Spectral Regression for Efficient Regularized Subspace Learning

Abstract

Subspace learning based face recognition methods have attracted considerable interests in recent years, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Locality Preserving Projection (LPP), Neighborhood Preserving Embedding (NPE) and MarginalFisher Analysis (MFA). However, a disadvantage of all these approaches is that their computations involve eigendecomposition of dense matrices which is expensive in both time and memory. In this paper, we propose a novel dimensionality reduction framework, called Spectral Regression (SR), for efficient regularized subspace learning. SR casts the problem of learning the projective functions into a regression framework, which avoids eigen-decomposition of dense matrices. Also, with the regression based framework, different kinds of regularizers can be naturally incorporated into our algorithm which makes it more flexible. Computational analysis shows that SR has only linear-time complexity which is a huge speed up comparing to the cubic-time complexity of the ordinary approaches. Experimental results on face recognition demonstrate the effectiveness and efficiency of our method.

Analyzing the Subspaces Obtained by Dimensionality Reduction for Human Action Recognition from 3d Data

“…Hence, the Spectral Regression framework reformulates the subspace learning problem as a two-step approach, namely graph embedding of the input data and regression for learning the parameters of projection functions [16]. Following this formulation, solely a small set of regularized least-square problems has to be solved, which runs with linear complexity.…”

Research on Multimode Biometric Features Recognition System Adopting Neural Network

“…The problem for artificial neural network is to understand the structure of an algorithm is very difficult, since too many factors may result in excessive filtration, and the best network structure can only be determined based on experiments [6]. Document [9] introduces a method to automatically detect and use of multi-level grades fusion multimode face and fingerprint recognition system, through the replacement of the score and the true score fusion to improve the face recognition system [16,17].…”

“…In addition to the input node, each node is a processing unit with non-linear activation function. MLP application is called to train network back-propagation supervised learning techniques and characteristics in relation with ability of expression of forward MLP has also been authenticated [9]. As for Arbitrary function study, in the three layer networks, it is able to learn any function by adoption of any precision.…”

Machine Learning Methods for Social Signal Processing

“…The input features were obtained by concatenation of the geometric and appearance features. Due to the excessive number of the features, the Spectral Regression (SR) (Cai et al 2007) was applied to select the most relevant features for the intensity estimation of each AU. The intensity classification was performed using AU-specific SVMs.…”