Technologies on Display

G3 - Automatic Face Recognition

(1) Automatic Face Recognition
Due to growing demands in areas such as law enforcement identification, banking and security system access authentication, and anti-terrorist video surveillance, automatic face recognition has attracted great attention in recent years. Comparing to other biometrics technologies such as fingerprint and iris recognition, face recognition is the best suited for surveillance of busy public places like airports and the customs, since it does not require those being watched to cooperate by looking into an iris scanner or putting a hand on a fingerprint reader. It is the only biometrical technology that does not require those being checked to cooperate. Face recognition devices can work with the video feeds from the cameras that are already ubiquitous in public places. It is also much easier for authorities to obtain a suspect's photo than to obtain his other biometrical identifiers. We recently developed a new face graph model based on discrete wavelet transformation. The method can locate key facial features such as nose tip and mouth corner at a rate a hundred times faster than the traditional elastic graph. In addition, we established a unified framework for the popular subspace based face recognition approaches. A unified subspace analysis algorithm is developed under this framework that gives higher recognition accuracy than all the traditional subspace methods.


  • Surveillance: Home security, anti-terrorist, airport, and casinos.
  • Access control: Bank, private company, and network security.
  • E-commerce: Fraud control, ATM, and internet banking ID.
  • Law Enforcement: Duplicate ID detection and mugshot database search.
  • Government: Customs control, social security service, motor vehicle registration, and immigration.
  • Multimedia: Information retrieval, video compression, and telecommunication.
  • Other: Finding lost relatives and personalized service.


  • No touching of any equipment.
  • Does not require those being checked to cooperate.
  • Easy to correct error and easy to obtain data.
  • Fast computation using discrete wavelet transform features.
  • Robust to lighting and expression changes.

(2) Face Sketch Recognition and Face Hallucination
An important application of facial recognition is to assist law enforcement. For example, automatic retrieving of photos of suspects from police mug-shot databases can help police narrow down the number of potential suspects quickly. However, in most cases, only the drawing based on the recollection of an eyewitness is available. Therefore searching an image database by using a sketch drawing is definitely useful. Despite the great need of such an automatic photo retrieval system using face sketches, limited research has been carried out in this area. We have developed a unique face sketch recognition system that achieves a higher recognition performance than human beings. Given a simple sketch of a person, the system can potentially search through hundreds of thousands of photos to hunt down the target person. Such a facial recognition system capable of automatic classification of both face photos and drawn sketches can be a crucial investigator support tool for practical law enforcement in any police departments and national security agencies around the world.
In video surveillance, the faces of interest are sometimes small in size because of the great distance between the camera and the target people. As such many detailed facial features are lost in the low-resolution images. For identification, especially by human, it is useful to render a high-resolution large face image from the low-resolution one. This technique is called face hallucination. Using the special property of face structure, we develop an effective algorithm that can learn from a training set containing high- and low- resolution image pairs. The reconstructed photo by our hallucination algorithm appears realistic when compared to the original high-resolution image.


  • Law Enforcement
  • Sketch generation for entertainment application
  • Video Surveillance


  • The first and only working system for sketch recognition.
  • Better performance than human being
  • Simple and fast algorithm


Principal Investigator
Prof. XiaoOu Tang
Department of Information Engineering