Please use this identifier to cite or link to this item:
Title: Hallucinated 3d face model from a single 2d low-resolution face using machine learning
Authors: Naleer, H.M.M
Yao, LU
Keywords: face hallucination
3D face model
2D low-resolution
Issue Date: 2-Aug-2014
Publisher: South Eastern University of Sri Lanka, University Park, Oluvil #32360, Sri Lanka
Citation: Proceedings of 4th International Symposium 2015 on " Emerging Trends and Challenges on Sustainable Development”, p. 58
Abstract: In this paper, the 3D face hallucination system is proposed on both 2D training face images as well as respective 3D training face models with grey-level. The proposed method hallucinates the 3D high-resolution model patch using same position of each image patches of interpolated 2D training image and 3D high-resolution training face model for low-resolution input image. Firstly, the optimal weights of the 2D input low-resolution image position-patches are estimated with the corresponding 2D low-resolution training image patches. The canonical correlation analysis (CCA) is used to learn the mapping between the 2D interpolated face training image and the 3D face model with respect to their weights. Secondly, the corresponding 3D face model patch with weight by matching high score among the 2D interpolated training image patches and 3D training face model is selected. Finally, the 3D high-resolution facial model is formed by integrating the hallucinated 3D patches which are obtained through mapping patches with respective weights. In order to evaluate the performances of the above approaches, we used example based learning methods to obtain the high-resolution output for a low-resolution input. In this approach, we used the available frontal data sets such as FERET, CAS-PERL and CMU to analysis the performance and some parameters are also considered, which may affect the results from the above proposed method.
ISBN: 978-955-627-053-2
Appears in Collections:4th International Symposium - 2014

Files in This Item:
File Description SizeFormat 
4 th Int Symp_2014_Article_65_Pages from 501-510.pdfArticle 65435.34 kBAdobe PDFThumbnail

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.