Image smoothing method based on global gradient sparsity and local relative gradient constraint optimization.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Image smoothing is crucial for various applications, but preserving structural details while removing texture remains a significant challenge.
• Existing methods often struggle to differentiate between texture and structure, leading to loss of important image information.

Purpose of the Study:

• To develop an advanced image smoothing method that effectively removes texture while preserving main image structures.
• To enhance edge preservation capabilities in image smoothing techniques.

Main Methods:

• A novel image smoothing method is proposed, utilizing global gradient sparsity and local relative gradient constraints optimization.
• A multi-directional difference constrained global gradient sparsity decomposition is employed to generate a guidance image with reduced texture gradients.
• An edge-aware operator, based on local gradient constraints and the luminance channel, is constructed to weaken repetitive texture gradients.

Main Results:

• The proposed method successfully reduces texture interference by weakening texture detail gradients.
• The edge-aware operator effectively captures structural information for guiding global image optimization.
• Experimental results demonstrate superior edge preservation compared to existing advanced smoothing methods.

Conclusions:

• The developed image smoothing method offers significant advantages in preserving image edges.
• The combination of global gradient sparsity and local relative gradient constraints provides an effective approach for texture removal while maintaining structural integrity.