Tampere University of Technology

TUTCRIS Research Portal

Color Constancy Convolutional Autoencoder

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

Details

Original languageEnglish
Title of host publication2019 IEEE Symposium Series on Computational Intelligence, SSCI 2019
PublisherIEEE
Pages1085-1090
Number of pages6
ISBN (Electronic)9781728124858
ISBN (Print)978-1-7281-2486-5
DOIs
Publication statusPublished - 2019
Publication typeA4 Article in a conference publication
EventIEEE Symposium Series on Computational Intelligence -
Duration: 1 Jan 1900 → …

Conference

ConferenceIEEE Symposium Series on Computational Intelligence
Abbreviated titleIEEE SSCI
Period1/01/00 → …

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

Publication forum classification

Field of science, Statistics Finland