Rigid-motion-invariant classification of 3-D textures

Area of Science:

• Computer Vision
• Image Processing
• Texture Analysis

Background:

• 3-D texture discrimination is challenging due to variations in orientation and position.
• Existing methods often fail to account for all 3-D rigid motions.
• Modeling textures as stationary Gaussian random fields provides a mathematical framework.

Purpose of the Study:

• To develop a 3-D rigid-motion-invariant texture discrimination method.
• To define a novel distance metric for 3-D textures robust to rotations and translations.
• To enable reliable statistical similarity testing of 3-D textures under arbitrary rigid motions.

Main Methods:

• Formulating 3-D texture rotations in the digital domain.
• Defining a texture distance invariant under 3-D rigid motions.
• Utilizing Kullback-Leibler divergence between 3-D Gaussian Markov random fields.
• Averaging divergences over all possible 3-D rotations using the Haar measure.

Main Results:

• A novel concept of 3-D texture rotations is formally developed.
• A 3-D rigid-motion-invariant texture distance is defined.
• An algorithm for computing this distance is presented.
• Experimental results demonstrate effective multiscale texture discrimination.

Conclusions:

• The proposed method provides robust 3-D texture discrimination invariant to rigid motion.
• The developed texture distance is effective for statistical similarity testing.
• The approach shows promise for applications involving general 3-D texture models.