Adaptive regularized constrained least squares image restoration.

Area of Science:

• Image restoration
• Signal processing
• Computational imaging

Background:

• Image degradation in noisy environments is a significant challenge.
• Traditional Constrained Least-Squares (CLS) methods often use a global regularization parameter, which can be suboptimal.
• Gaussian impulse response blurs are common in imaging systems.

Purpose of the Study:

• To develop an improved image restoration technique for noisy, blurred images.
• To introduce a localized regularization parameter approach for Constrained Least-Squares (CLS).
• To reduce border ringing and denoise the restored signal.

Main Methods:

• A novel Constrained Least-Squares (CLS) approach with point-wise regularization parameters.
• Utilizing wavelet transform coefficients on the finest scale for parameter determination.
• Constraining the weighted standard deviation of wavelet coefficients using a local intensity variation measure.
• Manipulating finest-scale wavelet coefficients to decrease border ringing.
• Applying wavelet denoising techniques for significant noise reduction.

Main Results:

• The proposed method demonstrates superior visual quality compared to global CLS.
• Quantitative evaluation shows lower mean squared error (MSE) than global CLS.
• Effective reduction of border ringing artifacts.
• Successful denoising of the inverse solution when noise is significant.

Conclusions:

• The adaptive, point-wise regularization parameter approach significantly enhances image restoration.
• This method offers improved performance over traditional global regularization techniques.
• The technique is effective for restoring images degraded by Gaussian blur and noise.