Uneven illumination removal based on fully convolutional network for dermoscopy images

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

Abstract

For the dermoscopy image, uneven illumination will influence segmentation accuracy and lead to wrong aided diagnosis result. In this paper, an uneven illumination removal method based on deep learning is proposed for dermoscopy images. Different from the traditional Retinex based methods, which estimate the illumination component using statistical methods to obtain the reflectance component of the image (uneven illumination removal result), in this paper, the illumination component is regarded as a black box to be learned by a designed fully convolutional neural network(FCN) model. The designed FCN model has more scales and can mine more effective features to obtain good illumination correction results. Experiment results show that, compared with 7 other state-of-art algorithms, our method can remove uneven illumination more effectively, and with our method, the segmentation performance is improved greatly.

Original languageEnglish
Title of host publication2016 13th International Computer Conference on Wavelet Active Media Technology and Information Processing, ICCWAMTIP 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages243-247
Number of pages5
ISBN (Electronic)9781509061259
DOIs
StatePublished - 20 Oct 2017
Event13th International Computer Conference on Wavelet Active Media Technology and Information Processing, ICCWAMTIP 2017 - Chengdu, China
Duration: 16 Dec 201618 Dec 2016

Publication series

Name2016 13th International Computer Conference on Wavelet Active Media Technology and Information Processing, ICCWAMTIP 2017

Conference

Conference13th International Computer Conference on Wavelet Active Media Technology and Information Processing, ICCWAMTIP 2017
Country/TerritoryChina
CityChengdu
Period16/12/1618/12/16

Keywords

  • deep learning
  • dermoscopy images
  • fully convolutional network
  • Retinex
  • Uneven illumination

Fingerprint

Dive into the research topics of 'Uneven illumination removal based on fully convolutional network for dermoscopy images'. Together they form a unique fingerprint.

Cite this