Image Regularizations Based on the Sparsity of Corner Points.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Traditional image regularizations often assume sparse edges, which limits performance in textured or complex edge areas.
• Existing methods struggle to preserve fine details and edges during noise or blur removal.

Purpose of the Study:

• To propose a new regularization prior based on the sparsity of corner points in natural images.
• To develop and evaluate image denoising and deblurring methods using this novel corner point prior.

Main Methods:

• Introduced a new prior assuming corner points are sparse in natural images.
• Developed a novel regularization technique leveraging Noble's corner measure.
• Experimentally validated the method on image denoising and deblurring tasks.

Main Results:

• The proposed corner point sparsity prior outperforms traditional edge sparsity priors.
• Demonstrated superior performance in preserving edges and textures, especially in challenging regions.
• Achieved excellent results in both image denoising and deblurring applications.

Conclusions:

• The sparsity of corner points is a more robust prior for image regularization than edge sparsity.
• The proposed method offers significant improvements for image inverse problems, particularly in preserving image details.