Files in this item



application/pdfHuang_Jia-Bin.pdf (263kB)
(no description provided)PDF


Title:Face Representation Learning
Author(s):Huang, Jia-Bin
Subject(s):Electrical and Computer Engineering
Abstract:The pursuit of beauty never stops. However, the strikingly similar faces of the twenty Miss Daegu 2013 contestants cause a serious discussion and debate over the internet. I first took the portrait photography of the twenty contestants and aligned them together to a common reference frame. I then computationally find the landmark positions of their facial features and warp each image to their "mean shape". This"average face" shown above visualizes the mean shape and appearance among the twenty contestants. With the notion of "average", we could further analyze and quantify the similarity among these contestants. The visual similarity among images is central to my research topic in computer vision. It plays an important role in computer vision because the visual appearance of an object captured under different illumination, scale or viewpoint could be dramatically different. To reliably infer the underlying identities and properties of the object or compare the similarity of among different objects, we often need to find a representation that is invariant or robust to such intra- and inter-class variations. The average face thus exemplifies such an approach by seeking a better representation.
Issue Date:2014-05
Rights Information:Copyright 2014 Jia-Bin Huang
Date Available in IDEALS:2014-05-16

This item appears in the following Collection(s)

Item Statistics