Pattern Recognition

A new metric (joint maximal information entropy (JMIE)) is defined to measure a feature subset.

A new feature selection method combining the joint maximal information entropy among features (FS-JMIE) and binary particle swarm optimization (BPSO) algorithm is proposed in this paper.

Experimental results on 5 UCI datasets show the efficiency of the proposed feature selection method.

The proposed method manifests advantage in feature selection with multiple classes.

FS-JMIE shows higher consistency and better time-efficiency than BPSO-SVM algorithm.

We propose a novel non-negative matrix tri-factorization model based on cosparsity regularization to enable the co-feature-selection for co-clustering. It aims to learn the inter-correlation among the multi-way features while co-shrinking the irrelevant ones by encouraging the co-sparsity of the model parameters.

We propose an efficient algorithm to solve the non-smooth optimization problem. It works in an iteratively update fashion, and is guaranteed to converge.

Experimental results on various data sets show the effectiveness of the proposed approach.

Showing that combining original data with a proper nonlinear embedding could be a better basis for adaptive graph learning.

Development of dual representations, i.e., the original data and a nonlinear embedding obtained by an Extreme Learning Machine-based neural network.

Proposing a novel adaptive graph learning method for clustering based on the dual representation.

Extensive experiments on both synthetic and real-world benchmark datasets verified the effectiveness of the proposed method.

We investigate theoretical behaviors of the EM and DA-EM algorithms.

We propose a theoretical lower bound for initialization of the annealing parameter in the DA-EM algorithm.

We theoretically prove that, the EM algorithm for Gaussian mixtures actually exhibits the self-annealing behavior.

Experimental results actually indicate that our theoretical results are applicable in practice.

In this paper, we proposed a new algorithm called LDMLCS to build a family of decomposable models that are appropriate for classification process.

LDMLCS can generate numerous decomposable models, including simple models such as Tree Augmented Naive Bayes (TAN) and complex models with large joint marginal distribution, and some models that could be used for classification on a given dataset.

This algorithm can address the problem of over-fitting and also capture the interaction between different features effectively and consequently, the obtained model is easily interpretable.

A new model for the wring style evolution is considered.

The stylistic structure of a corpus is visualized by the appropriate dendrogram.

The methodology is capable of identifying style changes over time.

The emotional state of a person can be obtained by extracting individual stress index, which provides useful information to health field.

This paper directs the frontier of industrializing stress recognition by focusing on establishing a set of non-contact imaging based classifications for emotional stress (ES) and Physical stress (PS).

The proposed algorithm successfully extracts weak signal from thermal imaging for stress classification.

The classification algorithm is effective with accuracy as high as 90%. This study can lead to a practical system for the noninvasive assessment of stress states for practical applications.

The main contribution of this manuscript is a new training algorithm for binary classification using neural networks.

The training algorithm is based on the minimization of an estimate of the Bayes risk.

Parzen windows method is used to estimate the conditional distributions necessary to compute the probabilities of error included in the Bayes risk.

A new set of training algorithms emerge from this Bayes risk minimization formulation using Parzen windows.

Some interesting relationships with classical training methods are discovered.

Some novel tensor representation for 3D skeleton are proposed.

Tensor-based linear dynamical system is proposed.

Higher recognition accuracy for different datasets are achieved.

This paper advocates the local steering kernel to encode the image patch.

A novel local adaptive contrast measure using local steering kernel is proposed.

This paper presents a patch-based infrared small target detection approach.

The proposed infrared small target detection method achieves promising results.

A novel large margin method for white matter hyperintensity segmentation is proposed.

A supervised large margin algorithm is proposed to learn a global classifier.

A semi-supervised large margin classifier is learned for refinement on test subject.

The proposed model shows competitive performance on subjects with vascular disease.

We propose directional-max-gradient flow to describe gradient propagation process.

We design operators to classify source points into off-plane and in-plane edges.

We propose labeled iterative depth propagation to get better all-in-focus image.

Experiments show effectiveness of proposed algorithm on synthesized and real data.

Modeling contourlet coefficients using t Location-Scale distribution.

Study adaptation between t Location-Scale distribution and contourlet coefficients.

Designing a contourlet domain multiplicative watermark detector using LRT.

Deriving ROC of the watermark detector theoretically and analyzing its performance.

We developed a gradient based decision boundary feature extraction algorithm for neural networks and its incremental version.

The proposed method updates the decision boundaries from sequentially added samples and obtains discriminately informative vectors based on the updated decision boundaries.

When applied to real world databases, it showed noticeably better classification performance than some existing incremental algorithms.

We leverage the manifold structure of visual data in order to improve performance in general optimization problems subject to linear constraints.

As the main theoretical result, we show that manifold constraints can be transferred from the data to the optimized variables if these are linearly correlated.

We also show that the resulting optimization problem can be solved with an efficient alternating direction method of multipliers that can consistently integrate the manifold constraints during the optimization process.

We obtain a simple approach, which instead of directly optimizing on the manifold, and can iteratively recast the problem as the projection over the manifold via an embedding method.