Nonconvex Structural Sparsity Residual Constraint for Image Restoration

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Structural sparse representation (SSR) effectively utilizes local sparsity and nonlocal self-similarity in natural images for restoration.
• Existing SSR methods can be further improved with more flexible and effective sparse modeling techniques.

Purpose of the Study:

• To propose a novel nonconvex structural sparsity residual constraint (NSSRC) model for superior image restoration.
• To integrate a nonconvex sparsity residual constraint (NC-SRC) prior into the SSR framework.
• To develop an efficient alternating minimizing framework for solving the NSSRC model.

Main Methods:

• The proposed NSSRC model combines Structural Sparse Representation (SSR) with a novel Nonconvex Sparsity Residual Constraint (NC-SRC).
• An alternating minimizing framework is employed to optimize the NSSRC model for image restoration tasks.
• The approach applies a more flexible sparse representation scheme for enhanced natural image modeling.

Main Results:

• The NSSRC model demonstrates improved sparse modeling capabilities for natural images.
• Experimental results on image denoising and deblocking show superior performance compared to existing methods.
• The proposed method achieves high-quality restored images across various datasets.

Conclusions:

• The novel NSSRC model offers a significant advancement in image restoration techniques.
• Integrating NC-SRC into SSR provides a more powerful tool for image processing.
• The developed alternating minimizing framework effectively solves the proposed restoration problems, validating its practical utility.