Multiscale discriminant saliency for visual attention

Anh Cat Le Ngo, Li-Minn Ang, Guoping Qiu, Kah Phooi Seng

Research output: Book chapter/Published conference paperConference paper

Abstract

The bottom-up saliency, an early stage of humans' visual attention, can be considered as a binary classification problem between center and surround classes. Discriminant power of features for the classification is measured as mutual information between features and two classes distribution. The estimated discrepancy of two feature classes very much depends on considered scale levels; then, multi-scale structure and discriminant power are integrated by employing discrete wavelet features and Hidden markov tree (HMT). With wavelet coefficients and Hidden Markov Tree parameters, quad-tree like label structures are constructed and utilized in maximum a posterior probability (MAP) of hidden class variables at corresponding dyadic sub-squares. Then, saliency value for each dyadic square at each scale level is computed with discriminant power principle and the MAP. Finally, across multiple scales is integrated the final saliency map by an information maximization rule. Both standard quantitative tools such as NSS, LCC, AUC and qualitative assessments are used for evaluating the proposed multiscale discriminant saliency method (MDIS) against the well-know information-based saliency method AIM on its Bruce Database wity eye-tracking data. Simulation results are presented and analyzed to verify the validity of MDIS as well as point out its disadvantages for further research direction. © 2013 Springer-Verlag Berlin Heidelberg.
Original languageEnglish
Title of host publicationICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V
Place of PublicationHeidelberg
PublisherSpringer
Pages464-484
Number of pages20
Volume7971
ISBN (Print)9783642396366
DOIs
Publication statusPublished - 2013
EventThe 13th International Conference on Computational Science and Its Applications : ICCSA 2013 - International University, Hi Chi Minh City, Viet Nam
Duration: 24 Jun 201327 Jun 2013
https://web.archive.org/web/20120922002912/http://iccsa.org/

Conference

ConferenceThe 13th International Conference on Computational Science and Its Applications
CountryViet Nam
CityHi Chi Minh City
Period24/06/1327/06/13
Internet address

Fingerprint

Labels

Cite this

Le Ngo, A. C., Ang, L-M., Qiu, G., & Seng, K. P. (2013). Multiscale discriminant saliency for visual attention. In ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V (Vol. 7971, pp. 464-484). Heidelberg: Springer. https://doi.org/10.1007/978-3-642-39637-3_37
Le Ngo, Anh Cat ; Ang, Li-Minn ; Qiu, Guoping ; Seng, Kah Phooi. / Multiscale discriminant saliency for visual attention. ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V. Vol. 7971 Heidelberg : Springer, 2013. pp. 464-484
@inproceedings{9a6eca615de7416ba6f2c0a9aa5414d8,
title = "Multiscale discriminant saliency for visual attention",
abstract = "The bottom-up saliency, an early stage of humans' visual attention, can be considered as a binary classification problem between center and surround classes. Discriminant power of features for the classification is measured as mutual information between features and two classes distribution. The estimated discrepancy of two feature classes very much depends on considered scale levels; then, multi-scale structure and discriminant power are integrated by employing discrete wavelet features and Hidden markov tree (HMT). With wavelet coefficients and Hidden Markov Tree parameters, quad-tree like label structures are constructed and utilized in maximum a posterior probability (MAP) of hidden class variables at corresponding dyadic sub-squares. Then, saliency value for each dyadic square at each scale level is computed with discriminant power principle and the MAP. Finally, across multiple scales is integrated the final saliency map by an information maximization rule. Both standard quantitative tools such as NSS, LCC, AUC and qualitative assessments are used for evaluating the proposed multiscale discriminant saliency method (MDIS) against the well-know information-based saliency method AIM on its Bruce Database wity eye-tracking data. Simulation results are presented and analyzed to verify the validity of MDIS as well as point out its disadvantages for further research direction. {\circledC} 2013 Springer-Verlag Berlin Heidelberg.",
author = "{Le Ngo}, {Anh Cat} and Li-Minn Ang and Guoping Qiu and Seng, {Kah Phooi}",
year = "2013",
doi = "10.1007/978-3-642-39637-3_37",
language = "English",
isbn = "9783642396366",
volume = "7971",
pages = "464--484",
booktitle = "ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V",
publisher = "Springer",
address = "United States",

