Notice the great importance of landmark localization, we have concentrate on this problem from several aspects including the rule-based methods, Deformable templates, AdaBoost-based method, Example-based Shape Learning, and Several improvements on Active Shape Models.

• Region Growing Search for Eye Localization

A novel approach to localize the human iris centers in facial images is proposed. From the detected face center, our method searches the possible iris edge by gradually growing a rectangle, and then localizes the iris center in an expected region by searching the pixel with the minimal intensity. Our experimental results show that the proposed method has good performance in both detection accuracy and the speed to localize. The algorithm has been successfully applied to a face recognition system to provide the initial locations for aligning faces. See our ICME2002 paper or contact Mr. Cao Bo.

• Facial Landmarks Localization based on AdaBoost

The eye centers, nose tip and moth corners are most critical facial landmarks. Especially the eye centers are used by most of the systems to align the faces. Therefore it is very important to localize the critical facial landmarks precisely and efficiently and it is necessary to exploit elaborate algorithm to guarantee the precision. Benefiting from the experience of the face detection method based on AdaBoost, we attempt to adopt the localization algorithm based on AdaBoost and obtain good results. [See our IWBRS2005 paper or contact Mr.Zhiheng Niu]

• Example-based Shape Learning

A novel Example-based Shape Learning (ESL) strategy has been proposed for facial feature alignment. The method is motivated by an intuitive and experimental observation that there exists an approximate linearity relationship between the image difference and the shape difference, that is, similar face images imply similar face shapes. Therefore, given a learning set of face images with their corresponding face landmarks labeled, the shape of any novel face image can be learned by estimating its similarities to the training images in the learning set and applying these similarities to the shape reconstruction of the novel face image. Concretely, if the novel face image is expressed by an optimal linear combination of the training images, the same linear combination coefficients can be directly applied to the linear combination of the training shapes to construct the optimal shape for the novel face image. Our experiments have convincingly shown the effectiveness and efficiency of the proposed approach in both speed and accuracy performance compared with other methods. See our ICASSP2003 paper or contact Ms.Xiujuan Chai.

• Enhancement of ASM

Motivation:

ASM models all the landmarks with the same local texture modeling called profile. The simple 1D profile model may not localize landmarks accurately enough, especially some key landmarks.

Methods:

We argue that some elaborate local texture matching and searching method should improve the final performance of the ASM. At least we can design complex algorithm for better precision of some critical landmarks (e.g. eye center). We have considered the following three improvements:

(1) Edge information is exploited to construct better local texture models for the landmarks on the face contour [See our ICMI2002 paper or contact Dr. Shiguang Shan];

(2) Elastic Graph Matching is introduced to localize the key landmarks [See our ICBA2004 paper or contact Dr. Shiguang Shan];

(3) AdaBoost is used to localize the key landmarks, and the Landmarks located by local texture matching are exploited in a confidence-constraint manner, i.e., only landmarks with high confidence are used for PCA-based shape modeling. [See our IWBRS2005 paper or contact Mr.Zhiheng Niu].