Scene Categorization with Spectral Features

Salman H. Khan, Munawar Hayat, Fatih Porikli

Research output: A Conference proceeding or a Chapter in BookConference contribution

7 Citations (Scopus)

Abstract

Spectral signatures of natural scenes were earlier found to be distinctive for different scene types with varying spatial envelope properties such as openness, naturalness, ruggedness, and symmetry. Recently, such handcrafted features have been outclassed by deep learning based representations. This paper proposes a novel spectral description of convolution features, implemented efficiently as a unitary transformation within deep network architectures. To the best of our knowledge, this is the first attempt to use deep learning based spectral features explicitly for image classification task. We show that the spectral transformation decorrelates convolutional activations, which reduces co-adaptation between feature detections, thus acts as an effective regularizer. Our approach achieves significant improvements on three large-scale scene-centric datasets (MIT-67, SUN-397, and Places-205). Furthermore, we evaluated the proposed approach on the attribute detection task where its superior performance manifests its relevance to semantically meaningful characteristics of natural scenes.

Original languageEnglish
Title of host publicationProceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages5639-5649
Number of pages11
Volume2017-October
ISBN (Electronic)9781538610329
ISBN (Print)9781538610336
DOIs
Publication statusPublished - 22 Dec 2017
Event16th IEEE International Conference on Computer Vision, ICCV 2017 - Venice, Italy
Duration: 22 Oct 201729 Oct 2017

Publication series

NameProceedings of the IEEE International Conference on Computer Vision
Volume2017-October

Conference

Conference16th IEEE International Conference on Computer Vision, ICCV 2017
CountryItaly
CityVenice
Period22/10/1729/10/17

Fingerprint

Image classification
Network architecture
Convolution
Chemical activation
Deep learning

Cite this

Khan, S. H., Hayat, M., & Porikli, F. (2017). Scene Categorization with Spectral Features. In Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017 (Vol. 2017-October, pp. 5639-5649). [8237863] (Proceedings of the IEEE International Conference on Computer Vision; Vol. 2017-October). IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ICCV.2017.601
Khan, Salman H. ; Hayat, Munawar ; Porikli, Fatih. / Scene Categorization with Spectral Features. Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017. Vol. 2017-October IEEE, Institute of Electrical and Electronics Engineers, 2017. pp. 5639-5649 (Proceedings of the IEEE International Conference on Computer Vision).
@inproceedings{85a9c4af312e4886a86e56790e1d807f,
title = "Scene Categorization with Spectral Features",
abstract = "Spectral signatures of natural scenes were earlier found to be distinctive for different scene types with varying spatial envelope properties such as openness, naturalness, ruggedness, and symmetry. Recently, such handcrafted features have been outclassed by deep learning based representations. This paper proposes a novel spectral description of convolution features, implemented efficiently as a unitary transformation within deep network architectures. To the best of our knowledge, this is the first attempt to use deep learning based spectral features explicitly for image classification task. We show that the spectral transformation decorrelates convolutional activations, which reduces co-adaptation between feature detections, thus acts as an effective regularizer. Our approach achieves significant improvements on three large-scale scene-centric datasets (MIT-67, SUN-397, and Places-205). Furthermore, we evaluated the proposed approach on the attribute detection task where its superior performance manifests its relevance to semantically meaningful characteristics of natural scenes.",
keywords = "Spectral Features, Scene categorization, Deep learning",
author = "Khan, {Salman H.} and Munawar Hayat and Fatih Porikli",
year = "2017",
month = "12",
day = "22",
doi = "10.1109/ICCV.2017.601",
language = "English",
isbn = "9781538610336",
volume = "2017-October",
series = "Proceedings of the IEEE International Conference on Computer Vision",
publisher = "IEEE, Institute of Electrical and Electronics Engineers",
pages = "5639--5649",
booktitle = "Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017",
address = "United States",

}

Khan, SH, Hayat, M & Porikli, F 2017, Scene Categorization with Spectral Features. in Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017. vol. 2017-October, 8237863, Proceedings of the IEEE International Conference on Computer Vision, vol. 2017-October, IEEE, Institute of Electrical and Electronics Engineers, pp. 5639-5649, 16th IEEE International Conference on Computer Vision, ICCV 2017, Venice, Italy, 22/10/17. https://doi.org/10.1109/ICCV.2017.601

Scene Categorization with Spectral Features. / Khan, Salman H.; Hayat, Munawar; Porikli, Fatih.

Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017. Vol. 2017-October IEEE, Institute of Electrical and Electronics Engineers, 2017. p. 5639-5649 8237863 (Proceedings of the IEEE International Conference on Computer Vision; Vol. 2017-October).

Research output: A Conference proceeding or a Chapter in BookConference contribution

TY - GEN

T1 - Scene Categorization with Spectral Features

AU - Khan, Salman H.

AU - Hayat, Munawar

AU - Porikli, Fatih

PY - 2017/12/22

Y1 - 2017/12/22

N2 - Spectral signatures of natural scenes were earlier found to be distinctive for different scene types with varying spatial envelope properties such as openness, naturalness, ruggedness, and symmetry. Recently, such handcrafted features have been outclassed by deep learning based representations. This paper proposes a novel spectral description of convolution features, implemented efficiently as a unitary transformation within deep network architectures. To the best of our knowledge, this is the first attempt to use deep learning based spectral features explicitly for image classification task. We show that the spectral transformation decorrelates convolutional activations, which reduces co-adaptation between feature detections, thus acts as an effective regularizer. Our approach achieves significant improvements on three large-scale scene-centric datasets (MIT-67, SUN-397, and Places-205). Furthermore, we evaluated the proposed approach on the attribute detection task where its superior performance manifests its relevance to semantically meaningful characteristics of natural scenes.

AB - Spectral signatures of natural scenes were earlier found to be distinctive for different scene types with varying spatial envelope properties such as openness, naturalness, ruggedness, and symmetry. Recently, such handcrafted features have been outclassed by deep learning based representations. This paper proposes a novel spectral description of convolution features, implemented efficiently as a unitary transformation within deep network architectures. To the best of our knowledge, this is the first attempt to use deep learning based spectral features explicitly for image classification task. We show that the spectral transformation decorrelates convolutional activations, which reduces co-adaptation between feature detections, thus acts as an effective regularizer. Our approach achieves significant improvements on three large-scale scene-centric datasets (MIT-67, SUN-397, and Places-205). Furthermore, we evaluated the proposed approach on the attribute detection task where its superior performance manifests its relevance to semantically meaningful characteristics of natural scenes.

KW - Spectral Features

KW - Scene categorization

KW - Deep learning

UR - http://www.scopus.com/inward/record.url?scp=85041915133&partnerID=8YFLogxK

UR - http://www.mendeley.com/research/scene-categorization-spectral-features

U2 - 10.1109/ICCV.2017.601

DO - 10.1109/ICCV.2017.601

M3 - Conference contribution

SN - 9781538610336

VL - 2017-October

T3 - Proceedings of the IEEE International Conference on Computer Vision

SP - 5639

EP - 5649

BT - Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017

PB - IEEE, Institute of Electrical and Electronics Engineers

ER -

Khan SH, Hayat M, Porikli F. Scene Categorization with Spectral Features. In Proceedings - 2017 IEEE International Conference on Computer Vision, ICCV 2017. Vol. 2017-October. IEEE, Institute of Electrical and Electronics Engineers. 2017. p. 5639-5649. 8237863. (Proceedings of the IEEE International Conference on Computer Vision). https://doi.org/10.1109/ICCV.2017.601