L₀ Gradient-Regularization and Scale Space Representation Model for Cartoon and Texture Decomposition.

Area of Science:

• Computer Vision
• Image Processing
• Applied Mathematics

Background:

• Traditional image decomposition methods struggle with scale variations in cartoon and texture components.
• Existing techniques often misclassify small, high-contrast textures or large, low-contrast structures.

Purpose of the Study:

• To develop an improved image decomposition model addressing limitations of traditional methods.
• To accurately separate cartoon and texture components while preserving scale features.

Main Methods:

• Introduced a variational model incorporating an L0-based total variation norm for the cartoon component.
• Utilized an L2 norm for the scale-space representation of the texture component.
• Applied a quadratic penalty function to manage the non-separable L0 norm minimization.

Main Results:

• The proposed model effectively decomposes images into cartoon and texture layers.
• Demonstrated superior handling of both small-scale textures and large-scale structures.
• Validated the effectiveness and efficiency of the approach through numerical experiments.

Conclusions:

• The new variational model offers a significant advancement in image decomposition.
• It overcomes the limitations of gradient amplitude-based methods by considering scale features.
• The approach provides accurate and efficient separation of image components.