Active Colorization for Cartoon Line Drawings.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Image colorization is crucial for the animation industry but is labor-intensive.
• Existing methods often struggle with maintaining color consistency across multiple related images.

Purpose of the Study:

• To develop a novel semi-automatic method for image colorization using a single reference image.
• To achieve coherent and accurate colorization for raw sketch images in animation production.

Main Methods:

• An active-learning framework for semantic region matching between images.
• Mixed-integer quadratic programming (MIQP) to refine matches using spatial context.
• Efficient integration of user interactions to guide the colorization process.

Main Results:

• The proposed method successfully colorizes sets of images based on a single reference.
• Achieved superior color coherency compared to state-of-the-art deep learning methods.
• Demonstrated the potential for applying the region matching framework to other tasks like color transfer.

Conclusions:

• The novel semi-automatic colorization method offers an efficient and accurate solution for the animation industry.
• The active-learning and MIQP approach provides robust semantic region matching.
• The framework shows promise for broader applications in image processing and computer vision.