Neural Image Color Correction for Boundary Bleeding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for coloring black-and-white images to color images using artificial neural networks often result in color bleeding effects at unclear image boundaries.
Innovation Solution
A method and apparatus using two artificial neural networks to correct color bleeding in images by inputting a sketch image to a first neural network, receiving user input data, and using a second neural network to alleviate the color bleeding effect through training with pseudo user input data generated by detecting and comparing boundaries of colored images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single artificial neural network is used for image coloring, then the coloring process is simple and fast, but color bleeding occurs at unclear boundaries
Solution Approach 1:
The patent divides the coloring process into two stages using two separate neural networks: a first neural network for initial coloring and a second neural network for boundary correction. This segmentation allows each network to specialize in specific tasks, preventing color bleeding at boundaries while maintaining efficient processing speed.
Solution Approach 2:
The patent implements a feedback mechanism where the output of the first neural network is fed into the second neural network for refinement. The second network receives both the original sketch image and the initial colored image, allowing it to correct boundary issues based on the feedback from the first network's coloring results.
2Object-generated harmful factors
If manual boundary correction is performed, then color bleeding can be eliminated, but the process becomes time-consuming and complex
Solution Approach 1:
The patent enables the system to automatically correct color bleeding through the second neural network, which processes the initial colored image and self-corrects boundary issues without requiring manual intervention. The network learns from training data to automatically identify and fix coloring errors at boundaries.
Solution Approach 2:
The first neural network performs preliminary coloring of the entire image before the second network intervenes for boundary correction. This preliminary action completes the basic coloring task efficiently, and only the boundary regions requiring correction are processed by the second network, saving time compared to full manual correction.
3Object-generated harmful factors
If boundary detection accuracy is improved, then color bleeding is reduced, but the complexity of the system increases
Solution Approach 1:
The patent introduces an intermediary mechanism where the second neural network acts as a mediator between the first network's coloring output and the final corrected image. This intermediary processes boundary information and mediates the correction process, improving boundary detection accuracy without requiring complex manual control systems.
Data Source
AI summary
A method of correcting a colored image includes obtaining a colored image by inputting a sketch image to a first artificial neural network, receiving user input data for a target area in the colored image, and obtaining a corrected image in which a color state of the target area has been changed, by inputting the sketch image and the user input data to a second artificial neural network.