From heuristic optimization to dictionary learning: a review and comprehensive comparison of image denoising...

Area of Science:

• Image Processing
• Computer Vision
• Signal Processing

Background:

• Image denoising is a critical task in image processing.
• Advances in denoising are linked to improved natural image modeling.
• Existing techniques lack a unified classification framework.

Purpose of the Study:

• To introduce a novel taxonomy for image denoising techniques.
• To provide a structured understanding of state-of-the-art methods.
• To facilitate comparison and analysis of different denoising approaches.

Main Methods:

• Development of a new taxonomy based on image representations.
• Selection and evaluation of representative algorithms within each category.
• Comparative analysis of local versus nonlocal methods.
• Assessment of methods utilizing learned dictionaries.

Main Results:

• Nonlocal methods generally outperform local methods in image denoising.
• Image denoising techniques employing overcomplete representations with learned dictionaries show superior performance.
• The proposed taxonomy effectively categorizes and compares various denoising algorithms.

Conclusions:

• The developed taxonomy offers a valuable framework for understanding image denoising.
• Nonlocal strategies and learned dictionary representations are key to advanced denoising.
• This work serves as a reference and catalyst for future research in image denoising.