Deep Learning based Super-Resolution for Improved Action Recognition

Kamal Nasrollahi, Sergio Escalera Guerrero, Pejman Rasti, Gholamreza Anbarjafari, Xavier Baro, Hugo J. Escalante, Thomas B. Moeslund

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

17 Citations (Scopus)
1001 Downloads (Pure)

Abstract

Action recognition systems mostly work with videos of proper quality and resolution. Even most challenging bench- mark databases for action recognition, hardly include videos of low-resolution from, e.g., surveillance cameras. In videos recorded by such cameras, due to the distance between people and cameras, people are pictured very small and hence challenge action recognition algorithms. Simple upsampling methods, like bicubic interpolation, cannot retrieve all the detailed information that can help the recognition. To deal with this problem, in this paper we combine results of bicubic interpolation with results of a state-of- the-art deep learning-based super-resolution algorithm, through an alpha-blending approach. The experimental results obtained on down-sampled version of a large subset of Hoolywood2 benchmark database show the importance of the proposed system in increasing the recognition rate of a state-of-the-art action recognition system for handling low-resolution videos.
Original languageEnglish
Title of host publicationInternational Conference on Image Processing Theory, Tools and Applications (IPTA), 2015
Number of pages6
PublisherIEEE Signal Processing Society
Publication date2015
Pages67 - 72
ISBN (Print)978-1-4799-8636-1, 978-1-4799-8637-8
DOIs
Publication statusPublished - 2015
EventIEEE International Conference on Image Processing Theory, Tools and Applications - Orleans, France
Duration: 10 Nov 201513 Nov 2015
Conference number: 5th

Conference

ConferenceIEEE International Conference on Image Processing Theory, Tools and Applications
Number5th
CountryFrance
CityOrleans
Period10/11/201513/11/2015

Fingerprint

Cameras
Interpolation
Deep learning

Cite this

Nasrollahi, K., Guerrero, S. E., Rasti, P., Anbarjafari, G., Baro, X., J. Escalante, H., & Moeslund, T. B. (2015). Deep Learning based Super-Resolution for Improved Action Recognition. In International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015 (pp. 67 - 72). IEEE Signal Processing Society. https://doi.org/10.1109/IPTA.2015.7367098
Nasrollahi, Kamal ; Guerrero, Sergio Escalera ; Rasti, Pejman ; Anbarjafari, Gholamreza ; Baro, Xavier ; J. Escalante, Hugo ; Moeslund, Thomas B. / Deep Learning based Super-Resolution for Improved Action Recognition. International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015. IEEE Signal Processing Society, 2015. pp. 67 - 72
@inproceedings{fee52140d2e54962bd125a5db6581f34,
title = "Deep Learning based Super-Resolution for Improved Action Recognition",
abstract = "Action recognition systems mostly work with videos of proper quality and resolution. Even most challenging bench- mark databases for action recognition, hardly include videos of low-resolution from, e.g., surveillance cameras. In videos recorded by such cameras, due to the distance between people and cameras, people are pictured very small and hence challenge action recognition algorithms. Simple upsampling methods, like bicubic interpolation, cannot retrieve all the detailed information that can help the recognition. To deal with this problem, in this paper we combine results of bicubic interpolation with results of a state-of- the-art deep learning-based super-resolution algorithm, through an alpha-blending approach. The experimental results obtained on down-sampled version of a large subset of Hoolywood2 benchmark database show the importance of the proposed system in increasing the recognition rate of a state-of-the-art action recognition system for handling low-resolution videos.",
author = "Kamal Nasrollahi and Guerrero, {Sergio Escalera} and Pejman Rasti and Gholamreza Anbarjafari and Xavier Baro and {J. Escalante}, Hugo and Moeslund, {Thomas B.}",
year = "2015",
doi = "10.1109/IPTA.2015.7367098",
language = "English",
isbn = "978-1-4799-8636-1",
pages = "67 -- 72",
booktitle = "International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015",
publisher = "IEEE Signal Processing Society",
address = "United States",

}

Nasrollahi, K, Guerrero, SE, Rasti, P, Anbarjafari, G, Baro, X, J. Escalante, H & Moeslund, TB 2015, Deep Learning based Super-Resolution for Improved Action Recognition. in International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015. IEEE Signal Processing Society, pp. 67 - 72, IEEE International Conference on Image Processing Theory, Tools and Applications , Orleans, France, 10/11/2015. https://doi.org/10.1109/IPTA.2015.7367098

Deep Learning based Super-Resolution for Improved Action Recognition. / Nasrollahi, Kamal; Guerrero, Sergio Escalera; Rasti, Pejman; Anbarjafari, Gholamreza; Baro, Xavier; J. Escalante, Hugo; Moeslund, Thomas B.

International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015. IEEE Signal Processing Society, 2015. p. 67 - 72.

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

TY - GEN

T1 - Deep Learning based Super-Resolution for Improved Action Recognition

AU - Nasrollahi, Kamal

AU - Guerrero, Sergio Escalera

AU - Rasti, Pejman

AU - Anbarjafari, Gholamreza

AU - Baro, Xavier

AU - J. Escalante, Hugo

AU - Moeslund, Thomas B.

PY - 2015

Y1 - 2015

N2 - Action recognition systems mostly work with videos of proper quality and resolution. Even most challenging bench- mark databases for action recognition, hardly include videos of low-resolution from, e.g., surveillance cameras. In videos recorded by such cameras, due to the distance between people and cameras, people are pictured very small and hence challenge action recognition algorithms. Simple upsampling methods, like bicubic interpolation, cannot retrieve all the detailed information that can help the recognition. To deal with this problem, in this paper we combine results of bicubic interpolation with results of a state-of- the-art deep learning-based super-resolution algorithm, through an alpha-blending approach. The experimental results obtained on down-sampled version of a large subset of Hoolywood2 benchmark database show the importance of the proposed system in increasing the recognition rate of a state-of-the-art action recognition system for handling low-resolution videos.

AB - Action recognition systems mostly work with videos of proper quality and resolution. Even most challenging bench- mark databases for action recognition, hardly include videos of low-resolution from, e.g., surveillance cameras. In videos recorded by such cameras, due to the distance between people and cameras, people are pictured very small and hence challenge action recognition algorithms. Simple upsampling methods, like bicubic interpolation, cannot retrieve all the detailed information that can help the recognition. To deal with this problem, in this paper we combine results of bicubic interpolation with results of a state-of- the-art deep learning-based super-resolution algorithm, through an alpha-blending approach. The experimental results obtained on down-sampled version of a large subset of Hoolywood2 benchmark database show the importance of the proposed system in increasing the recognition rate of a state-of-the-art action recognition system for handling low-resolution videos.

U2 - 10.1109/IPTA.2015.7367098

DO - 10.1109/IPTA.2015.7367098

M3 - Article in proceeding

SN - 978-1-4799-8636-1

SN - 978-1-4799-8637-8

SP - 67

EP - 72

BT - International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015

PB - IEEE Signal Processing Society

ER -

Nasrollahi K, Guerrero SE, Rasti P, Anbarjafari G, Baro X, J. Escalante H et al. Deep Learning based Super-Resolution for Improved Action Recognition. In International Conference on Image Processing Theory, Tools and Applications (IPTA), 2015. IEEE Signal Processing Society. 2015. p. 67 - 72 https://doi.org/10.1109/IPTA.2015.7367098