A Hybrid Strategy for Illuminant Estimation Targeting Hard Images

Zakizadeh, Roshanak, Brown, Michael and Finlayson, Graham (2015) A Hybrid Strategy for Illuminant Estimation Targeting Hard Images. In: Proceedings of the IEEE International Conference on Computer Vision Workshops. IEEE Conference Publications, pp. 49-56. ISBN 978-1-4673-8390-5

[img]
Preview
PDF (Zakizadeh_A_Hybrid_Strategy_ICCV_2015_paper) - Published Version
Download (1MB) | Preview

Abstract

Illumination estimation is a well-studied topic in computer vision. Early work reported performance on benchmark datasets using simple statistical aggregates such as mean or median error. Recently, it has become accepted to report a wider range of statistics, e.g. top 25%, mean, and bottom 25% performance. While these additional statistics are more informative, their relationship across different methods is unclear. In this paper, we analyse the results of a number of methods to see if there exist ‘hard’ images that are challenging for multiple methods. Our findings indicate that there are certain images that are difficult for fast statistical-based methods, but that can be handled with more complex learning-based approaches at a significant cost in time-complexity. This has led us to design a hybrid method that first classifies an image as ‘hard’ or ‘easy’ and then uses the slower method when needed, thus providing a balance between time-complexity and performance. In addition, we have identified dataset images that almost no method is able to process. We argue, however, that these images have problems with how the ground truth is established and recommend their removal from future performance evaluation.

Item Type: Book Section
Additional Information: ISSN: 1550-5499
Faculty \ School: Faculty of Science > School of Computing Sciences
Depositing User: Pure Connector
Date Deposited: 01 Jun 2016 11:02
Last Modified: 22 Jul 2020 03:23
URI: https://ueaeprints.uea.ac.uk/id/eprint/59160
DOI: 10.1109/ICCVW.2015.16

Actions (login required)

View Item View Item