Work Experience

• NEC Laboratories America
  Summer Intern, May, 2008 - August, 2008. Cupertino, CA
  
  Project: Emotion Recognition Based on Infrared Image.
  
  The task of this project was to explore the relation between the temperature and the human emotion. The final goal is to embed this technology into vehicle safety system to evaluate driver's emotion status. Because under the different emotion status, the level and the distribution of the temperature on facial components are different. However, the temperature is subject dependent, the normal classification methods are not proper for this application. I introduced latent variable into model learning to assign the coming subject to the proper model and got better performance than the regular classification methods.

• General Electric Global Research Center
  Summer Intern, June, 2007 - August, 2007. Niskayuna, NY
  
  Project: Photo ID based Face Verification System.
  
  This project was a long term one, and began from my summer intern work. The key problems of photo ID based face verification are: 1) The irregular watermark on the face; 2) Age gap between the current subject and the one on the ID; 3) Accessary and moustache impaction on recognition. My duty included two parts: data collection and build the recognition algorithm. For the data set,I collected the data from GE employees. The photo IDs came from the driver licenses and the normal face images were captured by digital camcorder. For the recognition, I used wavelets to remove some of the watermarks, and used Gabor features inside the patches to do recognition. We got around 70\% recognition rate, but it needs to be improved a lot for the practical system.

• Microsoft Research Asia
  Summer Intern, June, 2003 - August, 2003. Beijing, China
  
  Project: Designed and realized one new face recognition algorithm.
  
  Based on the concept of the intra and extra space, the difference of the face images from the same subject are used to build the intra space and the difference of the faces from different subjects constructs the extra space. These two spaces are built in high dimension space which is decided by the feature number. Boosting was used to do dimension reduction and build classifier. We got the best good result on FERET face database by that time.

• GLOBAL ePROCURE
  
  Summer Intern, June, 2005 - August, 2005. Clark, NJ
  
  Web developing. Javascript and VBscript were used for IT support.