Separation Surfaces in the Spectral TV Domain for Texture Decomposition

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Image decomposition is crucial for analyzing and manipulating image content.
• Existing methods struggle with textures exhibiting gradual variations in scale, contrast, or illumination.
• The total-variation (TV) spectral framework offers a promising approach for multi-scale image analysis.

Purpose of the Study:

• To introduce a novel concept of separation surfaces for image decomposition.
• To develop a method for separating textures with spatially varying characteristics.
• To enable natural and visually convincing texture enhancement or attenuation.

Main Methods:

• A surface is embedded in the spectral total-variation (TV) 3D domain to encode a spatially varying separation scale.
• The TV spectral framework is utilized to decompose images into a continuum of textural scales.
• A texture stratum, defined by a band around a fitted surface, determines the adaptive scale range for each pixel.

Main Results:

• The proposed method effectively separates textures with gradually varying pattern size, contrast, or illumination.
• A texture stratum adaptively defines the scale range for texture identification at each pixel.
• The decomposition allows for natural and visually convincing attenuation or enhancement of image textures.

Conclusions:

• Separation surfaces provide a powerful new tool for image decomposition, particularly for complex textures.
• The method offers precise control over texture scale and properties.
• This approach has significant potential for image editing and enhancement applications.