Age estimation via pose-invariant 3D face alignment feature in 3 streams of CNN

Li Sun, Song Qiu, Qingli Li, Hongying Liu, Mei Zhou

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

This paper proposes an algorithm for age estimation intentionally considering the pose variation and local deformation of faces. Pose-invariant patches are extracted in face region, and they are located from the landmarks’ neighborhood in 2D image coordinate. The landmarks can be regarded as the projections of the points on 3D face model, and the projection parameters are estimated by Convolution Neural Network (CNN). Two different structures of CNN are designed for age estimation task. One way is to stack individual patch in the spatial domain, and the stacked image is given to a CNN to make the estimation. The second is to design CNN for each particular patch and CNNs for different patches do not share weights. Together with another CNN trained on the original face region, the three streams for age estimation are combined by late fusion of the output layer. Experiments show that the proposed scheme outperforms other state-of-the-art methods.

Original languageEnglish
Title of host publicationAdvances in Multimedia Information Processing – PCM 2017 - 18th Pacific-Rim Conference on Multimedia, Revised Selected Papers
EditorsBing Zeng, Hongliang Li, Abdulmotaleb El Saddik, Xiaopeng Fan, Shuqiang Jiang, Qingming Huang
PublisherSpringer Verlag
Pages172-183
Number of pages12
ISBN (Print)9783319773797
DOIs
StatePublished - 2018
Event18th Pacific-Rim Conference on Multimedia, PCM 2017 - Harbin, China
Duration: 28 Sep 201729 Sep 2017

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10735 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th Pacific-Rim Conference on Multimedia, PCM 2017
Country/TerritoryChina
CityHarbin
Period28/09/1729/09/17

Fingerprint

Dive into the research topics of 'Age estimation via pose-invariant 3D face alignment feature in 3 streams of CNN'. Together they form a unique fingerprint.

Cite this