Effective Alternating Direction Optimization Methods for Sparsity-Constrained Blind Image Deblurring.

Area of Science:

• Computer Vision and Image Processing
• Signal Processing
• Internet of Things (IoT) Imaging

Background:

• Single-image blind deblurring is an ill-conditioned inverse problem crucial for IoT imaging sensors.
• Existing methods struggle with accurately estimating blur kernels from single, degraded images.

Purpose of the Study:

• To develop a robust and accurate method for estimating blur kernels in single-image blind deblurring.
• To enhance image restoration quality for IoT imaging applications.

Main Methods:

• Proposed a hybrid regularization method leveraging the sparse and piecewise smooth properties of blur kernels.
• Incorporated L1-norm of kernel intensity and squared L2-norm of intensity derivative for blur kernel estimation.
• Employed a variational image restoration model with L1-norm data-fidelity and second-order Total Generalized Variation (TGV) regularizer for non-blind deconvolution.
• Utilized Alternating Direction Method of Multipliers (ADMM) for solving non-smooth optimization problems.

Main Results:

• The proposed method accurately estimates blur kernels, simplifying the deblurring process.
• Demonstrated superior performance compared to state-of-the-art methods on synthetic and realistic datasets.
• Achieved satisfactory quantitative and qualitative imaging performance.

Conclusions:

• The hybrid regularization approach effectively addresses the challenges of single-image blind deblurring in IoT.
• The method offers a robust solution for accurate blur kernel estimation and high-quality image restoration.