Combining LBP difference and feature correlation for texture description.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Effective texture characterization relies on multiple visual cues.
• Local Binary Patterns (LBP) are successful but are discrete, hindering combination with other features.
• A numerical variant is needed to overcome LBP's non-numerical constraint.

Purpose of the Study:

• To propose a numerical variant of LBP, named LBP difference (LBPD).
• To develop a compact descriptor combining LBPD with other features for enhanced texture analysis.
• To evaluate the performance of the proposed descriptor on public datasets.

Main Methods:

• Introduced LBP difference (LBPD) as a numerical texture descriptor.
• Developed the covariance and LBPD descriptor (COV-LBPD) by combining LBPD with other features using a covariance matrix.
• Validated the descriptor's effectiveness on publicly available datasets.

Main Results:

• The proposed LBPD is simple, rotation invariant, and computationally efficient.
• The COV-LBPD descriptor effectively captures correlations between LBPD and other features in a compact representation.
• Experimental results demonstrate promising performance of COV-LBPD on texture datasets.

Conclusions:

• The LBPD offers a numerical solution to combine LBP with other texture features.
• The COV-LBPD descriptor shows significant potential for advanced texture characterization.
• The proposed method advances texture analysis by enabling compact and discriminative feature representation.