Understanding how image quality affects deep neural networks

Samuel Dodge, Lina Karam

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

72 Citations (Scopus)

Abstract

Image quality is an important practical challenge that is often overlooked in the design of machine vision systems. Commonly, machine vision systems are trained and tested on high quality image datasets, yet in practical applications the input images can not be assumed to be of high quality. Recently, deep neural networks have obtained state-of-the-art performance on many machine vision tasks. In this paper we provide an evaluation of 4 state-of-the-art deep neural network models for image classification under quality distortions. We consider five types of quality distortions: blur, noise, contrast, JPEG, and JPEG2000 compression. We show that the existing networks are susceptible to these quality distortions, particularly to blur and noise. These results enable future work in developing deep neural networks that are more invariant to quality distortions.

Original languageEnglish (US)
Title of host publication2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781509003549
DOIs
StatePublished - Jun 23 2016
Event8th International Conference on Quality of Multimedia Experience, QoMEX 2016 - Lisbon, Portugal
Duration: Jun 6 2016Jun 8 2016

Other

Other8th International Conference on Quality of Multimedia Experience, QoMEX 2016
CountryPortugal
CityLisbon
Period6/6/166/8/16

Fingerprint

neural network
Image quality
Computer vision
Image classification
Deep neural networks
evaluation
performance

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Media Technology
  • Communication

Cite this

Dodge, S., & Karam, L. (2016). Understanding how image quality affects deep neural networks. In 2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016 [7498955] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/QoMEX.2016.7498955

Understanding how image quality affects deep neural networks. / Dodge, Samuel; Karam, Lina.

2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016. Institute of Electrical and Electronics Engineers Inc., 2016. 7498955.

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

Dodge, S & Karam, L 2016, Understanding how image quality affects deep neural networks. in 2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016., 7498955, Institute of Electrical and Electronics Engineers Inc., 8th International Conference on Quality of Multimedia Experience, QoMEX 2016, Lisbon, Portugal, 6/6/16. https://doi.org/10.1109/QoMEX.2016.7498955
Dodge S, Karam L. Understanding how image quality affects deep neural networks. In 2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016. Institute of Electrical and Electronics Engineers Inc. 2016. 7498955 https://doi.org/10.1109/QoMEX.2016.7498955
Dodge, Samuel ; Karam, Lina. / Understanding how image quality affects deep neural networks. 2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016. Institute of Electrical and Electronics Engineers Inc., 2016.
@inproceedings{1746a6f3349c4edd97f8a43f2a5c0205,
title = "Understanding how image quality affects deep neural networks",
abstract = "Image quality is an important practical challenge that is often overlooked in the design of machine vision systems. Commonly, machine vision systems are trained and tested on high quality image datasets, yet in practical applications the input images can not be assumed to be of high quality. Recently, deep neural networks have obtained state-of-the-art performance on many machine vision tasks. In this paper we provide an evaluation of 4 state-of-the-art deep neural network models for image classification under quality distortions. We consider five types of quality distortions: blur, noise, contrast, JPEG, and JPEG2000 compression. We show that the existing networks are susceptible to these quality distortions, particularly to blur and noise. These results enable future work in developing deep neural networks that are more invariant to quality distortions.",
author = "Samuel Dodge and Lina Karam",
year = "2016",
month = "6",
day = "23",
doi = "10.1109/QoMEX.2016.7498955",
language = "English (US)",
booktitle = "2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - Understanding how image quality affects deep neural networks

AU - Dodge, Samuel

AU - Karam, Lina

PY - 2016/6/23

Y1 - 2016/6/23

N2 - Image quality is an important practical challenge that is often overlooked in the design of machine vision systems. Commonly, machine vision systems are trained and tested on high quality image datasets, yet in practical applications the input images can not be assumed to be of high quality. Recently, deep neural networks have obtained state-of-the-art performance on many machine vision tasks. In this paper we provide an evaluation of 4 state-of-the-art deep neural network models for image classification under quality distortions. We consider five types of quality distortions: blur, noise, contrast, JPEG, and JPEG2000 compression. We show that the existing networks are susceptible to these quality distortions, particularly to blur and noise. These results enable future work in developing deep neural networks that are more invariant to quality distortions.

AB - Image quality is an important practical challenge that is often overlooked in the design of machine vision systems. Commonly, machine vision systems are trained and tested on high quality image datasets, yet in practical applications the input images can not be assumed to be of high quality. Recently, deep neural networks have obtained state-of-the-art performance on many machine vision tasks. In this paper we provide an evaluation of 4 state-of-the-art deep neural network models for image classification under quality distortions. We consider five types of quality distortions: blur, noise, contrast, JPEG, and JPEG2000 compression. We show that the existing networks are susceptible to these quality distortions, particularly to blur and noise. These results enable future work in developing deep neural networks that are more invariant to quality distortions.

UR - http://www.scopus.com/inward/record.url?scp=84979701671&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84979701671&partnerID=8YFLogxK

U2 - 10.1109/QoMEX.2016.7498955

DO - 10.1109/QoMEX.2016.7498955

M3 - Conference contribution

BT - 2016 8th International Conference on Quality of Multimedia Experience, QoMEX 2016

PB - Institute of Electrical and Electronics Engineers Inc.

ER -