$L_{0}$ Gradient Projection.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• L0 gradient minimization is a powerful edge-preserving filtering technique.
• Existing methods struggle with parameter interpretation, hindering intuitive control over image flatness.
• The parameter in traditional L0 gradient minimization lacks physical meaning, making settings difficult.

Purpose of the Study:

• To propose a new edge-preserving filtering method with an intuitive parameter for controlling image flatness.
• To address the limitations of traditional L0 gradient minimization by reformulating the problem.
• To provide an efficient algorithm for solving the proposed nonconvex optimization problem.

Main Methods:

• Formulated edge-preserving filtering as minimizing quadratic data-fidelity subject to a hard constraint on the L0 gradient (parameter α).
• Introduced a novel use of L0 gradient with a clear physical meaning for the parameter α.
• Developed an efficient algorithm using the alternating direction method of multipliers (ADMM) to solve the nonconvex problem, involving decomposition into subproblems with closed-form solutions.

Main Results:

• The proposed method allows direct control over image flatness via the parameter α, which represents the output image's L0 gradient.
• An efficient ADMM-based algorithm was successfully implemented to find approximate solutions.
• Experimental results demonstrated the advantages and effectiveness of the new edge-preserving filtering method.

Conclusions:

• The novel L0 gradient-based edge-preserving filtering method offers intuitive parameter control and effective image smoothing.
• The proposed formulation and efficient algorithm provide a practical solution for edge-preserving image filtering.
• This approach enhances image processing capabilities by simplifying the control of image flatness and preserving important edges.