Improving deep neural networks classification by preprocessing images
Nenhuma Miniatura disponível
Citações na Scopus
0
Tipo de produção
Artigo de evento
Data
2016
Autores
ERDMANN, H.
ITO, F. T.
TAKABAYASHI, D.
DOS SANTOS, D. N.
ITO, F. T.
TAKABAYASHI, D.
DOS SANTOS, D. N.
Orientador
Periódico
Computational Vision and Medical Image Processing V - Proceedings of 5th Eccomas Thematic Conference on Computational Vision and Medical Image Processing, VipIMAGE 2015
Título da Revista
ISSN da Revista
Título de Volume
Citação
ERDMANN, H.; ITO, F. T.; TAKABAYASHI, D.; DOS SANTOS, D. N. Improving deep neural networks classification by preprocessing images. Computational Vision and Medical Image Processing V - Proceedings of 5th Eccomas Thematic Conference on Computational Vision and Medical Image Processing, VipIMAGE, p. 313-319, oct. 2015.
Texto completo (DOI)
Palavras-chave
Resumo
© 2016 Taylor & Francis Group, London.The amount of images used to train a classifier has a great impact in the algorithm’s performance. In the domain of e-commerce, most of the dataset is formed by studio images and we demonstrate in this work that by multiplying the amount of images with different transformations can provide a significant boost on the overall performance of the classifier. Moreover, we list the challenges encountered in such task and present the improvements obtained with a classifier trained with additional images that were synthetically generated by applying several transformations on the original dataset.