Deep learning optical image denoising research based on principal component estimation.

Area of Science:

• Optics and Image Processing
• Computational Imaging
• Machine Learning Applications in Optics

Background:

• High-quality denoising of optical interference images typically requires accurate noise level prediction.
• Blind denoising methods, while not needing noise prediction, often result in significant loss of phase information.
• Existing algorithms face challenges in balancing denoising effectiveness with computational efficiency.

Purpose of the Study:

• To develop a fast and high-quality denoising algorithm for optical interference images.
• To preserve phase information while effectively reducing noise.
• To introduce a comprehensive evaluation factor considering both denoising performance and time complexity.

Main Methods:

• A hybrid approach combining Principal Component Analysis (PCA) for noise analysis and residual neural networks for image denoising.
• PCA is utilized to establish an accurate mapping between estimated and true noise levels.
• A residual neural network model is selected based on the noise level mapping for optimal image processing.

Main Results:

• The proposed algorithm effectively denoises optical interference images while maximizing the retention of essential image information, particularly phase data.
• A comprehensive evaluation factor was proposed and validated, demonstrating its utility in assessing algorithm performance.
• Benchmarking against other denoising algorithms showed superior performance in optical interference image preprocessing.

Conclusions:

• The combined PCA and residual neural network algorithm offers a significant advancement in optical interference image denoising.
• The proposed evaluation criterion provides a practical measure for algorithm selection in real-world applications.
• This approach presents a new paradigm for denoising challenges in various optical imaging techniques.