Color Constancy Convolutional Autoencoder

Firas Laakom; Jenni Raitoharju; Alexandros Iosifidis; Jarno Nikkanen; Moncef Gabbouj

doi:10.1109/SSCI44817.2019.9002684

Color Constancy Convolutional Autoencoder

Firas Laakom, Jenni Raitoharju, Alexandros Iosifidis, Jarno Nikkanen, Moncef Gabbouj

Computing Sciences

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

5 Citations (Scopus)

Abstract

In this paper, we study the importance of pretraining for the generalization capability in the color constancy problem. We propose two novel approaches based on convolutional autoencoders: an unsupervised pre-training algorithm using a fine-tuned encoder and a semi-supervised pre-training algorithm using a novel composite-loss function. This enables us to solve the data scarcity problem and achieve competitive, to the state-of-the-art, results while requiring much fewer parameters on ColorChecker RECommended dataset. We further study the over-fitting phenomenon on the recently introduced version of INTEL-TUT Dataset for Camera Invariant Color Constancy Research, which has both field and non-field scenes acquired by three different camera models.

Original language	English
Title of host publication	2019 IEEE Symposium Series on Computational Intelligence, SSCI 2019
Publisher	IEEE
Pages	1085-1090
Number of pages	6
ISBN (Electronic)	9781728124858
ISBN (Print)	978-1-7281-2486-5
DOIs	https://doi.org/10.1109/SSCI44817.2019.9002684
Publication status	Published - 2019
Publication type	A4 Article in conference proceedings
Event	IEEE Symposium Series on Computational Intelligence - Duration: 1 Jan 1900 → …

Conference

Conference	IEEE Symposium Series on Computational Intelligence
Abbreviated title	IEEE SSCI
Period	1/01/00 → …

Keywords

color constancy
convolutional autoencoders
illumination
pre-training

Publication forum classification

Publication forum level 1

ASJC Scopus subject areas

Artificial Intelligence
Computer Science Applications
Modelling and Simulation

Access to Document

10.1109/SSCI44817.2019.9002684

Cite this

@inproceedings{014286315c92497a993cfc825e1263ae,

title = "Color Constancy Convolutional Autoencoder",

abstract = "In this paper, we study the importance of pretraining for the generalization capability in the color constancy problem. We propose two novel approaches based on convolutional autoencoders: an unsupervised pre-training algorithm using a fine-tuned encoder and a semi-supervised pre-training algorithm using a novel composite-loss function. This enables us to solve the data scarcity problem and achieve competitive, to the state-of-the-art, results while requiring much fewer parameters on ColorChecker RECommended dataset. We further study the over-fitting phenomenon on the recently introduced version of INTEL-TUT Dataset for Camera Invariant Color Constancy Research, which has both field and non-field scenes acquired by three different camera models.",

keywords = "color constancy, convolutional autoencoders, illumination, pre-training",

author = "Firas Laakom and Jenni Raitoharju and Alexandros Iosifidis and Jarno Nikkanen and Moncef Gabbouj",

note = "EXT={"}Iosifidis, Alexandros{"}; IEEE Symposium Series on Computational Intelligence, IEEE SSCI ; Conference date: 01-01-1900",

year = "2019",

doi = "10.1109/SSCI44817.2019.9002684",

language = "English",

isbn = "978-1-7281-2486-5",

pages = "1085--1090",

booktitle = "2019 IEEE Symposium Series on Computational Intelligence, SSCI 2019",

publisher = "IEEE",

}

TY - GEN

T1 - Color Constancy Convolutional Autoencoder

AU - Laakom, Firas

AU - Raitoharju, Jenni

AU - Iosifidis, Alexandros

AU - Nikkanen, Jarno

AU - Gabbouj, Moncef

N1 - EXT="Iosifidis, Alexandros"

PY - 2019

Y1 - 2019

N2 - In this paper, we study the importance of pretraining for the generalization capability in the color constancy problem. We propose two novel approaches based on convolutional autoencoders: an unsupervised pre-training algorithm using a fine-tuned encoder and a semi-supervised pre-training algorithm using a novel composite-loss function. This enables us to solve the data scarcity problem and achieve competitive, to the state-of-the-art, results while requiring much fewer parameters on ColorChecker RECommended dataset. We further study the over-fitting phenomenon on the recently introduced version of INTEL-TUT Dataset for Camera Invariant Color Constancy Research, which has both field and non-field scenes acquired by three different camera models.

AB - In this paper, we study the importance of pretraining for the generalization capability in the color constancy problem. We propose two novel approaches based on convolutional autoencoders: an unsupervised pre-training algorithm using a fine-tuned encoder and a semi-supervised pre-training algorithm using a novel composite-loss function. This enables us to solve the data scarcity problem and achieve competitive, to the state-of-the-art, results while requiring much fewer parameters on ColorChecker RECommended dataset. We further study the over-fitting phenomenon on the recently introduced version of INTEL-TUT Dataset for Camera Invariant Color Constancy Research, which has both field and non-field scenes acquired by three different camera models.

KW - color constancy

KW - convolutional autoencoders

KW - illumination

KW - pre-training

U2 - 10.1109/SSCI44817.2019.9002684

DO - 10.1109/SSCI44817.2019.9002684

M3 - Conference contribution

SN - 978-1-7281-2486-5

SP - 1085

EP - 1090

BT - 2019 IEEE Symposium Series on Computational Intelligence, SSCI 2019

PB - IEEE

T2 - IEEE Symposium Series on Computational Intelligence

Y2 - 1 January 1900

ER -

Color Constancy Convolutional Autoencoder

Abstract

Conference

Keywords

Publication forum classification

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this