Rosenblum Undergraduate Research Opportunity Award (RUROA)

To provide more research opportunities for undergraduate students, the Department of Computer Science and Engineering at Michigan State University has initiated a "Rosenblum Undergraduate Research Opportunity Award (RUROA)" starting Fall 2013. This award is a generous gift from Mr. Paul Rosenblum, who has a long term passion for supporting undergraduates to involve in research projects. This fall we have a list of research projects available to undergraduate students, especially in the areas of biometrics, computer vision, machine learning, and graphics.

To apply for this award, a undergraduate student should select one of the research project listed below, meet with the faculty advisor, and submit an one-page proposal to Prof. Matt Mutka (mutka@cse.msu.edu) before September 10th, 2013.  Prof. Mutka will forward proposals to a CSE RUROA award committee that will review the proposals and select the award(s) by September 15th, 2013. The awarded undergraduate student will work on the awarded project during Fall 2013 and Spring 2014 semesters, under the supervision of the faculty advisor.  Each awarded project will support the undergraduate student with $3000/semester ($6000/year).  The faculty advisor will receive discretionary funds to support the student.

"3D Correspondence Inference" Faculty Advisor: Yiying Tong Contact e-mail: ytong@msu.edu Research Area(s): Computer Graphics

This project aims at developing tools for correspondence inference, which can be used to facilitate shape retrieval and shape analysis. As 3D scanners and 3D printers have become commonplace, there is a dire need of algorithms to search within and analyze the large 3D geometric datasets. The first step to these goals is often building a correspondence among different shapes. Most existing methods for correspondence inference focus on spatial representations of the correspondences, lacking direct access to global structures. We plan to extend the functional map representation, a compact form storing the linear mapping between low-frequency functions defined on the shapes. Our group is building a new representation based on vector fields. We seek a candidate to help build test datasets, convert various 3D model formats, and design convenient user interface, as well as contribute to the core algorithm. Experience in C++, basic geometric modeling, and digital signal processing is required.

 

"Empirical Downscaling of Model-Simulated Data for Climate Research" Faculty Advisor: Pang-Ning Tan Contact e-mail: tpan@cse.msu.edu Research Area(s): Data Mining

Projections of future climate scenarios is crucial for assessing and understanding the potential impacts of climate change on various human and natural systems, including agriculture, water resource, urban development, tourism, human health, and biodiversity. Recent advances in computer-simulated regional and global climate models have produced vast amounts of data that could be harnessed to improve our ability to generate more reliable projections of the future climate scenarios. However, the spatial scales of these projections (ranging from 10km to 200km resolution) are still far too coarse to be effectively used by various impact assessment studies. The goal of this project is to develop data mining approaches that integrate historical climate observations with model simulated data to obtain more reliable high resolution climate projections. These high-resolutions projections will be utilized by crop models to simulate their future yield production in response to changing climate conditions.

 

"Facial Analysis in First Person Vision" Faculty Advisor: Xiaoming Liu Contact e-mail: liuxm@cse.msu.edu Research Area(s): Computer Vision, Biometrics, Pattern Recognition

The conventional cameras, either stationary camera or PTZ camera, capture visual content from a third-person’s perspective. In contrast, the recently developed first-person vision (FPV) senses the environment and the surrounding people’s activities from a wearable sensor, which are taken from his/her viewpoint. There are a number of advantages in the videos of FPV over those from conventional cameras, such as better viewing angles, close-up distance, multi-view observations, etc. Hence, how to take advantage of them in the visual content analysis is an important question. In the context of human-to-human interaction, this project will explore the technical and practical approaches of performing facial analysis on the wearable platform. For example, while wearing a camera on my head, I am approached by another person, whom I could not remember his/her name. Can the wearable camera take the face image to perform a face matching with a list of subjects that I have encountered before? This project will involve studying various communication schemes between wearable sensors and PC/smartphone, implementing existing visual analysis (such as face recognition or expression recognition) algorithms on smartphone.

