Pei, Yuruis an associate professor in the Department of Machine Intelligence, School of EECS. She has received the Ph.D. degree in Computer Science from Peking University, the M.S. degree from Zhejiang University, and the B.S. degree from Central South University. She was a visiting professor in Queen Mary, University of London and the Imperial College, London in 2011-2012. Her research interests include action recognition, image processing, and 3D reconstruction.

Dr. Pei has published more than 30 research papers in international journal and conference, including IEEE TPAMI, IEEE TBME, IEEE TVCG, ICCV. She has several research projects including NSFC, 973 programs, and 863 programs. She has got second prize of Science and Technology Progress Award (Ministry of public security) in 2011. Her main research work is summarized as follows:

1) Unsupervised Image Matching Based on Manifold Alignment. The work addressed the issue of automatic matching between two image sets with similar intrinsic structures and different appearances, especially when there is no prior correspondence. An unsupervised manifold alignment framework was proposed to establish correspondence between data sets by a mapping function in the mutual embedding space. To avoid potential confusions in image matching, she proposed an extended affine transformation to solve the nonrigid alignment in the embedding space. The comparatively tight alignments and the structure preservation can be obtained simultaneously.

2) Registration and Superimposition of Cone-Beam CT Images. The superimposition of cone-beam computed tomography (CBCT) images is an essential step to evaluate shape variations of pre and post-orthodontic operations due to pose variations and the bony growth. A CBCT superimposition method based on the joint embedding of subsets extracted from the CBCT images was proposed. The integration of sparse subsets with context-aware spherical intensity integral (SII) descriptors and correspondence establishment by joint embedding enables the reliable and efficient CBCT superimposition.

3) Unsupervised Random Forest for Affinity Construction. The affinity matrix construction for a large and high-dimensional dataset is a challenging task. She presented an unsupervised random-forest-based metric for the affinity construction of heterogeneous data. Novel criteria for the node splitting are proposed to avoid the rank-deficiency that is encountered while growing the trees. The forest-based combined metric is defined by the length of the common traversing path from the root to leaves as well as the cardinality of the smallest shared parent node. Further, a pseudo-leaf-splitting (PLS) or called affinity propagation algorithm is proposed to consider the spatial transformation to regularize the affinity measures.