketch image and a color based on the color guide information. A related training method and apparatus are also provided to develop models for this image processing technique.