TY - CHAP

T1 - A Comparative Study for Fisheye Image Classification

T2 - 12th International Conference on Soft Computing and Pattern Recognition, SoCPaR 2020 and 16th International Conference on Information Assurance and Security, IAS 2020

AU - Chen, Zhen

AU - Georgiadis, Anthimos

N1 - Conference code: 12 & 16

PY - 2021/1/1

Y1 - 2021/1/1

N2 - The comparison between the feature-based method and the learning-based method is conducted in the training time, the accuracy and the generalization capacity, to address the optimisation for the multi-style fisheye imagery classification. We construct an srd-SIFT descriptor based SVM classifier to present the feature-based method for describing the influence of the dataset scale and the visual word scale on the classifier. The SVM classifier achieves 15.98% accuracy on the test set after 162 h training, with the condition that includes 800 images per class in 12 classes and 1500 visual words. For the learning-based method, we propose to expand training samples’ style variety, via style transformation, to facilitate the contemporary architecture retraining. Following this approach, we retrain the ResNet-50 by an artificial multi-style fisheye image dataset without complementing new training labels. The performance of the obtained ResNet classifier is evaluated on 6000 images collected in the real-world. The result shows that the retrained classifier has great generalization capacity and reaches 97.19% top-3 accuracy.

AB - The comparison between the feature-based method and the learning-based method is conducted in the training time, the accuracy and the generalization capacity, to address the optimisation for the multi-style fisheye imagery classification. We construct an srd-SIFT descriptor based SVM classifier to present the feature-based method for describing the influence of the dataset scale and the visual word scale on the classifier. The SVM classifier achieves 15.98% accuracy on the test set after 162 h training, with the condition that includes 800 images per class in 12 classes and 1500 visual words. For the learning-based method, we propose to expand training samples’ style variety, via style transformation, to facilitate the contemporary architecture retraining. Following this approach, we retrain the ResNet-50 by an artificial multi-style fisheye image dataset without complementing new training labels. The performance of the obtained ResNet classifier is evaluated on 6000 images collected in the real-world. The result shows that the retrained classifier has great generalization capacity and reaches 97.19% top-3 accuracy.

KW - DNN

KW - Fisheye image

KW - Style expansion

KW - Super vector machine

KW - Engineering

UR - http://www.scopus.com/inward/record.url?scp=85105858576&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/2d6c1b60-da08-3bbe-b5b5-9d4c1c4207ab/

U2 - 10.1007/978-3-030-73689-7_41

DO - 10.1007/978-3-030-73689-7_41

M3 - Article in conference proceedings

AN - SCOPUS:85105858576

SN - 978-3-030-73688-0

T3 - Advances in Intelligent Systems and Computing

SP - 424

EP - 433

BT - Proceedings of the 12th International Conference on Soft Computing and Pattern Recognition, SoCPaR 2020

A2 - Abraham, Ajith

A2 - Ohsawa, Yukio

A2 - Gandhi, Niketa

A2 - Jabbar, M. A.

A2 - Haqiq, Abdelkrim

A2 - McLoone, Seán

A2 - Issac, Biju

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 15 December 2020 through 18 December 2020

ER -