Fast cartoon + texture image filters.

Area of Science:

• Image Processing
• Computer Vision
• Mathematical Analysis

Background:

• Variational models for image decomposition separate geometric and textural parts.
• Meyer's models utilize functions with bounded variation and oscillatory distributions.
• Numerical solutions for these models have been historically challenging.

Purpose of the Study:

• To develop a computationally efficient and effective method for image decomposition into geometric and textural components.
• To adapt Meyer's theoretical framework into a practical numerical method.
• To improve the separation of cartoon and texture in images.

Main Methods:

• The study begins with a linear model, equivalent to a low-pass/high-pass filter pair.
• A novel nonlinear filter pair is derived from the linear model by incorporating total variation.
• The proposed method uses a single parameter: the texture scale.

Main Results:

• The new nonlinear filter pair effectively separates geometric and textural image components.
• Experiments demonstrate faster and improved performance compared to existing methods.
• The method shows promise for applications like edge detection.

Conclusions:

• The proposed nonlinear filter pair offers a computationally efficient and effective solution for image decomposition.
• This approach retains the core principles of Meyer's models while improving practicality.
• The method provides a robust tool for image analysis tasks, including edge detection.