Dual-Transfer Face Sketch-Photo Synthesis

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Face sketch identification is crucial for law enforcement but challenging.
• Current methods often fail to preserve identity-specific information, impacting recognition accuracy.
• Automatic sketch synthesis from photos is desired to reduce manual drawing costs.

Purpose of the Study:

• To address the limitations of conventional face sketch-photo synthesis methods.
• To propose a novel dual-transfer framework for improved identity-specific information recovery.
• To enhance face recognition performance in both sketch and photo domains.

Main Methods:

• A novel dual-transfer framework combining inter-domain and intra-domain transfer processes.
• Inter-domain transfer recovers common facial structures by transferring regressors.
• Intra-domain transfer suppresses identity information loss by transferring identity-specific mappings.

Main Results:

• The proposed framework effectively recovers identity-specific information lost in conventional methods.
• Experimental results show more identifiable facial structures compared to state-of-the-art methods.
• The framework yields significantly higher face recognition performance on a benchmark dataset.

Conclusions:

• The dual-transfer framework offers a robust solution for face sketch-photo synthesis.
• Preserving identity-specific details is key to improving face recognition accuracy.
• This approach advances the development of intelligent sketched face identification systems.