Automatic Gray Image Coloring Method Based on Convolutional Network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Automatic image coloring is crucial for quality but traditionally time-consuming.
• Existing methods like prior knowledge, reference pictures, and interactive coloring have limitations, especially with multiple objects.
• Deep learning has shown promise in automatic coloring, but regional control remains a challenge.

Purpose of the Study:

• To develop an automatic image coloring method capable of regional mixed coloring.
• To address the limitation of applying different reference colors to multiple objects within a single image.
• To enhance the quality and efficiency of automatic image coloring through a novel deep learning approach.

Main Methods:

• A deep learning-based method using Convolutional Neural Networks (CNNs) for image coloring.
• Combines instance color image segmentation and image fusion techniques.
• Implements regional color transfer by extracting semantic information and feature maps (content and style) using CNNs.

Main Results:

• The proposed CNN-based method effectively identifies multiple objects and background areas for targeted coloring.
• Successfully transfers colors from designated areas of reference images to corresponding areas in grayscale images.
• Achieves superior coloring effects with advantages in network volume and overall coloring quality compared to existing methods.

Conclusions:

• The developed method offers a significant advancement in automatic image coloring, particularly for complex scenes with multiple distinct regions.
• Provides users with greater control over regional coloring, allowing for foreground and background color differentiation based on references and prior knowledge.
• Demonstrates the potential of deep learning, specifically CNNs, for sophisticated image manipulation tasks like controlled, multi-region image recoloring.