Weighted full binary tree-sliced binary pattern: An RGB-D image descriptor.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Inter-pixel similarity is crucial for image analysis and pattern recognition.
• Traditional methods struggle to exploit pixel similarity effectively in decimal form.
• Binary patterns offer a more robust way to represent and compare pixel information.

Purpose of the Study:

• To propose a novel algorithm, Weighted Full Binary Tree-Sliced Binary Pattern (WFBT-SBP), for analyzing inter-pixel similarity in images.
• To leverage the properties of Full Binary Trees to enhance image pattern analysis.
• To apply the developed method for facial similarity verification using RGB-D texture analysis.

Main Methods:

• A 3x3 pixel block is converted into a binarized form to analyze inter-pixel similarity.
• The Sliced Binary Pattern method compares the binary pattern of a central pixel with its 8 nearest neighbors.
• Weights are assigned based on the positional match of binary patterns, forming a Weighted Full Binary Tree structure.
• The WFBT-SBP algorithm traces similarity paths within RGB image dimensions.

Main Results:

• The WFBT-SBP algorithm effectively analyzes inter-pixel similarity patterns in images.
• The method achieved classification accuracies ranging from 76.89% to 77.98% on the KinfaceW-I and KinfaceW-II datasets.
• The algorithm demonstrated strong performance in facial similarity verification tasks.

Conclusions:

• The proposed WFBT-SBP algorithm provides an effective approach for inter-pixel similarity analysis.
• The method shows significant potential for applications in facial recognition and verification.
• The use of binary patterns and weighted trees enhances the ability to capture subtle image similarities.