Word Recognition by Combining Outline Emphasis and Synthesize Background
Character recognition collects item keywords from images from e-commerce websites; however, it requires a huge amount of training data. In this paper, we propose an efficient method to collect the training data by generating synthesis images and emphasizing outlines to obtain realistic images. The proposed method improves recognition accuracy on both generated images and real images from e-commerce websites.
KeywordsCharacter recognition Synthesis image CNN
- 1.Jaderberg, M., Simonyan, K., Vedaldi, A., Zisserman, A.: Synthetic data and artificial neural networks for natural scene text recognition. In: NIPS Deep Learning Workshop, arXiv 2014 (2014)Google Scholar
- 2.Kobayashi, T., Nakagawa, M.: A pattern classification method of linear-time learning and constant-time classification. IEICE 89(11), 981–992 (2006)Google Scholar
- 5.Wang, T., Wu, D.J., Coates, A., Ng, A.Y.: End-to-end text recognition with convolutional neural networks. In: ICPR (2012)Google Scholar
- 6.Nair, V., Hinton, G.E.: Rectified linear units improve restricted boltzmann machines. In: ICML, pp. 807–814 (2010)Google Scholar
- 7.Hinton, G.E., Srivastava, N., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.R.: Improving neural networks by preventing co-adaptation of feature detectors. Clinical Orthopaedics and Related Research, abs/1207.0850 (2012)Google Scholar