The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
La tâche d'annotation d'images devient extrêmement importante pour une récupération efficace d'images à partir du Web et d'autres grandes bases de données. Cependant, l’énorme quantité d’informations sémantiques et la dépendance complexe des étiquettes sur une image rendent la tâche difficile. Par conséquent, déterminer la similarité sémantique entre plusieurs étiquettes sur une image est utile pour comprendre toute attribution d’étiquette incomplète pour la récupération d’image. Ce travail propose une nouvelle méthode pour résoudre le problème de l'annotation d'images multi-étiquettes en unifiant deux types différents de termes de régularisation laplaciens dans un réseau neuronal convolutif profond (CNN) pour des performances d'annotation robustes. Le modèle de régularisation laplacien unifié est implémenté pour traiter efficacement les étiquettes manquantes en générant la similarité contextuelle entre les étiquettes à la fois en interne et en externe grâce à leurs similitudes sémantiques, ce qui constitue la principale contribution de cette étude. Plus précisément, nous générons des matrices de similarité entre les étiquettes en interne en utilisant la méthode de quantification de type III de Hayashi et en externe en utilisant la méthode word2vec. Les matrices de similarité générées par les deux méthodes différentes sont ensuite combinées sous la forme d'un terme de régularisation laplacien, qui est utilisé comme nouvelle fonction objectif du CNN profond. Le terme de régularisation mis en œuvre dans cette étude est capable de résoudre le problème de l'annotation multi-étiquettes, permettant un réseau neuronal formé plus efficacement. Les résultats expérimentaux sur des ensembles de données de référence publics révèlent que le modèle de régularisation unifié proposé avec CNN profond produit des résultats nettement meilleurs que le CNN de base sans régularisation ni autres méthodes de pointe pour prédire les étiquettes manquantes.
Jonathan MOJOO
Hiroshima University
Yu ZHAO
Hiroshima University
Muthu Subash KAVITHA
Hiroshima University
Junichi MIYAO
Hiroshima University
Takio KURITA
Hiroshima University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Jonathan MOJOO, Yu ZHAO, Muthu Subash KAVITHA, Junichi MIYAO, Takio KURITA, "Completion of Missing Labels for Multi-Label Annotation by a Unified Graph Laplacian Regularization" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 10, pp. 2154-2161, October 2020, doi: 10.1587/transinf.2019EDP7318.
Abstract: The task of image annotation is becoming enormously important for efficient image retrieval from the web and other large databases. However, huge semantic information and complex dependency of labels on an image make the task challenging. Hence determining the semantic similarity between multiple labels on an image is useful to understand any incomplete label assignment for image retrieval. This work proposes a novel method to solve the problem of multi-label image annotation by unifying two different types of Laplacian regularization terms in deep convolutional neural network (CNN) for robust annotation performance. The unified Laplacian regularization model is implemented to address the missing labels efficiently by generating the contextual similarity between labels both internally and externally through their semantic similarities, which is the main contribution of this study. Specifically, we generate similarity matrices between labels internally by using Hayashi's quantification method-type III and externally by using the word2vec method. The generated similarity matrices from the two different methods are then combined as a Laplacian regularization term, which is used as the new objective function of the deep CNN. The Regularization term implemented in this study is able to address the multi-label annotation problem, enabling a more effectively trained neural network. Experimental results on public benchmark datasets reveal that the proposed unified regularization model with deep CNN produces significantly better results than the baseline CNN without regularization and other state-of-the-art methods for predicting missing labels.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDP7318/_p
Copier
@ARTICLE{e103-d_10_2154,
author={Jonathan MOJOO, Yu ZHAO, Muthu Subash KAVITHA, Junichi MIYAO, Takio KURITA, },
journal={IEICE TRANSACTIONS on Information},
title={Completion of Missing Labels for Multi-Label Annotation by a Unified Graph Laplacian Regularization},
year={2020},
volume={E103-D},
number={10},
pages={2154-2161},
abstract={The task of image annotation is becoming enormously important for efficient image retrieval from the web and other large databases. However, huge semantic information and complex dependency of labels on an image make the task challenging. Hence determining the semantic similarity between multiple labels on an image is useful to understand any incomplete label assignment for image retrieval. This work proposes a novel method to solve the problem of multi-label image annotation by unifying two different types of Laplacian regularization terms in deep convolutional neural network (CNN) for robust annotation performance. The unified Laplacian regularization model is implemented to address the missing labels efficiently by generating the contextual similarity between labels both internally and externally through their semantic similarities, which is the main contribution of this study. Specifically, we generate similarity matrices between labels internally by using Hayashi's quantification method-type III and externally by using the word2vec method. The generated similarity matrices from the two different methods are then combined as a Laplacian regularization term, which is used as the new objective function of the deep CNN. The Regularization term implemented in this study is able to address the multi-label annotation problem, enabling a more effectively trained neural network. Experimental results on public benchmark datasets reveal that the proposed unified regularization model with deep CNN produces significantly better results than the baseline CNN without regularization and other state-of-the-art methods for predicting missing labels.},
keywords={},
doi={10.1587/transinf.2019EDP7318},
ISSN={1745-1361},
month={October},}
Copier
TY - JOUR
TI - Completion of Missing Labels for Multi-Label Annotation by a Unified Graph Laplacian Regularization
T2 - IEICE TRANSACTIONS on Information
SP - 2154
EP - 2161
AU - Jonathan MOJOO
AU - Yu ZHAO
AU - Muthu Subash KAVITHA
AU - Junichi MIYAO
AU - Takio KURITA
PY - 2020
DO - 10.1587/transinf.2019EDP7318
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 10
JA - IEICE TRANSACTIONS on Information
Y1 - October 2020
AB - The task of image annotation is becoming enormously important for efficient image retrieval from the web and other large databases. However, huge semantic information and complex dependency of labels on an image make the task challenging. Hence determining the semantic similarity between multiple labels on an image is useful to understand any incomplete label assignment for image retrieval. This work proposes a novel method to solve the problem of multi-label image annotation by unifying two different types of Laplacian regularization terms in deep convolutional neural network (CNN) for robust annotation performance. The unified Laplacian regularization model is implemented to address the missing labels efficiently by generating the contextual similarity between labels both internally and externally through their semantic similarities, which is the main contribution of this study. Specifically, we generate similarity matrices between labels internally by using Hayashi's quantification method-type III and externally by using the word2vec method. The generated similarity matrices from the two different methods are then combined as a Laplacian regularization term, which is used as the new objective function of the deep CNN. The Regularization term implemented in this study is able to address the multi-label annotation problem, enabling a more effectively trained neural network. Experimental results on public benchmark datasets reveal that the proposed unified regularization model with deep CNN produces significantly better results than the baseline CNN without regularization and other state-of-the-art methods for predicting missing labels.
ER -