Related Experiment Video
Updated: Nov 25, 2025
Measuring Connectivity in the Primary Visual Pathway in Human Albinism Using Diffusion Tensor Imaging and Tractography
Published on: August 11, 2016
Direct discriminant locality preserving projection with Hammerstein polynomial expansion
Xi Chen1, Jiashu Zhang, Defang Li
1School of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650093, China. biometrics@yeah.net
A new method, Hammerstein polynomial discriminant locality preserving projection (HPDDLPP), improves classification by directly optimizing discriminant vectors in high-dimensional space. This approach overcomes limitations of kernel-based methods, offering enhanced efficiency and effectiveness in pattern recognition tasks.
Area of Science:
- Machine Learning
- Pattern Recognition
- Dimensionality Reduction
Background:
- Discriminant Locality Preserving Projection (DLPP) is a linear method for classification.
- Kernel-based DLPP (KDLPP) enhances nonlinear description but faces computational challenges and lacks explicit mapping functions.
- KDLPP's inability to obtain optimal discriminant vectors hinders performance.
Purpose of the Study:
- To propose a novel method, Hammerstein Polynomial Discriminant Locality Preserving Projection (HPDDLPP), to address KDLPP's limitations.
- To achieve optimal discriminant vectors with reduced computational burden.
- To enhance the classification accuracy in dimensionality reduction techniques.
Main Methods:
- HPDDLPP directly implements DLPP objectives in a high-dimensional second-order Hammerstein polynomial space.
- The method avoids matrix inversion, simplifying computation.
- It extracts optimal discriminant vectors efficiently.
Main Results:
- HPDDLPP demonstrates effectiveness in face and palmprint recognition tasks.
- Experimental results show superior performance compared to related classical methods.
- The proposed method overcomes the computational burden and mapping issues of KDLPP.
Conclusions:
- HPDDLPP offers an efficient and effective solution for dimensionality reduction and classification.
- The method successfully extracts optimal discriminant vectors without significant computational overhead.
- HPDDLPP represents a significant advancement over existing DLPP and KDLPP techniques for pattern recognition.
