Spatio-spectral binary patterns based on multispectral filter arrays for texture classification

Area of Science:

• Computer Vision
• Image Processing
• Multispectral Imaging

Background:

• Local Binary Patterns (LBP) are effective for gray-level texture analysis.
• Extending LBP to color images increases computational and memory demands.
• Bayer filter arrays offer raw image texture descriptors, reducing processing steps.

Purpose of the Study:

• To develop an LBP operator for multispectral images that extracts spatial and spectral texture.
• To process raw images directly, avoiding demosaicking and reducing descriptor size.
• To achieve reduced computation cost and high discriminative power.

Main Methods:

• Designed a novel Local Binary Pattern (LBP) operator for multispectral raw images.
• The operator jointly extracts spatial and spectral texture information.
• Evaluated performance on a large dataset using extensive experiments.

Main Results:

• The proposed LBP descriptor significantly reduces computation cost.
• It demonstrates high discriminative power compared to classical LBP on demosaicked images.
• Effective texture analysis directly from multispectral raw data is achieved.

Conclusions:

• The new LBP operator offers an efficient and powerful method for multispectral texture analysis.
• Direct processing of raw multispectral images is advantageous for texture descriptor extraction.
• This approach enhances texture discrimination while minimizing computational overhead.