Tracking facial feature points with prediction-assisted view-based active shape model

Chao Wang, Xubo Song

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Citations (Scopus)

Abstract

Facial feature tracking is a key step in facial dynamics modeling and affect analysis. Active Shape Model (ASM) has been a popular tool for detecting facial features. However, ASM has its limitations. Due to the finiteness of the training set, it cannot handle large variations in facial pose exhibited in video sequences. In addition, it requires accurate initiation. In order to address these limitations, we propose a novel approach that is capable of providing a more accurate shape initiation as well as automatically switching on multi-view models. We categorize the apparent 2D motions of facial feature points into global motion (the rigid part) and local motion (the non-rigid part) by whether they have relative movement in the image plane. We use the Kalman framework to predict the global motion, and then use adaptive block matching to refine the search for local motion. This will provide an initial shape closer to the real position for ASM. From this initial shape, we can estimate a rough head pose (yaw rotation), which in turn helps choose a suitable view-specific model automatically for ASM. We compare our method with the original ASM as well as with a newly developed competing method. The experimental results demonstrate that our approach have a higher flexibility and accuracy.

Original languageEnglish (US)
Title of host publication2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011
Pages259-264
Number of pages6
DOIs
StatePublished - 2011
Event2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011 - Santa Barbara, CA, United States
Duration: Mar 21 2011Mar 25 2011

Other

Other2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011
CountryUnited States
CitySanta Barbara, CA
Period3/21/113/25/11

Keywords

  • adaptive block matching
  • facial feature points
  • Kalman filter
  • prediction-assisted active shape model

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Vision and Pattern Recognition

Cite this

Wang, C., & Song, X. (2011). Tracking facial feature points with prediction-assisted view-based active shape model. In 2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011 (pp. 259-264). [5771408] https://doi.org/10.1109/FG.2011.5771408

Tracking facial feature points with prediction-assisted view-based active shape model. / Wang, Chao; Song, Xubo.

2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011. 2011. p. 259-264 5771408.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Wang, C & Song, X 2011, Tracking facial feature points with prediction-assisted view-based active shape model. in 2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011., 5771408, pp. 259-264, 2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011, Santa Barbara, CA, United States, 3/21/11. https://doi.org/10.1109/FG.2011.5771408
Wang C, Song X. Tracking facial feature points with prediction-assisted view-based active shape model. In 2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011. 2011. p. 259-264. 5771408 https://doi.org/10.1109/FG.2011.5771408
Wang, Chao ; Song, Xubo. / Tracking facial feature points with prediction-assisted view-based active shape model. 2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011. 2011. pp. 259-264
@inproceedings{eb39bd899ad543dfa47184acb47d4dd9,
title = "Tracking facial feature points with prediction-assisted view-based active shape model",
abstract = "Facial feature tracking is a key step in facial dynamics modeling and affect analysis. Active Shape Model (ASM) has been a popular tool for detecting facial features. However, ASM has its limitations. Due to the finiteness of the training set, it cannot handle large variations in facial pose exhibited in video sequences. In addition, it requires accurate initiation. In order to address these limitations, we propose a novel approach that is capable of providing a more accurate shape initiation as well as automatically switching on multi-view models. We categorize the apparent 2D motions of facial feature points into global motion (the rigid part) and local motion (the non-rigid part) by whether they have relative movement in the image plane. We use the Kalman framework to predict the global motion, and then use adaptive block matching to refine the search for local motion. This will provide an initial shape closer to the real position for ASM. From this initial shape, we can estimate a rough head pose (yaw rotation), which in turn helps choose a suitable view-specific model automatically for ASM. We compare our method with the original ASM as well as with a newly developed competing method. The experimental results demonstrate that our approach have a higher flexibility and accuracy.",
keywords = "adaptive block matching, facial feature points, Kalman filter, prediction-assisted active shape model",
author = "Chao Wang and Xubo Song",
year = "2011",
doi = "10.1109/FG.2011.5771408",
language = "English (US)",
isbn = "9781424491407",
pages = "259--264",
booktitle = "2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011",

}

TY - GEN

T1 - Tracking facial feature points with prediction-assisted view-based active shape model

AU - Wang, Chao

AU - Song, Xubo

PY - 2011

Y1 - 2011

N2 - Facial feature tracking is a key step in facial dynamics modeling and affect analysis. Active Shape Model (ASM) has been a popular tool for detecting facial features. However, ASM has its limitations. Due to the finiteness of the training set, it cannot handle large variations in facial pose exhibited in video sequences. In addition, it requires accurate initiation. In order to address these limitations, we propose a novel approach that is capable of providing a more accurate shape initiation as well as automatically switching on multi-view models. We categorize the apparent 2D motions of facial feature points into global motion (the rigid part) and local motion (the non-rigid part) by whether they have relative movement in the image plane. We use the Kalman framework to predict the global motion, and then use adaptive block matching to refine the search for local motion. This will provide an initial shape closer to the real position for ASM. From this initial shape, we can estimate a rough head pose (yaw rotation), which in turn helps choose a suitable view-specific model automatically for ASM. We compare our method with the original ASM as well as with a newly developed competing method. The experimental results demonstrate that our approach have a higher flexibility and accuracy.

AB - Facial feature tracking is a key step in facial dynamics modeling and affect analysis. Active Shape Model (ASM) has been a popular tool for detecting facial features. However, ASM has its limitations. Due to the finiteness of the training set, it cannot handle large variations in facial pose exhibited in video sequences. In addition, it requires accurate initiation. In order to address these limitations, we propose a novel approach that is capable of providing a more accurate shape initiation as well as automatically switching on multi-view models. We categorize the apparent 2D motions of facial feature points into global motion (the rigid part) and local motion (the non-rigid part) by whether they have relative movement in the image plane. We use the Kalman framework to predict the global motion, and then use adaptive block matching to refine the search for local motion. This will provide an initial shape closer to the real position for ASM. From this initial shape, we can estimate a rough head pose (yaw rotation), which in turn helps choose a suitable view-specific model automatically for ASM. We compare our method with the original ASM as well as with a newly developed competing method. The experimental results demonstrate that our approach have a higher flexibility and accuracy.

KW - adaptive block matching

KW - facial feature points

KW - Kalman filter

KW - prediction-assisted active shape model

UR - http://www.scopus.com/inward/record.url?scp=79958738905&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=79958738905&partnerID=8YFLogxK

U2 - 10.1109/FG.2011.5771408

DO - 10.1109/FG.2011.5771408

M3 - Conference contribution

SN - 9781424491407

SP - 259

EP - 264

BT - 2011 IEEE International Conference on Automatic Face and Gesture Recognition and Workshops, FG 2011

ER -