Machine Learning Colorization of Line Drawings via Element-Specific Estimation Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automatic colorization technologies are unable to accurately colorize line drawings of characters and elements in games and animations.
Innovation Solution
An image processing method that uses machine learning to generate an estimation model from pairs of line-drawing and colorized images, allowing for accurate colorization of line drawings by identifying elements and modifying colorization layers based on user input, while storing learning data for iterative improvement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional automatic colorization technologies are used, then the process is simple and fast, but the colorization accuracy of line drawings in games and animations is poor
Solution Approach 1:
The system segments the colorization task by creating separate estimation models for different elements (characters, backgrounds, objects). Each model is trained on element-specific learning data, allowing accurate colorization of line drawings while maintaining manageable system complexity through modular architecture
Solution Approach 2:
The system performs preliminary actions by pre-training estimation models using extensive learning data pairs of line drawings and colorized images. These pre-trained models are stored and can be directly applied to new line drawings, achieving high accuracy without requiring complex real-time processing
2Measurement precision
If element-specific estimation models are generated through machine learning, then colorization accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The system performs model training in advance using accumulated learning data, generating element-specific estimation models before they are needed. These pre-trained models are stored for reuse, eliminating the need to retrain models for each colorization task and reducing processing time while maintaining high accuracy
Solution Approach 2:
The system creates and stores copies of trained estimation models for different elements. Once a model is trained on learning data, the model copy can be reused multiple times for colorizing similar elements, significantly reducing the time and computational resources required for actual colorization operations
3Measurement precision
If learning data is stored and reused for model generation, then colorization quality improves, but data management complexity increases
Solution Approach 1:
The system segments learning data by element type (characters, backgrounds, objects), organizing data into element-specific datasets. This segmentation allows the system to manage complex data through modular organization, where each element's learning data is handled independently by its corresponding estimation model
Solution Approach 2:
The system implements feedback mechanisms where colorization results are evaluated and used to generate additional learning data. This feedback loop continuously improves model accuracy by incorporating real-world performance data, managing complexity through automated feedback processing rather than manual data curation
Data Source
AI summary
Provided is an image processing method wherein a computer: generates an estimation model for estimating the colorized image from the line-drawing image for each element through machine learning using the learning data of that element; identifies the element corresponding to the subject line-drawing image; generates the colorized image that is to be paired with the subject line-drawing image, on the basis of the estimation model corresponding to the identified element and the subject line-drawing image; generates a colorization layer of an image file including a line-drawing layer and the colorization layer by using the subject colorized image; extracts the modified colorization layer and the corresponding line-drawing layer as the image pair for learning; and stores a pair of the line-drawing image of the extracted line-drawing layer and the colorized image of the extracted colorization layer, as the learning data, in association with the element corresponding to the estimation model.