Introducing oriented Laplacian diffusion into a variational decomposition model

Area of Science:

• Image processing
• Computer vision
• Applied mathematics

Background:

• The Osher, Solé, and Vese (OSV) model offers effective image denoising by emphasizing lower image frequencies.
• A limitation of the OSV model is its tendency to misclassify high-frequency textures as noise.
• Existing models like the Mean Curvature Model (MCM) also face challenges with texture preservation during denoising.

Purpose of the Study:

• To develop an improved image denoising method that preserves high-frequency textures.
• To introduce an oriented Laplacian operator to address the limitations of traditional Laplacian operators in denoising.
• To evaluate the performance of the new oriented Laplacian model against established denoising techniques.

Main Methods:

• Implementation of a novel diffusion process utilizing an oriented Laplacian operator.
• Modification of the OSV model to incorporate the oriented Laplacian for texture-aware denoising.
• Comparative analysis of the proposed model against the OSV model, MCM, and non-local means (NLM) using test images with oriented textures.

Main Results:

• The proposed oriented Laplacian model effectively reduces the inclusion of high-frequency texture in the noise component.
• Quantitative results show higher signal-to-noise ratios (SNRs) for the oriented Laplacian model compared to OSV and MCM.
• Visual assessment indicates superior denoising performance and perceptual quality, especially for images with oriented textures, compared to OSV and MCM.

Conclusions:

• The oriented Laplacian model represents a significant advancement in image denoising, particularly for textured images.
• This method offers a better balance between noise reduction and texture preservation than previous approaches.
• While SNRs may be slightly lower than NLM in some cases, the proposed model provides enhanced visual fidelity.