 

"Facial Expression Recognition in HCI" Faculty Advisor: Anil Jain Contact e-mail: jain@msu.edu Research Area(s): Biometrics, Pattern Recognition

Consider a typical family room, sensors such as one RGB camera and/or one 3D sensor such as Microsoft Kinect are mounted on top of a TV. When a user is watching TV or playing game, it would be nice to automatically estimate the emotional status of the user. Typically this is achieved by expression recognition from conventional RGB videos of faces. In this project, we will utilize both the RGB camera and 3D sensor to develop a real-time expression recognition system. The student is expected to review the literature, gather or collect relevant datasets, study the pattern recognition algorithms, and develop a demo system on expression recognition.

 

"Fine-grained Ethnicity Estimation from Facial Images" Faculty Advisor: Xiaoming Liu Contact e-mail: liuxm@cse.msu.edu Research Area(s): Computer Vision, Biometrics, Pattern Recognition

We, as humans, have the capability to predict the demographic estimation, such as age, gender, and ethnicity, of the person we encounter in our daily life. As computer scientists, we always like to make a computer has the same intelligence as humans do. In the case of demographic estimation, the goal is to develop algorithms and software to automatic estimate demographic, such as age, gender, and ethnicity, from face images. This is a research topic not only for fulfilling the dream of machine intelligence, but also driven by its potential applications in law enforcement, security control, and human-computer interaction. In the research community, automatic estimation of age, gender, and basic ethnicity groups has been well studied. However, how to estimate fine-grained ethnicity groups (Chinese, Arbs, Japanese, Korea, white American, African American, Hispanic, etc) is still an unaddressed problem. This research project will involve identifying major ethnicity groups, collecting face databases, developing algorithms, and performance evaluation.

 

"Gesture Recognition Using Smart Phones" Faculty Advisor: Arun Ross Contact e-mail: rossarun@cse.msu.edu Research Area(s): Biometrics, Computer Vision, Pattern Recognition

The goal of this project is to develop a gesture recognition application for a smart phone. The application on the smart phone should be able to decipher the hand motion of a user holding it (i.e., the smart phone) and perform certain operations based on the executed motion. For example, if the user were to grasp the smart phone and move it from a horizontal to a vertical position, the application should first recognize this gesture and then connect with another smart phone device in order to transfer data. The application should be able to recognize at least 5 distinct hand gestures based on the accelerometer data.

 

"Iris Recognition Using Smart Phones" Faculty Advisor: Arun Ross Contact e-mail: rossarun@cse.msu.edu Research Area(s): Biometrics, Computer Vision, Pattern Recognition

The goal of this project is to develop an application that can perform iris recognition in a smart phone. The application should capture an image of the iris using the camera in the smart phone, extract features from the image, and compare the extracted features with the template data stored in the smart phone in order to determine if the acquired iris corresponds to the stored template. The feature extraction and matching have to be performed within the smart phone’s computing environment.

 

"Knows Who You Are from Your Hands" Faculty Advisor: Xiaoming Liu Contact e-mail: liuxm@cse.msu.edu Research Area(s): Computer Vision, Biometrics, Pattern Recognition

We hypothesize that an individual computer user has a unique and consistent pattern of hand appearance, shape, and movements, independent of the text, while typing on a keyboard. We have developed a novel biometric modality named “Typing Behavior (TB)” for continuous user authentication. Given a webcam pointing toward a keyboard, real-time computer vision algorithms are developed to automatically extract hand movement patterns from the video stream. For one typing video, the hands are segmented in each frame and a unique descriptor is extracted based on the shape and position of hands, as well as their temporal dynamics in the video sequence. Experimental results have shown that this is an accurate and efficient biometric modality. Now we would like to extend this work toward using hand appearance information for authentication. A large database of typing has been captured in our prior study. This project will involve studying various feature representations for hand texture, develop the system component, and evaluate the system performance.

"Online Movie Recommender System" Faculty Advisor: Rong Jin Contact e-mail: rongjin@cse.msu.edu Research Area(s):  Machine learning

The objective of online recommender system is to automatically recommend items for an individual user by effectively leveraging the rating information provided by other users. The most well known example is book recommendation in Amazon. Another example is movie recommendation in Netflix. In this project, we will focus on developing the state-of-the-art online recommendation system by exploiting the matrix completion theory, which yielded the best performance in the Netflix contest. The figure on the right shows the basic idea of recommendation by matrix completion. We will utilize the movie rating data provided in Netflix contest and meta-data from IMDB to develop an online movie recommendation system.