Combining multi-level loss for image denoising

Fei Li, Nailiang Kuang, Jiangbin Zheng, Qianru Wei, Yue Xi, Yanrong Guo

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

Abstract

The image processing has witnessed remarkable progress in image denoising. Nevertheless, restoring the visual quality of the image remains a great challenge. Existing methods might fail to obtain the denoised images with high visual quality since they ignore the potential connection with the high-level feature and result in over-smoothing results. Aiming to research whether high-level feature could influence the performance of denoising task, we propose an end-to-end multi-module neural network architecture, which introduces the combination of the high-level and low-level feature in the training process, for image denoising. In order to guide model preserve structural information efficiently, we introduce a hybrid loss, which is designed to restore details from both pixel and feature space. The experimental results show our method improves the visual quality of images and performs well compared with state-of-the-art methods on three benchmarks.

Original languageEnglish
Title of host publicationAdvances in Brain Inspired Cognitive Systems - 10th International Conference, BICS 2019, Proceedings
EditorsJinchang Ren, Amir Hussain, Huimin Zhao, Jun Cai, Rongjun Chen, Yinyin Xiao, Kaizhu Huang, Jiangbin Zheng
PublisherSpringer
Pages423-432
Number of pages10
ISBN (Print)9783030394301
DOIs
StatePublished - 2020
Event10th International Conference on Brain Inspired Cognitive Systems, BICS 2019 - Guangzhou, China
Duration: 13 Jul 201914 Jul 2019

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11691 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference10th International Conference on Brain Inspired Cognitive Systems, BICS 2019
Country/TerritoryChina
CityGuangzhou
Period13/07/1914/07/19

Keywords

  • Edge loss
  • Hybrid loss function
  • Image denoising

Fingerprint

Dive into the research topics of 'Combining multi-level loss for image denoising'. Together they form a unique fingerprint.

Cite this