Sorted consecutive local binary pattern for texture classification.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Conventional texture classification methods have limitations in encoding spatial transitions and rotation invariance.
• Existing methods often fail to capture discriminative power from patterns with more than two spatial transitions.

Purpose of the Study:

• To propose a novel texture classification method, sorted consecutive local binary pattern (scLBP), that overcomes limitations of conventional approaches.
• To introduce a framework combining scLBP with kd-tree for enhanced texture analysis.

Main Methods:

• Developed scLBP to encode all spatial transitions irrespective of their count, ensuring rotation invariance through pattern sorting.
• Integrated dictionary learning with scLBP using kd-tree for efficient data separation and discriminative coding.
• Proposed a unified framework for utilizing scLBP and kd-tree in texture classification tasks.

Main Results:

• The proposed scLBP framework achieved superior classification rates on CUReT, UMD, and KTH-TIPS2-a datasets.
• Demonstrated competitive performance on UIUC and Outex datasets, with only a marginal difference compared to conventional methods.
• Experimental evaluations confirmed the effectiveness of scLBP in capturing discriminative texture features.

Conclusions:

• The scLBP method offers a significant advancement in texture classification by effectively encoding complex spatial patterns.
• The integration with kd-tree provides a powerful and efficient framework for texture analysis.
• The proposed approach shows strong potential for various computer vision applications requiring robust texture recognition.