Jump to: Home | Biography | Research | Publications | Course |  Labs & groups |Links | Some interesting images| My Photo


1.   Face Detection

The goal of face detection is to determine whether there are any faces in a given image, and return the location and extent of each face in the image if one or more faces are present. I engaged in a project of National Hi-tech Program (863), and a project sponsored by the Natural Sciences Fund of China. (Details on the  methods)

An example of detected faces

One of my dreams is to find the faces in the following two images by face detectors although they are hard for human being.

If you can find the 9th face, your IQ hits 180.

If you can find the 11th face, your IQ hits 180.5.  :o)

2.   Dynamic Texture Segmentation

Dynamic texture (DT) is an extension of texture to the temporal domain. We address the problem of segmenting DT into disjoint regions in an unsupervised way. Each region is characterized by histograms of local binary patterns and contrast in a spatiotemporal mode. It combines the motion and appearance of DT together.  (Link)

Dynamic Texture Segmentation

Three steps of our methods





Pixelwise classification

Some experimental results performed on a real challenging sequence

3.   Face Synthesis

The problem to be discussed is that the enrollment face samples and query face samples are in the different lighting conditions in face recognition domain. Herein, the former is captured in the visual lighting (VIS) condition and the later in the near infrared (NIR) condition. It is difficult to directly match the face samples captured in these two lighting conditions since they are imaged in the completely different lighting conditions. To this task, we propose a new method for synthesizing a face image captured in VIS condition from the one in NIR. (PDF)

(a) Problem: Synthesize a virtual face sample x′  in S4 for an input face image x in S3 given the training set S1 and S2.

Motivation: Although using the active NIR imaging leads to truly illumination invariant face recognition and excellent results for indoor, cooperative user applications, it requires that the face templates of enrollment be also created from NIR face images, in addition to the use of NIR images for query. However, in many applications, templates of users were produced from VIS images taken in the visible light spectrum, such as passport and driver license photos. The ICAO (International Civil Aviation Organization), the International Organization for Standardization (ISO), and the International Electrotechnical Commission (IEC) standards recommend for taking VIS photos. A straightforward matching between the two types of templates is not effective mainly because of their different spectral properties.

(b)Training procedure. (c) Testing procedure

4. Spatiotemporal interest volume descriptor

Visual motion carries information about the dynamics of a scene. Automatic interpretation of this information is important when designing computer systems for visual navigation, video indexing, surveillance, human-computer interaction, browsing of video databases and other growing applications. This wide area of applications motivates the development of generic methods for video analysis that do not rely on specific assumptions about the particular types of motion, environments and imaging conditions. Analyzing and interpreting video is a growing topic in computer vision and its applications. Video data contains information about changes in the environment. One issue of motion representation is for the purpose of detecting and recognizing motion patterns in video sequences. Some examples of human actions are shown in the following figure.

Realistic samples for three classes of human actions: (a) kissing; (b) answering a phone; and (c) getting out of a car.



Jump to: Home | Biography | Research | Publications | Course | Labs & groups |Links | Some interesting images| My Photo

Any question about this page? Turn to its host!

This page is maintained currently by chenjie. Last modified: 2010-2-11, 11:47