Learning Non-Metric Visual Similarity for Image Retrieval

Noa Garcia, George Vogiatzis

Research output: Contribution to journalArticle

Abstract

Measuring visual similarity between two or more instances within a data distribution is a fundamental task in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the system. In this work, we explore neural networks models for learning a non-metric similarity function for instance search. We argue that non-metric similarity functions based on neural networks can build a better model of human visual perception than standard metric distances. As our proposed similarity function is differentiable, we explore a real end-to-end trainable approach for image retrieval, i.e. we learn the weights from the input image pixels to the final similarity score. Experimental evaluation shows that non-metric similarity networks are able to learn visual similarities between images and improve performance on top of state-of-the-art image representations, boosting results in standard image retrieval datasets with respect standard metric distances.
Original languageEnglish
Pages (from-to)18-25
Number of pages8
JournalImage and Vision Computing
Volume82
DOIs
Publication statusPublished - 10 Feb 2019

Fingerprint

Image retrieval
Neural networks
Pixels

Bibliographical note

© 2019, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

Keywords

  • Image retrieval
  • Non-metric learning
  • Visual similarity

Cite this

@article{ea97a2ad580043d7a5e8e89dabc4d5e8,
title = "Learning Non-Metric Visual Similarity for Image Retrieval",
abstract = "Measuring visual similarity between two or more instances within a data distribution is a fundamental task in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the system. In this work, we explore neural networks models for learning a non-metric similarity function for instance search. We argue that non-metric similarity functions based on neural networks can build a better model of human visual perception than standard metric distances. As our proposed similarity function is differentiable, we explore a real end-to-end trainable approach for image retrieval, i.e. we learn the weights from the input image pixels to the final similarity score. Experimental evaluation shows that non-metric similarity networks are able to learn visual similarities between images and improve performance on top of state-of-the-art image representations, boosting results in standard image retrieval datasets with respect standard metric distances.",
keywords = "Image retrieval, Non-metric learning, Visual similarity",
author = "Noa Garcia and George Vogiatzis",
note = "{\circledC} 2019, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/",
year = "2019",
month = "2",
day = "10",
doi = "10.1016/j.imavis.2019.01.001",
language = "English",
volume = "82",
pages = "18--25",
journal = "Image and Vision Computing",
issn = "0262-8856",
publisher = "Elsevier",

}

Learning Non-Metric Visual Similarity for Image Retrieval. / Garcia, Noa; Vogiatzis, George.

In: Image and Vision Computing, Vol. 82, 10.02.2019, p. 18-25.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Learning Non-Metric Visual Similarity for Image Retrieval

AU - Garcia, Noa

AU - Vogiatzis, George

N1 - © 2019, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

PY - 2019/2/10

Y1 - 2019/2/10

N2 - Measuring visual similarity between two or more instances within a data distribution is a fundamental task in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the system. In this work, we explore neural networks models for learning a non-metric similarity function for instance search. We argue that non-metric similarity functions based on neural networks can build a better model of human visual perception than standard metric distances. As our proposed similarity function is differentiable, we explore a real end-to-end trainable approach for image retrieval, i.e. we learn the weights from the input image pixels to the final similarity score. Experimental evaluation shows that non-metric similarity networks are able to learn visual similarities between images and improve performance on top of state-of-the-art image representations, boosting results in standard image retrieval datasets with respect standard metric distances.

AB - Measuring visual similarity between two or more instances within a data distribution is a fundamental task in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the system. In this work, we explore neural networks models for learning a non-metric similarity function for instance search. We argue that non-metric similarity functions based on neural networks can build a better model of human visual perception than standard metric distances. As our proposed similarity function is differentiable, we explore a real end-to-end trainable approach for image retrieval, i.e. we learn the weights from the input image pixels to the final similarity score. Experimental evaluation shows that non-metric similarity networks are able to learn visual similarities between images and improve performance on top of state-of-the-art image representations, boosting results in standard image retrieval datasets with respect standard metric distances.

KW - Image retrieval

KW - Non-metric learning

KW - Visual similarity

UR - https://linkinghub.elsevier.com/retrieve/pii/S0262885619300071

UR - http://www.scopus.com/inward/record.url?scp=85061662074&partnerID=8YFLogxK

U2 - 10.1016/j.imavis.2019.01.001

DO - 10.1016/j.imavis.2019.01.001

M3 - Article

VL - 82

SP - 18

EP - 25

JO - Image and Vision Computing

JF - Image and Vision Computing

SN - 0262-8856

ER -