Zernike Pooling: Generalizing Average Pooling Using Zernike Moments
Most of the established neural network architectures in computer vision are essentially composed of the same building blocks (e.g., convolutional, normalization, regularization, pooling layers, etc.), with their main difference being the connectivity of these components within the architecture and n...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2021-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9524712/ |
| _version_ | 1818690797876281344 |
|---|---|
| author | Thomas Theodoridis Kostas Loumponias Nicholas Vretos Petros Daras |
| author_facet | Thomas Theodoridis Kostas Loumponias Nicholas Vretos Petros Daras |
| author_sort | Thomas Theodoridis |
| collection | DOAJ |
| description | Most of the established neural network architectures in computer vision are essentially composed of the same building blocks (e.g., convolutional, normalization, regularization, pooling layers, etc.), with their main difference being the connectivity of these components within the architecture and not the components themselves. In this paper we propose a generalization of the traditional average pooling operator. Based on the requirements of <italic>efficiency</italic> (to provide information without repetition), <italic>equivalence</italic> (to be able to produce the same output as average pooling) and <italic>extendability</italic> (to provide a natural way of obtaining novel information), we arrive at a formulation that generalizes average pooling using the Zernike moments. Experimental results on <italic>Cifar 10</italic>, <italic>Cifar 100</italic> and <italic>Rotated MNIST</italic> data-sets showed that the proposed method was able to outperform the two baseline approaches, global average pooling and average pooling <inline-formula> <tex-math notation="LaTeX">$2 \times 2$ </tex-math></inline-formula>, as well as the two variants of Stochastic pooling and AlphaMEX in every case. A worst-case performance analysis on <italic>Cifar-100</italic> showed that significant gains in classification accuracy can be realised with only a modest 10% increase in training time. |
| first_indexed | 2024-12-17T12:31:43Z |
| format | Article |
| id | doaj.art-cb6cc30132b548919ef42b7435c1b074 |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-12-17T12:31:43Z |
| publishDate | 2021-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-cb6cc30132b548919ef42b7435c1b0742022-12-21T21:48:32ZengIEEEIEEE Access2169-35362021-01-01912112812113610.1109/ACCESS.2021.31086309524712Zernike Pooling: Generalizing Average Pooling Using Zernike MomentsThomas Theodoridis0https://orcid.org/0000-0003-2955-7331Kostas Loumponias1https://orcid.org/0000-0002-6268-3893Nicholas Vretos2https://orcid.org/0000-0003-3604-9685Petros Daras3https://orcid.org/0000-0003-3814-6710Information Technologies Institute (ITI), Centre for Research and Technology Hellas (CERTH), Thessaloniki, GreeceInformation Technologies Institute (ITI), Centre for Research and Technology Hellas (CERTH), Thessaloniki, GreeceInformation Technologies Institute (ITI), Centre for Research and Technology Hellas (CERTH), Thessaloniki, GreeceInformation Technologies Institute (ITI), Centre for Research and Technology Hellas (CERTH), Thessaloniki, GreeceMost of the established neural network architectures in computer vision are essentially composed of the same building blocks (e.g., convolutional, normalization, regularization, pooling layers, etc.), with their main difference being the connectivity of these components within the architecture and not the components themselves. In this paper we propose a generalization of the traditional average pooling operator. Based on the requirements of <italic>efficiency</italic> (to provide information without repetition), <italic>equivalence</italic> (to be able to produce the same output as average pooling) and <italic>extendability</italic> (to provide a natural way of obtaining novel information), we arrive at a formulation that generalizes average pooling using the Zernike moments. Experimental results on <italic>Cifar 10</italic>, <italic>Cifar 100</italic> and <italic>Rotated MNIST</italic> data-sets showed that the proposed method was able to outperform the two baseline approaches, global average pooling and average pooling <inline-formula> <tex-math notation="LaTeX">$2 \times 2$ </tex-math></inline-formula>, as well as the two variants of Stochastic pooling and AlphaMEX in every case. A worst-case performance analysis on <italic>Cifar-100</italic> showed that significant gains in classification accuracy can be realised with only a modest 10% increase in training time.https://ieeexplore.ieee.org/document/9524712/Neural networkspoolingZernike momentsimage classification |
| spellingShingle | Thomas Theodoridis Kostas Loumponias Nicholas Vretos Petros Daras Zernike Pooling: Generalizing Average Pooling Using Zernike Moments IEEE Access Neural networks pooling Zernike moments image classification |
| title | Zernike Pooling: Generalizing Average Pooling Using Zernike Moments |
| title_full | Zernike Pooling: Generalizing Average Pooling Using Zernike Moments |
| title_fullStr | Zernike Pooling: Generalizing Average Pooling Using Zernike Moments |
| title_full_unstemmed | Zernike Pooling: Generalizing Average Pooling Using Zernike Moments |
| title_short | Zernike Pooling: Generalizing Average Pooling Using Zernike Moments |
| title_sort | zernike pooling generalizing average pooling using zernike moments |
| topic | Neural networks pooling Zernike moments image classification |
| url | https://ieeexplore.ieee.org/document/9524712/ |
| work_keys_str_mv | AT thomastheodoridis zernikepoolinggeneralizingaveragepoolingusingzernikemoments AT kostasloumponias zernikepoolinggeneralizingaveragepoolingusingzernikemoments AT nicholasvretos zernikepoolinggeneralizingaveragepoolingusingzernikemoments AT petrosdaras zernikepoolinggeneralizingaveragepoolingusingzernikemoments |