Diffusion Tensor Images (DTI) Classification

Diffusion tensor imaging (DTI) provides rich information about brain tissue structure, especially white matter and has therefore gained attention in studying pathology of several diseases, such as schizophrenia, in a group-based analysis. In such research, it is important to accurately identify abnormal brain tissue that can classify patients and controls into two groups, for diagnosis, treatment and prognosis. We present a novel group analysis framework, rooted in the pattern classification theory.  Our method directly estimates the overlap between different groups using Bayes error rate to demonstrate separation of different groups when the data could have highly nonlinear structure, which is the case in DTI data. This framework also has the capability of combining multiple measurements that are extracted from tensor data, thus being able to handle multivariate testing.

Related Publications

• Peng Wang, and Ragini Verma, On Classifying Disease-induced Patterns in the Brain using Diffusion Tensor Images, Medical Image Computing and Computer Assisted Intervention (MICCAI), 2008 

• Peng Wang, and Ragini Verma, A Novel Framework for Identifying DTI-Based Brain Patterns of Schizophrenia, Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), 2008

Facial Expression Analysis of Schizophrenia

This research aims to develop computer vision and pattern recognition methods to study the facial expression impairments of patients with neuropsychiatric disorders, such as schizophrenia,  using video and 3D data. Patients with schizophrenia usually have impaired expressions in the form of "flat"  or "inappropriate" affects. In this research, I apply computer vision, machine learning and pattern recognition methods to quantify such affective deficits, with the use of different types of data .

The Wiki page of this project (Internal Users Only)

Background

   What is Schizophrenia?                                                        An Overview of Schizophrenia from NIMH

   Schizophrenia Research in the University of Pennsylvania:     Brain Behavior Center

Methods

2.     Video Based Methods:     analyzing temporal facial expression abnormality using video data

3.     Surface Based Methods:     combining geometric features and texture feature to quantify group differences between patients and healthy controls.

4.     Others:  applying machine learning methods, including Bayesian networks and manifold learning algorithms, to help understand the facial expression impairments in patients

Related Publications

• Peng Wang, F. Barrett, E. Martin, M. Milanova, Raquel E. Gur, Ruben C. Gur, Christian Kohler, and Ragini Verma, Automated Video Based Facial Expression Analysis of Neuropsychiatric Disorders, Journal of Neuroscience Methods, 2008

• Peng Wang, Christian Kohler, Elizabeth Martin, Neal Stolar, and Ragini Verma, Learning-based Analysis of Emotional Impairments in Schizophrenia, Mathematical Methods in Biomedical Image Analysis (MMBIA, with CVPR), 2008 

• Peng Wang, Christian Kohler, Fred Barrett, Raquel Gur, Ruben Gur, and Ragini Verma, Quantifying Facial Expression Abnormality in Schizophrenia by Combining 2D and 3D Features, IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2007  

• Peng Wang, Christian Kohler,and Ragini Verma, Estimating Cluster Overlap on Manifolds and its Application to Neuropsychiatric Disorders, IEEE  Workshop on Component Analysis Methods (with CVPR), 2007 

 Performance Modeling and Prediction of Face Recognition Systems

Face recognition algorithms and systems (and any other biometric systems) always have failures in correctly recognizing input images. Besides improving current algorithms, it is also desirable to predict when the systems and algorithms could fail, and to improve systems based on the performance modeling and prediction. It is a challenging task due to the complicated factors affecting the system performance.

Background

1.     Empirical evaluation in face recognition vendor testing: FERET FRGC,and more ...

2.     Error propagation for computer vision algorithms: S. Yi, R.H. Haralick and L. Shapiro, "Error propagation  in machine vision", Machine Vision and Applications, Vol. 7, pp. 93--114, 1994

3.     Some related work on fingerprint image quality, 2D points recognition, and face recognition.

Methods

In this work, we define the performance of a face recognition (FR) system as its recognition accuracy, and consider the intrinsic and extrinsic factors affecting its performance. The intrinsic factors of an FR system include its gallery images, the FR algorithm and the tuning parameters of the algorithm. The extrinsic factors include mainly query image conditions. Given such definitions, we present methods for performance modeling and prediction. For performance modeling, we propose the concept of "perfect recognition", based on which a performance metric is extracted from the perfect recognition similarity scores (PRSS) to relate the FR system performance to its intrinsic factors. The PRSS performance metric allows tuning FR algorithm parameters offline for near optimal performance. In addition, the performance metric extracted from query images can be used to adjust face alignment parameters online for improved performance. For online prediction of the performance of an FR system on query images, features are extracted from the actual similarity scores and their corresponding PRSS. Using such features, we can predict online if an individual query image can be correctly matched by the FR system, based on which we can reduce the incorrect match rates. Experimental results demonstrate that the performance of an FR system can be significantly improved using the presented methods.

Related Publications:

• Peng Wang, Qiang Ji, and James Wayman, Modeling and Predicting Face Recognition System Performance Based on Analysis of Similarity Scores, IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI),  Vol. 29, No.4, April 2007 

• Peng Wang ,and Qiang Ji, Performance Modeling and Prediction of Face Recognition Systems, CVPR, 2006

• Peng Wang, Lam Cam Tran and Qiang Ji, Improving Face Recognition by Online Image Alignment, ICPR, 2006

  Multi-View Face Detection

This work presents a recursive nonparametric discriminant analysis (RNDA) method to extract powerful features for multi-view face and eye detection. Based on extracted RNDA features, probabilistic classifiers are constructed and combined together to form a strong classifier using AdaBoost. Compared to commonly used Haar wavelet features, RNDA features show better accuracy in detecting complex objects, such as profile faces and eyes.

Related Publications:

• Peng Wang, and Qiang Ji, Multi-View Face and Eye Detection Using Discriminant Features, Computer Vision and Image Understanding (CVIU), Volume 105, Issue 2, February 2007

• Peng Wang, and Qiang Ji, Learning Discriminant Features for Multi-View Face and Eye Detection, CVPR, 2005

• Peng Wang, and Qiang Ji, Multi-View Face Detection under Complex Scene based on Combined SVMs, ICPR, 2004

Eye Detection

We detect eyes from near-frontal faces using discriminant features combined by AdaBoost. Experiments demonstrate that face recognitions using our automatically localized eyes provide accuracy comparable to those using manually marked eyes.

Fig. Summary of face recognition accuracy with automatic and manual eye positions on FRGC V1.0 database

 A Real-time System

I build a near real-time system which can perform face and eye detection, and online face recognition simultaneously. The system can run near real-time at a laptop using a simple CCD camera.  The demo video captured from screen is shown as follows. In this demo, the name of recognized person is shown above his/her face.

Demo: Video

Face Tracking and Pose Estimation

To address the drifting problems associated with the traditional object tracking under significant changes in both objects and environmental conditions, a novel collaborative tracking algorithm is presented. The collaborative tracking method probabilistically combines measurements from specific face models with measurements from a generic face model in a dynamic Bayesian network for robust multi-view face tracking. In addition, the probabilistic tracking results are used to incrementally adapt the specific face models to individuals. The collaborative tracking method can handle large face pose changes, and can efficiently build specific face models online.

Demo: Video1, Video2

Related Publications:

• Peng Wang and Qiang Ji, Multi-View Face Tracking with Factorial and Switching HMM, WACV, 2005

• Peng Wang and Qiang Ji, Robust Face Tracking via Collaboration of Generic and Specific Models, IEEE Trans. on Image Processing, 2008 Jul. 17(7):1189-99