NovemE - Color Space Net for Image Classification

Autor:	Urvi Oza, Pankaj Kumar, Sarangi Patel
Rok vydání:	2021
Předmět:	Contextual image classification Computer science business.industry Training time Pattern recognition State (computer science) Artificial intelligence Color space Base (topology) Transfer of learning business Net (mathematics) Convolutional neural network
Zdroj:	Intelligent Information and Database Systems ISBN: 9783030732790 ACIIDS
DOI:	10.1007/978-3-030-73280-6_42
Popis:	Image classification is one of the basic applications of computer vision where several deep convolutional neural networks (DCNN) have been able to achieve state of the art results. In this paper, we explore the efficacious effect of color space transformations on image classification. After empirically establishing that color space transforms indeed affects the accuracy of the classification in DCNN, we propose NovemE, an ensemble-based model made up of nine (novem) base learners. We use transfer learning with significantly reduced training time and improved accuracy of classification. This model integrates different color spaces and DCNNs in order to achieve a higher accuracy in classifying the given data. We experimented with CINIC10 and Street View House Number (SVHN) datasets and were successful in achieving significant improvement on the current state of the art results on these datasets.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::e13874f88e52ec95d1d4159cb6d548de https://doi.org/10.1007/978-3-030-73280-6_42 Zobrazit plný text záznamu