A two-dimensional Partial Least Squares with application to biological image recognition

The Partial Least Squares(PLS) is a novel multivariate data analysis method developed from practical applications in real world. It is not influenced by the total scatter matrices of training samples being singular or not. So PLS can efficiently deal with the case of high-dimensional space with only small sample size such as biological feature recognition. The standard PLS firstly reshapes images into vectors. In order not to destroy the inherent structure information, a two-dimensional PLS is proposed which can extract features being more discriminative and dramatically reduces the computational complexity compared to the standard PLS. The proposed method is applied to face and palm biometrics and is examined using the FERET and PolyU palmprint database. Experimental results show that 2DPLS is a good choice for real-world biometrics recognition.