Blind image quality assessment via deep recursive convolutional network with skip connection

Qingsen Yan, Jinqiu Sun, Shaolin Su, Yu Zhu, Haisen Li, Yanning Zhang

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

1 Scopus citations

Abstract

The performance of traditional image quality assessment (IQA) methods are not robust, due to those methods exploit shallow hand-designed features. It has been demonstrated that deep neural network can learn more effective features compared with the traditional methods. In this paper we propose a multi-scale recursive deep neural network to accurately predict image quality. In order to learn more effective feature representations for IQA, many deep learning based works focus on using more layers and deeper network structure. However, deeper network layers introduce large numbers of parameters, which causes huge difficulty in training. The proposed recursive convolution layer ensures both the depth of the network and the light of parameters, which guarantees the convergence of training procedure. Moreover, extracting multi-scale features is the most prevalent approach in IQA. Based on this criteria, we using skip connection to combine information among layers, and it further enriches the coarse and fine features for quality assessment. The experimental results on the LIVE, CISQ and TID2013 databases show that the proposed algorithm outperforms all of the state-of-the-art methods, which verifies the effectiveness of our network architecture.

Original languageEnglish
Title of host publicationPattern Recognition and Computer Vision - First Chinese Conference, PRCV 2018, Proceedings
EditorsCheng-Lin Liu, Tieniu Tan, Jie Zhou, Jian-Huang Lai, Xilin Chen, Nanning Zheng, Hongbin Zha
PublisherSpringer Verlag
Pages51-61
Number of pages11
ISBN (Print)9783030033347
DOIs
StatePublished - 2018
Event1st Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2018 - Guangzhou, China
Duration: 23 Nov 201826 Nov 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11257 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference1st Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2018
Country/TerritoryChina
CityGuangzhou
Period23/11/1826/11/18

Keywords

  • Convolutional neural networks (CNN)
  • Deep learning
  • Feature extraction
  • Image quality assessment (IQA)
  • No-reference (NR)
  • Skip layer

Fingerprint

Dive into the research topics of 'Blind image quality assessment via deep recursive convolutional network with skip connection'. Together they form a unique fingerprint.

Cite this