~ Lexing Xie / Research / ObjRecg
 

Semi-supervised Learning of Patch Based Appearance Model

Overview
This work addresses the problem of learning object appearance from image patches under a semi-supervised scenario. We develop an original version of Gaussian mixture model (GMM) learning algorithm based on partially labeled data; we also propose techniques for saliency-based ranking and selection of the learned mixture components. Experiments show effective yet fast learning in identifying salient patches in the appearance of a class of objects. This work has promising extensions for both advancing state-of-the-art part-based object detectors, and incorporating partial knowledge into multimedia analysis systems in a principled manner.
 
Publications and Reports
L. Xie, P. Pérez (2004). Slightly Supervised Learning of Part-Based Appearance Models, IEEE Workshop on Learning in Computer Vision and Pattern Recognition, in conjunction with CVPR 2004, Washington DC, June 2004 (PDF, Slides)
Last update: August 5, 2004