Michigan State University
Designing convolutional neural networks for face alignment and anti-spoofing
Abstract
dc:description"Face alignment is the process of detecting a set of fiducial points on a face image, such as mouth corners, nose tip, etc. Face alignment is a key module in the pipeline of most facial analysis tasks, normally after face detection and before subsequent feature extraction and classification. As a result, improving the face alignment accuracy is helpful for numerous facial analysis tasks. Recently, face alignment works are popular in top vision venues and achieve a lot of attention. In spite of the fruitful prior work and ongoing progress of face alignment, pose-invariant face alignment is still challenging. To address the inherent challenges associated with this problem, we propose pose-invariant face alignment by fitting a dense 3DMM, and integrating estimation of 3D shape and 2D facial landmarks from a single face image in a single CNN. We introduce a new layer, called visualization layer, which is differentiable and allows backpropagation of an error from a later block to an earlier one. Another application of facial analysis is the face anti-spoofing, which has recently achieved a lot of attention. While face recognition systems serve as a verification portal for various devices (i.e., phone unlock, access control, and transportation security), attackers present face spoofs (i.e., presentation attacks, PA) to the system and attempt to be authenticated as the genuine user. We present our proposed deep models for face anti-spoofing that use the supervision from both the spatial and temporal auxiliary information, for the purpose of robustly detecting face PA from a face video."--Page ii.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jourabloo, Amin
- Contributors dc:contributor
-
- Liu, Xiaoming
- Ross, Arun
- Morris, Daniel
- Boddeti, Vishnu
Subjects
dc:subject × 4Rights
dc:rights- Statement dc:rights
-
- In Copyright
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
etd:47821
isbn:9781392163481
isbn:139216348X
oclc:1117774670
umi:13866123
local:Jourabloo_grad.msu_0128D_16763