The staircasing effect in neighborhood filters and its solution

Area of Science:

• Image Processing
• Computer Vision
• Partial Differential Equations

Background:

• Classical image denoising methods, such as neighborhood and bilateral filters, average pixel values to enhance signal-to-noise ratio.
• These widely used filters, including variants like SUSAN and sigma filters, often introduce undesirable "staircase effects"—flat regions with artifact boundaries.

Purpose of the Study:

• To explain the origin of the staircase effect in neighborhood filters.
• To propose a modified filter that resolves these artifacts.
• To extend the solution to other algorithms like NL-means.

Main Methods:

• Analyzing the partial differential equation (PDE) underlying neighborhood filters to understand the staircase effect.
• Deriving a stable PDE for a modified neighborhood filter variant.
• Applying the correction strategy to the NL-means algorithm.

Main Results:

• The staircase effect is linked to an ill-posed PDE, similar to the Perona-Malik equation.
• A simple variant of the neighborhood filter effectively eliminates the staircase artifacts.
• The proposed correction method is also applicable to the NL-means algorithm, resolving its similar artifact issues.

Conclusions:

• The staircase effect in image denoising is a consequence of the underlying PDE.
• A modified neighborhood filter and its stable PDE offer a robust solution to staircase artifacts.
• This approach enhances the performance of various image denoising algorithms, including NL-means.