A unified approach to superresolution and multichannel blind deconvolution.

Area of Science:

• Image processing
• Computer vision
• Signal processing

Background:

• Multiple low-resolution frames often suffer from unknown degradations.
• Blind deconvolution and superresolution are challenging inverse problems.
• Prior information about blur kernels is typically required for accurate reconstruction.

Purpose of the Study:

• To develop a new approach for blind deconvolution and superresolution.
• To reconstruct a high-resolution image from multiple degraded low-resolution frames without assuming prior blur information.
• To enhance image quality under severe noise conditions.

Main Methods:

• A regularized energy function is formulated and minimized.
• Regularization is applied in both the image and blur domains.
• Variational principles are used for image regularization, while inter-frame differences guide blur regularization.

Main Results:

• The proposed method achieves stable performance even with significant noise.
• Blur regularization ensures solution consistency by leveraging variations across frames.
• Experiments on synthetic and real data validate the technique's robustness.

Conclusions:

• The presented approach effectively addresses blind deconvolution and superresolution.
• It offers a robust solution for reconstructing high-resolution images from degraded inputs.
• The method shows practical utility in real-world image restoration applications.