Face Sketch Synthesis Using Regularized Broad Learning System.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Existing face sketch synthesis methods fall into data-driven and model-driven categories.
• Data-driven methods risk detail loss, while model-driven methods involve time-consuming training.
• Refining deep structures in model-driven approaches is particularly slow.

Purpose of the Study:

• To propose an efficient and effective face sketch synthesis method.
• To preserve rich details during the transformation from photographs to sketches.
• To reduce the computational complexity associated with training and testing.

Main Methods:

• A Regularized Broad Learning System (RBLS) is proposed for direct photograph-to-sketch transformation.
• The incremental learning scheme of the Broad Learning System (BLS) allows easy feature mapping expansion without full retraining.
• Bayesian estimation-based regularization is integrated for enhanced feature selection, generalization, and robustness.

Main Results:

• The RBLS method successfully synthesizes high-quality face sketches with rich details preserved.
• Experiments on CUHK student and AR datasets validate the effectiveness and efficiency of the RBLS method.
• The proposed method significantly reduces computational complexity compared to existing approaches.

Conclusions:

• The RBLS offers a superior alternative for face sketch synthesis.
• It achieves high-quality results with significantly improved efficiency.
• The method demonstrates robustness and strong generalization capabilities.