}

Le Ngo, AC, Ang, L-M, Qiu, G & Seng, KP 2013, Multiscale discriminant saliency for visual attention. in ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V. vol. 7971, Springer, Heidelberg, pp. 464-484, The 13th International Conference on Computational Science and Its Applications , Hi Chi Minh City, Viet Nam, 24/06/13. https://doi.org/10.1007/978-3-642-39637-3_37

Multiscale discriminant saliency for visual attention. / Le Ngo, Anh Cat; Ang, Li-Minn; Qiu, Guoping; Seng, Kah Phooi.

ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V. Vol. 7971 Heidelberg : Springer, 2013. p. 464-484.

Research output: Book chapter/Published conference paperConference paper

TY - GEN

T1 - Multiscale discriminant saliency for visual attention

AU - Le Ngo, Anh Cat

AU - Ang, Li-Minn

AU - Qiu, Guoping

AU - Seng, Kah Phooi

PY - 2013

Y1 - 2013

N2 - The bottom-up saliency, an early stage of humans' visual attention, can be considered as a binary classification problem between center and surround classes. Discriminant power of features for the classification is measured as mutual information between features and two classes distribution. The estimated discrepancy of two feature classes very much depends on considered scale levels; then, multi-scale structure and discriminant power are integrated by employing discrete wavelet features and Hidden markov tree (HMT). With wavelet coefficients and Hidden Markov Tree parameters, quad-tree like label structures are constructed and utilized in maximum a posterior probability (MAP) of hidden class variables at corresponding dyadic sub-squares. Then, saliency value for each dyadic square at each scale level is computed with discriminant power principle and the MAP. Finally, across multiple scales is integrated the final saliency map by an information maximization rule. Both standard quantitative tools such as NSS, LCC, AUC and qualitative assessments are used for evaluating the proposed multiscale discriminant saliency method (MDIS) against the well-know information-based saliency method AIM on its Bruce Database wity eye-tracking data. Simulation results are presented and analyzed to verify the validity of MDIS as well as point out its disadvantages for further research direction. © 2013 Springer-Verlag Berlin Heidelberg.

AB - The bottom-up saliency, an early stage of humans' visual attention, can be considered as a binary classification problem between center and surround classes. Discriminant power of features for the classification is measured as mutual information between features and two classes distribution. The estimated discrepancy of two feature classes very much depends on considered scale levels; then, multi-scale structure and discriminant power are integrated by employing discrete wavelet features and Hidden markov tree (HMT). With wavelet coefficients and Hidden Markov Tree parameters, quad-tree like label structures are constructed and utilized in maximum a posterior probability (MAP) of hidden class variables at corresponding dyadic sub-squares. Then, saliency value for each dyadic square at each scale level is computed with discriminant power principle and the MAP. Finally, across multiple scales is integrated the final saliency map by an information maximization rule. Both standard quantitative tools such as NSS, LCC, AUC and qualitative assessments are used for evaluating the proposed multiscale discriminant saliency method (MDIS) against the well-know information-based saliency method AIM on its Bruce Database wity eye-tracking data. Simulation results are presented and analyzed to verify the validity of MDIS as well as point out its disadvantages for further research direction. © 2013 Springer-Verlag Berlin Heidelberg.

U2 - 10.1007/978-3-642-39637-3_37

DO - 10.1007/978-3-642-39637-3_37

M3 - Conference paper

SN - 9783642396366

VL - 7971

SP - 464

EP - 484

BT - ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V

PB - Springer

CY - Heidelberg

ER -

Le Ngo AC, Ang L-M, Qiu G, Seng KP. Multiscale discriminant saliency for visual attention. In ICCSA 2013: Computational Science and Its Applications – ICCSA 2013 Proceedings Part V. Vol. 7971. Heidelberg: Springer. 2013. p. 464-484 https://doi.org/10.1007/978-3-642-39637-3_37