Facial Sketch Synthesis Using 2D Direct Combined Model-Based Face-Specific Markov Network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Facial sketch synthesis is crucial for digital forensics and human-computer interaction.
• Existing systems often lack the ability to replicate specific artistic styles.
• Learning and reproducing unique drawing styles remains a significant challenge in sketch synthesis.

Purpose of the Study:

• To develop a facial sketch synthesis system capable of reproducing a specific artist's drawing style.
• To enhance the similarity between synthesized sketches and input images.
• To demonstrate the framework's effectiveness across diverse facial variations and its applicability in facial recognition.

Main Methods:

• A novel system combining a 2D direct combined model (2DDCM) with a face-specific Markov network.
• Three modules: global (2DDCM for geometry/texture), local (parametric 2DDCM and non-parametric Markov random field for detailed texture), and enhancement (post-processing for lighting).
• Training on a large dataset of image/sketch pairs to learn artistic drawing styles.

Main Results:

• Synthesized sketches exhibit high similarity to input images and ground-truth sketches.
• The system successfully reproduces the unique drawing style of the training artist.
• Effective synthesis across varied facial poses, lighting, and races, even with unseen data.
• Positive results in automatic facial recognition tests.

Conclusions:

• The proposed framework offers a robust method for artist-specific facial sketch synthesis.
• The system achieves superior qualitative and quantitative agreement compared to existing methods.
• The framework demonstrates significant potential for practical applications in facial recognition and digital art.