PCA Feature Research Articles

Customer churn prediction on kernel principal component analysis feature abstraction

ion method was presented. The prediction model was designed by combining Kernel Principal Component Analysis (KPCA) and Logistic regression. Experimental results of customer churn prediction for a telecommunication carrier show that the proposed method is superior to original attributes and PCA feature abstraction in hit rate, covering rate, accuracy rate and lift coefficient, which indicates that KPCA abstracts nonlinear features of customer data and provides an effective measurement for customer churn prediction.

Towards a measure of biometric feature information

This paper develops an approach to measure the information content of a biometric feature representation. We define biometric information as the decrease in uncertainty about the identity of a person due to a set of biometric measurements. We then show that the biometric feature information for a person may be calculated by the relative entropy $${D(p\|q)}$$ between the population feature distribution q and the person’s feature distribution p. The biometric information for a system is the mean $${D(p\|q)}$$ for all persons in the population. In order to practically measure $${D(p\|q)}$$ with limited data samples, we introduce an algorithm which regularizes a Gaussian model of the feature covariances. An example of this method is shown for PCA and Fisher linear discriminant (FLD) based face recognition, with biometric feature information calculated to be 45.0 bits (PCA), 37.0 bits (FLD) and 55.6 bits (fusion of PCA and FLD features). Finally, we discuss general applications of this measure.

A method of illumination compensation for human face image based on quotient image

Illumination variation is one of the critical factors affecting face recognition rate. A novel approach for human face illumination compensation is presented in this paper. It constructs the nine-dimension face illumination subspace based on quotient image. In addition, with the aim to improve algorithm efficiency, a half-face illumination image is proposed and the low-dimension training set of the face image under different illumination conditions are obtained by means of PCA and wavelet transform. After processing, two different illumination compensation strategies are given: one is adding light, and the other is removing light. Based on the illumination compensation strategy, we implement the typical illumination sample image synthesis and the standard illumination sample image synthesis on a PCA feature subspace and a wavelet transform subspace, respectively, and the illumination compensation of the gray images and the color images are further realized. Experimental results based on the Yale Face Database B, the Extended Yale Face Database B and the CAS-PEAL Face Database indicate that execution time after compensation is approximately half the time and face recognition rate is improved by 20% compared with that of the original images.

NTF vs. PCA Features for Searching in a Spectral Image Database

NTF vs. PCA Features for Searching in a Spectral Image Database

Subspace-based gearbox condition monitoring by kernel principal component analysis

Feature extraction is a key step for gearbox condition monitoring. The statistical features of the measured vibrations can be used to characterise gearbox conditions; however, their regularity and sensitivity in pattern space are different and may vary considerably under different operating conditions. This paper addresses the non-linear feature extraction scheme from the time-domain features with wavelet packet preprocessing and frequency-domain features of the vibration signals using the kernel principal component analysis (KPCA). Then two different KPCA-based subspace structures are constructed for representing and classifying the gearbox conditions. The proposed methods can extract the non-linear features of gearbox conditions using KPCA effectively, and perform conveniently with low computational complexity based on subspace methods. Experimental analysis with a fatigue test of an automobile transmission gearbox shows that the KPCA features outperform PCA features in terms of clustering capability, and both the two KPCA-based subspace methods can be effectively applied to gearbox condition monitoring.

Feature selection in the independent component subspace for face recognition

This paper addresses the feature selection problem for face recognition in the independent component subspace. While there exists, at least, energy principle to guide the selection of the principle components, the independent components (ICs) are devoid of any energy ranking, and must therefore selected based on their discriminatory power. In addition the independent component features can be selected starting from a much larger pool, or from a combination pool of ICA and PCA features. Four feature selection schemes have been comparatively assessed, and feature subsets are tested on a face database constructed from CMU PIE and FERET databases. The discriminatory features from larger pools are observed to be concentrated around fiduciary spatial details of the nose, the eyes and the facial contour. Overall, face recognition benefits from the feature selection of ICA or PCA components and from the combination of ICA and PCA feature pools.

Face recognition using transform features and neural networks

In this paper we considered face recognition using two Radial Basis Function Network (RBFN) architectures and compared performance with the nearest neighbor algorithm. Performance was also evaluated for feature vectors extracted from face images by using principal component analysis as well as wavelet transform. Raw recognition rates as well as rates with confidence measures were considered. In the RBFN1 architecture, one network was used to discriminate among the classes, while the RBFN2 architecture used one network per class. From the point of view of computations RBFN2 was more efficient than RBFN1 with PCA feature vectors, but its recognition performance was slightly worse than RBFN1. Other experiments showed that RBFN1 was largely superior to the NNA when the amount of computations in both methods was similar. With the use of wavelet features, performance dropped 5–10% in relation to features extracted by PCA. However, in a given implementation, this must be weighed in conjunction with the advantages of using wavelet features, namely, no storage is required for eigenvectors, and they are simpler to compute.