Coupled Learning for Facial Deblur.

Area of Science:

• Computer Vision
• Image Processing
• Biometrics

Background:

• Facial image blur severely degrades recognition system efficiency.
• Existing blind deconvolution methods struggle with facial images due to reliance on strong edges, which are less common in faces.

Purpose of the Study:

• To develop a robust blind deconvolution method for restoring sharp facial images.
• To enhance the performance of facial recognition systems by improving image quality.

Main Methods:

• Representing point spread functions (PSFs) and estimated intrinsic (EI) sharp face images as linear combinations of pre-defined orthogonal sets.
• Employing a coupled learning algorithm to simultaneously estimate PSF and EI coefficients.
• Generating multiple candidate PSFs and EIs, followed by non-blind deconvolution and blind image quality assessment for optimal EI selection.

Main Results:

• The proposed method effectively restores intrinsic sharp face images from blurry inputs.
• Experimental validation across multiple benchmark face databases (e.g., Extended Yale Face Database B, CMU PIE) shows significant performance improvements.

Conclusions:

• The coupled learning approach offers a robust solution for blind facial image deblurring.
• Restored sharp facial images lead to demonstrably improved facial recognition accuracy.