Astronomical image denoising by means of improved adaptive backtracking-based matching pursuit algorithm

Area of Science:

• Signal Processing
• Image Denoising
• Sparse Coding

Background:

• Compressive sensing and sparse signal reconstruction are critical in signal processing.
• Existing methods like backtracking-based adaptive orthogonal matching pursuit (BAOMP) have limitations in precision and efficiency.
• Astronomical images often suffer from various noise types, requiring robust denoising techniques.

Purpose of the Study:

• To introduce an improved sparse coding algorithm, the improved adaptive backtracking-based OMP (ABOMP), for enhanced signal reconstruction.
• To address limitations of existing BAOMP methods by incorporating adaptive thresholds, residual feedback, and support set verification.
• To improve the denoising performance and visual quality of astronomical images.

Main Methods:

• Developed the improved adaptive backtracking-based OMP (ABOMP) algorithm with an adaptive threshold, residual feedback, and support set verification.
• Integrated an adaptive step-size mechanism for increased iteration efficiency.
• Combined ABOMP with the K-SVD dictionary learning algorithm for advanced denoising.
• Applied a contrast enhancement method to further improve visual results.

Main Results:

• The ABOMP algorithm demonstrates more precise atom selection compared to BAOMP.
• The adaptive step-size mechanism significantly reduces the number of iterations, leading to higher efficiency.
• Combining ABOMP with K-SVD yields superior denoising effects for astronomical images.
• Experimental results confirm effective removal of both Gaussian and Poisson noise.

Conclusions:

• The proposed ABOMP algorithm offers significant improvements in sparse signal reconstruction and denoising.
• ABOMP provides a more precise and efficient approach to atom selection in sparse coding.
• The algorithm effectively denoises astronomical images corrupted by Gaussian and Poisson noise, enhancing visual quality.