Multi scale supervised entropy weighted binary pattern for texture classification.

Area of Science:

• Computer Vision
• Image Analysis
• Machine Learning

Background:

• Texture is a vital visual attribute, but its analysis is challenged by complex imaging environments and scale variations.
• Existing multi-scale texture methods often suffer from complexity, redundancy, and neglect cross-scale feature correlations.

Purpose of the Study:

• To propose an efficient and robust multi-scale supervised entropy-weighted binary pattern for texture classification.
• To enhance the discriminative power and scale robustness of texture feature representations.

Main Methods:

• Introduced a local entropy-weighted histogram using 2D entropy to improve binary pattern operators.
• Developed a Local Entropy-based Optimal Selection Mechanism (LEOSM) for adaptive scale selection in Gaussian scale space.
• Proposed a Cross-Scale Uniformity Supervised Pattern Framework (CSUSPF) for compact, abstract, and discriminative multi-scale texture representation.

Main Results:

• The proposed method demonstrated superior performance on five public texture datasets (Outex, UIUC, CUReT, UMD, ALOT).
• Achieved consistent improvements of 1-5% over the baseline CLBP (Co-occurrence Local Binary Pattern).
• Outperformed several state-of-the-art texture classification approaches.

Conclusions:

• The proposed entropy-weighted binary pattern effectively captures multi-scale and cross-scale texture information.
• The method offers enhanced scale robustness and discriminative power for texture classification tasks.
• This approach provides a more efficient and accurate solution for complex texture analysis challenges.