Rethinking the Rotation Invariance of Local Convolutional Features for Content-Based Image Retrieval

Longjiao ZHAO; Yu WANG; Jien KATO

doi:10.1587/transinf.2020EDP7017

Rethinking the Rotation Invariance of Local Convolutional Features for Content-Based Image Retrieval

Longjiao ZHAO, Yu WANG, Jien KATO

Full Text Views

0

Cite this

Summary :

Recently, local features computed using convolutional neural networks (CNNs) show good performance to image retrieval. The local convolutional features obtained by the CNNs (LC features) are designed to be translation invariant, however, they are inherently sensitive to rotation perturbations. This leads to miss-judgements in retrieval tasks. In this work, our objective is to enhance the robustness of LC features against image rotation. To do this, we conduct a thorough experimental evaluation of three candidate anti-rotation strategies (in-model data augmentation, in-model feature augmentation, and post-model feature augmentation), over two kinds of rotation attack (dataset attack and query attack). In the training procedure, we implement a data augmentation protocol and network augmentation method. In the test procedure, we develop a local transformed convolutional (LTC) feature extraction method, and evaluate it over different network configurations. We end up a series of good practices with steady quantitative supports, which lead to the best strategy for computing LC features with high rotation invariance in image retrieval.

Publication: IEICE TRANSACTIONS on Information Vol.E104-D No.1 pp.174-182

Publication Date: 2021/01/01

Publicized: 2020/10/14

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2020EDP7017

Type of Manuscript: PAPER

Category: Image Processing and Video Processing

Authors

Longjiao ZHAO
  Nagoya University
Yu WANG
  Ritsumeikan University
Jien KATO
  Ritsumeikan University

Keyword

image retrieval, in-model data augmentation, in-model feature augmentation, post-model feature augmentation, local convolutional features, and rotation invariance

Cite this

Copy

Longjiao ZHAO, Yu WANG, Jien KATO, "Rethinking the Rotation Invariance of Local Convolutional Features for Content-Based Image Retrieval" in IEICE TRANSACTIONS on Information, vol. E104-D, no. 1, pp. 174-182, January 2021, doi: 10.1587/transinf.2020EDP7017.
Abstract: Recently, local features computed using convolutional neural networks (CNNs) show good performance to image retrieval. The local convolutional features obtained by the CNNs (LC features) are designed to be translation invariant, however, they are inherently sensitive to rotation perturbations. This leads to miss-judgements in retrieval tasks. In this work, our objective is to enhance the robustness of LC features against image rotation. To do this, we conduct a thorough experimental evaluation of three candidate anti-rotation strategies (in-model data augmentation, in-model feature augmentation, and post-model feature augmentation), over two kinds of rotation attack (dataset attack and query attack). In the training procedure, we implement a data augmentation protocol and network augmentation method. In the test procedure, we develop a local transformed convolutional (LTC) feature extraction method, and evaluate it over different network configurations. We end up a series of good practices with steady quantitative supports, which lead to the best strategy for computing LC features with high rotation invariance in image retrieval.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2020EDP7017/_p

Copy

@ARTICLE{e104-d_1_174,
author={Longjiao ZHAO, Yu WANG, Jien KATO, },
journal={IEICE TRANSACTIONS on Information},
title={Rethinking the Rotation Invariance of Local Convolutional Features for Content-Based Image Retrieval},
year={2021},
volume={E104-D},
number={1},
pages={174-182},
abstract={Recently, local features computed using convolutional neural networks (CNNs) show good performance to image retrieval. The local convolutional features obtained by the CNNs (LC features) are designed to be translation invariant, however, they are inherently sensitive to rotation perturbations. This leads to miss-judgements in retrieval tasks. In this work, our objective is to enhance the robustness of LC features against image rotation. To do this, we conduct a thorough experimental evaluation of three candidate anti-rotation strategies (in-model data augmentation, in-model feature augmentation, and post-model feature augmentation), over two kinds of rotation attack (dataset attack and query attack). In the training procedure, we implement a data augmentation protocol and network augmentation method. In the test procedure, we develop a local transformed convolutional (LTC) feature extraction method, and evaluate it over different network configurations. We end up a series of good practices with steady quantitative supports, which lead to the best strategy for computing LC features with high rotation invariance in image retrieval.},
keywords={},
doi={10.1587/transinf.2020EDP7017},
ISSN={1745-1361},
month={January},}

Copy

TY - JOUR
TI - Rethinking the Rotation Invariance of Local Convolutional Features for Content-Based Image Retrieval
T2 - IEICE TRANSACTIONS on Information
SP - 174
EP - 182
AU - Longjiao ZHAO
AU - Yu WANG
AU - Jien KATO
PY - 2021
DO - 10.1587/transinf.2020EDP7017
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 1
JA - IEICE TRANSACTIONS on Information
Y1 - January 2021
AB - Recently, local features computed using convolutional neural networks (CNNs) show good performance to image retrieval. The local convolutional features obtained by the CNNs (LC features) are designed to be translation invariant, however, they are inherently sensitive to rotation perturbations. This leads to miss-judgements in retrieval tasks. In this work, our objective is to enhance the robustness of LC features against image rotation. To do this, we conduct a thorough experimental evaluation of three candidate anti-rotation strategies (in-model data augmentation, in-model feature augmentation, and post-model feature augmentation), over two kinds of rotation attack (dataset attack and query attack). In the training procedure, we implement a data augmentation protocol and network augmentation method. In the test procedure, we develop a local transformed convolutional (LTC) feature extraction method, and evaluate it over different network configurations. We end up a series of good practices with steady quantitative supports, which lead to the best strategy for computing LC features with high rotation invariance in image retrieval.
ER -