Guided Adaptive Image Smoothing via Directional Anisotropic Structure Measurement

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Effective image smoothing requires metrics that distinguish structures from textures, adapting to intensity contrast.
• Existing methods often introduce artifacts like staircasing and blur, compromising structure preservation.

Purpose of the Study:

• To introduce a novel directional anisotropic structure measurement (DASM) for adaptive image smoothing.
• To develop a structure-aware image smoothing scheme that enhances structure preservation and reduces artifacts.

Main Methods:

• Developed Directional Anisotropic Structure Measurement (DASM) based on perceptual properties like anisotropy and local directionality.
• Designed a guided adaptive image smoothing scheme using DASM as a constraint, improving extrema localization and envelope construction.
• Implemented the algorithm on a space-filling curve for computational efficiency.

Main Results:

• DASM effectively characterizes structures and textures irrespective of contrast scales.
• The proposed adaptive smoothing approach significantly suppresses staircase artifacts and blur compared to prior methods.
• Experimental results validate the superiority of DASM for structure identification and the effectiveness of the smoothing scheme.

Conclusions:

• DASM is a valuable metric for identifying dominant structures in adaptive image smoothing.
• The proposed structure-aware adaptive image smoothing method offers superior performance, efficiency, and ease of implementation.