Image restoration for synthetic aperture systems with a non-blind deconvolution algorithm via a deep convolutional...

Area of Science:

• Optics and photonics
• Image processing
• Machine learning for imaging

Background:

• Optical synthetic aperture imaging systems enhance telescope resolution but suffer from image blurring due to sub-mirror dispersion and sparsity.
• Image restoration is crucial for mitigating these degradations and achieving high-quality images.

Purpose of the Study:

• To propose a non-blind deconvolution algorithm for restoring images from optical synthetic aperture systems.
• To enhance image clarity and contrast lost during the imaging process.

Main Methods:

• A synthetic aperture convolutional neural network (CNN) was trained as a denoiser.
• An improved half-quadratic splitting algorithm divided restoration into deconvolution and denoising subproblems.
• The CNN denoiser guided the deconvolution step using learned gradients.

Main Results:

• The proposed method significantly improved image quality compared to conventional algorithms.
• Peak signal-to-noise ratio increased from 23.7 dB to 30.8 dB at a 40 dB signal-to-noise ratio.
• Structural similarity index improved from 0.78 to 0.93.

Conclusions:

• The developed CNN-based non-blind deconvolution algorithm is effective for image restoration in optical synthetic aperture systems.
• Both quantitative and qualitative evaluations confirm the method's superior performance in enhancing image quality.