Please use this identifier to cite or link to this item: http://hdl.handle.net/10603/427404
Title: 3d face reconstruction and analysis
Researcher: Suganya Deve P R
Guide(s): Baskaran R
Keywords: Engineering and Technology
Computer Science
Computer Science Artificial Intelligence
Video surveillance cameras
Cosmetic surgeries
Model template
University: Anna University
Completed Date: 2022
Abstract: Three-dimensional human faces play an important role in applications newlinethat are related to fields such as entertainment, medical domain and newlinehuman-computer interactions. Besides, three-dimensional facial information is newlinerequired for the efficient working of various real-time applications such as face newlinerecognition, video surveillance and forensic analysis. With the use of video newlinesurveillance cameras, photo evidence containing the face images of suspects can newlinebe obtained. The obtained images could have been captured under poor lighting newlineconditions, at off-angle poses, and so on. Hence, it is mandatory to generate the 3D newlinemodels of the face images of suspects, so that the suspects can be identified easily. newlineIn the medical industry, 3D face models are greatly useful in planning cosmetic newlinesurgeries for burn patients. newline3D models can be obtained by using multiple views of a face captured newlineusing expensive equipment. But it is not always possible to obtain multiple newlineimages in real-time applications. So, researchers have started focusing on 3D newlineface reconstruction using just a single image. In the proposed framework for 3D newlineface reconstruction from a single frontal face image, as the initial step, landmarks newlineare localized on the database faces with the proposed landmark-mapping strategy newlineemploying a model template. Then, an autoencoder, which is assisted by the newlineproposed energy function to simultaneously learn the facial patch subspace and newlinethe keypoints positions is employed to predict the landmarks. Meta-parameters newlineare incorporated into the energy function during the training phase to enhance newlinethe performance of the autoencoder network in reconstructing the face model. newlineFinally, a unique 3D reconstruction is obtained with the proposed predicted newline
Pagination: xvi, 123p.
URI: http://hdl.handle.net/10603/427404
Appears in Departments:Faculty of Information and Communication Engineering

Files in This Item:
File Description SizeFormat 
01_title.pdfAttached File193.47 kBAdobe PDFView/Open
02_prelim pages.pdf2.36 MBAdobe PDFView/Open
03_content.pdf69.42 kBAdobe PDFView/Open
04_abstract.pdf56.7 kBAdobe PDFView/Open
05_chapter 1.pdf915.11 kBAdobe PDFView/Open
06_chapter 2.pdf128.38 kBAdobe PDFView/Open
07_chapter 3.pdf2.36 MBAdobe PDFView/Open
08_chapter 4.pdf2.81 MBAdobe PDFView/Open
09_chapter 5.pdf1.78 MBAdobe PDFView/Open
10_annexures.pdf607.14 kBAdobe PDFView/Open
80_recommendation.pdf127.3 kBAdobe PDFView/Open
Show full item record


Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

Altmetric Badge: