About me: I am a Ph.D. candidate in computer science at Rutgers University, New Jersey State University. I am a research assistant in Computational Biomedicine Imaging & Modeling Center working with advisor Dr. Dimitris N. Metaxas.

About my research: My research interests are computer vision and machine learning. More specifically, I am working on motion analysis for video sequences, including human action/activity recognition, human behavior analysis, saliency detection, background modeling, object detection and recognition. (Curriculum Vitae).

Summer Internship at Facebook, 2011.

Worked with Ads Optimization Team in Facebook. Researched and developed a new type of feature: Object Level Collaborative Features. It uses Facebook social network for Ads Recommendation System.

Summer Internship at Google Inc, 2010.

Worked with Google Public Data Explorer Team. Researched and developed five data mining algorithms for statistical time series data from statistical agencies, including anomaly detection, pattern matching and kernel methods to analyze the properties of time series data. These algorithms are used as the core algorithms for Public Data Suggestion System. Click my Google intern project poster for more details.

Research Assistant at Chinese Academy of Sciences, China 2006-2007.

Worked with Face recognition group of ICT. Researched in pedestrian detection in video sequences; Developed an automatic moving object detection algorithm; Proposed a powerful 3D features for modeling moving objects and published two papers.

Human Action/Activity Recognition

Working on human action/activity recognition in video sequences. Developed a new feautre Interaction Energy Potential to model people's social behaviors. This feature shows powerful performance on modeling group activities. Details are coming soon.

• [CVPR'11] Xinyi Cui, Qingshan Liu, Mingchen Gao, Dimitris N. Metaxas, Abnormal Detection Using Interaction Energy Potentials, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Colorado Springs, Colorado, June 20-25, 2011.

Fast and Efficient Background Subtraction.

This project addresses the issue of motion saliency detection in video sequences. Saliency detection has attracted a lot of attention in recent years. It aims at locating semantic regions in videos for further video understanding. This project focuses on the issue of motion saliency detection for video content analysis. I proposed a new method Temporal Spectral Residual (TSR) to capture the salient objects from video sequences. Based on the analysis on temporal slices, it can automatically separate foreground motion objects from the background. It also has an effective strategy for adaptive threshold selection and noise removal. Different from conventional background modeling methods with complex mathematical model, this method is simply based on Fourier spectrum analysis, which makes it simple and fast. The effectiveness of this method is demonstrated in the experiments of challenging videos with various types of dynamic background.

• [ACMMM'09] Xinyi Cui, Qingshan Liu, Dimitris N. Metaxas: Temporal Spectral Residual: Fast Motion Saliency Detection, ACM Multimedia (MM), Beijing, China, Oct. 19-24, 2009.

Facial Expression Recognition.

In this project, we proposed a novel framework for video-based facial expression recognition in video sequences, an important topic in human-computer interface, multimedia and visual surveillance. This framework is able to handle the issue of data with various time resolution including a single frame. It explores the intrinsic temporal patterns of facial expression and builds a model that is insensitive to time resolution.

We first use haar-like features to represent facial appearance, which serves as the basic appearance feature in the framework. Then K-Means clustering is performed on the facial appearance features to explore the intrinsic temporal patterns of each expression. Based on the temporal pattern models, we further map the facial appearance variations into dynamic binary patterns using a warping strategy. Finally, boosting learning is performed to construct the expression classifiers. Comparing to previous work, the dynamic binary patterns encode the intrinsic dynamics of expression. This method does not make any assumption on the time scale of facial expression data. Extensive experiments on the Cohn-Kanade database show the promising performance of our method.

• [CVPR'08] Peng Yang, Qingshan Liu, Xinyi Cui, Dimitris Metaxas, Facial Expression Recognition Based on Dynamic Binary Patterns, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jun 24-26, 2008, Alaska, USA.

Pedestrian Detection in Video Sequences

Pedestrian Detection in video sequences is an important application in computer vision. In this project, we developed a framework for robust pedestrian detection. We proposed a new feature 3D haar-like feature to capture the information of people. This feature does not only model the appearance of a person, but also capture the dynamic movement patterns. Boosting method is used to select the most representative features. We also proposed a new way to calculate the feature fast and efficiently. Experiments demonstrate that our approach can outperform Viola's well-known pedestrian detector in both detection accuracy and generalization ability. Other than pedestrian detection, this algorithm also works well for other types of video tasks. It achieves remarkable performance on human activity recognition.

---- 3D haar-like features (C++ Code): I am working on organizing the code and will release it soon. If you are interested in this feature, please leave me a message. Once the code is available, I will send you a notice.

• [ICME'07] Xinyi Cui, Yazhou Liu, Shiguang Shan, Xilin Chen, Wen Gao, 3D Haar-like Freatures for Pedestrian Detection, IEEE International Conference of Multimedia and Expo (ICME), Beijing, China, Jul. 02-05, 2007.

• [PRL'09] Yazhou Liu, Xilin Chen, Hongxun Yao, Xinyi Cui, Chaoran Liu, Wen Gao, Contour-Motion Feature (CMF): A Spatial-Time Approach for Robust Pedestrian Detection, Pattern Recognition Letter, Page 148-156, 2009.

Robust Face Recognition under various Lighting Conditions

In this project, I proposed a framework for robust face recognition under various lighting conditions. There is a tradeoff between lighting removal and details preserving. When the lighting effect is removed, the details of faces are lost too. And also noise is added. My goal in this project is to develop an algorithm to effectively remove lighting effect and at the same time, preserve the details of faces. An adaptive histogram equalization and statistical subspace algorithm are used together to achieve this goal. Experiments on Yale-B face dataset shows significant increase in performance, comparing to the algorithm without lighting removal.

• Technical Report (coming soon...)

Smart Board Game

A game that does not only have rules. It thinks and knows how to beat human players. It is a partially observable stochastic adversarial board game. Greedy strategy and Bayesian inference are embedded to make it smart. Code and slides are available.

---- Check my project page for more details.

Graphical Animations

Automatic generation of objects and motions. It uses procedural methods to produce structured objects, and behavior-based approach for animating crawling bugs. This model takes into account knowledge of object's formation, growth and biomechanics, to produce a great degree of realism. Written in Java and OpenGL.

---- Check my project page for more information and other fun images.

CS336, Database

Led classes for 90 undergraduate students. Introduced the state-of-art database management techniques. Guided them build an online database transaction system using HTML, JSP and mySQL. Had a lot of fun to see the brilliant projects from my students.

CS110, Intro to Computer Science

Led classes for 120 undergraduate students. Guided them how to think in a programming way and how to be a good programmer. Introduced the mainstream technologies of database management, the Internet and programming languages. Really enjoyed to see they started to like computing technologies.

I am a "work hard, enjoy life" type of person. I do a lot of fun activities in my spare time.
See my personal page.