A chi-squared-transformed subspace of LBP histogram for visual recognition

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Local Binary Pattern (LBP) is widely used in recognition tasks.
• Subspace methods applied to LBP are optimal for Gaussian distributions, but LBP features deviate from this.
• Non-Gaussian distribution of LBP features limits the effectiveness of traditional subspace approaches.

Purpose of the Study:

• To develop a feature transformation that makes LBP features better fit Gaussian distribution.
• To improve the performance of subspace methods on LBP features.
• To enhance recognition accuracy in various classification tasks using LBP.

Main Methods:

• Proposed a chi-squared transformation (CST) to convert LBP features towards a Gaussian distribution.
• Applied asymmetric principal component analysis (APCA) to the transformed features to handle their asymmetric nature and remove unreliable dimensions.
• Evaluated the combined CST-APCA method on spatial LBP for face recognition, protein cellular classification, and spatial-temporal LBP for dynamic texture recognition.

Main Results:

• The chi-squared transformation (CST) effectively adapts LBP features to better approximate a Gaussian distribution.
• Asymmetric principal component analysis (APCA) successfully addresses the asymmetric nature of the CST-transformed LBP features.
• Extensive experiments demonstrated significant improvements in recognition accuracy across diverse datasets, including face, protein, and dynamic texture recognition.

Conclusions:

• The proposed chi-squared transformation coupled with asymmetric principal component analysis (CST-APCA) significantly enhances recognition accuracy for LBP-based methods.
• This novel feature transformation approach effectively overcomes the limitations of applying standard subspace methods to non-Gaussian LBP features.
• CST-APCA offers a robust solution for improving performance in various pattern recognition applications utilizing LBP features.