2-D impulse noise suppression by recursive gaussian maximum likelihood estimation.

Area of Science:

• Image Processing
• Signal Processing
• Computer Vision

Background:

• 2-D impulse noise significantly degrades image quality.
• Existing noise reduction methods often lack efficiency or effectiveness.
• Advanced algorithms are needed for robust noise suppression.

Purpose of the Study:

• To develop an effective Recursive Gaussian Maximum Likelihood Estimation (RGMLE) method for 2-D impulse noise suppression.
• To introduce RGMLE-C and RGMLE-CS algorithms utilizing spatially-adaptive variances.
• To propose a novel recursion stopping strategy for reliable algorithm implementation.

Main Methods:

• Development of the Recursive Gaussian Maximum Likelihood Estimation (RGMLE) approach.
• Derivation of RGMLE-C and RGMLE-CS algorithms using certainty and joint certainty & similarity information for variance estimation.
• Implementation of a novel recursion stopping strategy based on uncorrupted pixel estimation error.
• GPU-based parallelization for efficient computation.

Main Results:

• RGMLE-C and RGMLE-CS algorithms demonstrate significantly better performance than traditional median filters.
• The proposed algorithms achieve effective suppression of 2-D impulse noise across various noise densities.
• The novel recursion stopping strategy ensures reliable implementation of the RGMLE algorithms.
• Efficient parallel implementation is achieved using GPU technology.

Conclusions:

• The proposed RGMLE-based algorithms provide a highly effective solution for 2-D impulse noise suppression.
• Spatially-adaptive variance estimation and a novel stopping strategy enhance noise reduction performance.
• The GPU-accelerated implementation makes these algorithms practical for real-world applications.