ABSTRACT
Since handwriting recognition is very sensitive to structural noise, like superimposed objects such as straight lines and other marks, it is necessary to remove noise in a preprocessing stage before recognition. Although numerous denoising approaches have been proposed, it remains a challenge. The difficulties are due to non-locality of structural noise and hard discernment between spurious and the meaningful regions. We propose a supervised approach using deep learning to remove structural noise. Specifically, we generalize the deep autoencoder into the deep denoising autoencoder (DDAE), which consists in training a neural network with noisy and clean pairs to minimize cross-entropy error. Inspired by recurrent neural networks, we introduce feedback loop from the output to enhance the "repaired" image well in the reconstruction stage in our framework. We test the DDAE model on three handwritten image data sets, and show advantages over Wiener filter, robust Boltzmann machines and deep autoencoder.
- M. Aharon, M. Elad, and A. Bruckstein. K-svd: An algorithm for designing of overcomplete dictionaries for sparse representation. IEEE Trans. On Signal Processing, 54(11):4311–5322, 2005. Google ScholarDigital Library
- J. Besag. On the statistical analysis of dirty pictures. Journal of the Royal Statistical Society. Series B, 48(3):259–302, 1986.Google Scholar
- A. Buades, B. Coll, and J.-M. Morel. A non-local algorithm for image denoising. In CVPR, pages 60–65, Washington, DC, USA, 2005. IEEE Computer Society. Google ScholarDigital Library
- G. Chen and S. N. Srihari. A noisy-or discriminative restricted boltzmann machine for recognizing handwriting style development. In ICFHR, 2014.Google ScholarCross Ref
- G. Chen, C. Xiong, and J. J. Corso. Dictionary transfer for image denoising via domain adaptation. In ICIP, 2012.Google ScholarCross Ref
- M. Elad and M. Aharon. Image denoising via learned dictionaries and sparse representation. In CVPR, pages 17–22, 2006. Google ScholarDigital Library
- J. L. Elman. Finding structure in time. COGNITIVE SCIENCE, 14(2):179–211, 1990.Google ScholarCross Ref
- K. Engan, S. O. Aase, and J. H. Husoy. Frame based signal compression using method of optimal directions (mod). 1999.Google ScholarCross Ref
- P. Gallinari, Y. LeCun, S. Thiria, and F. Fogelman-Soulie. Memoires associatives distribuees. In Proceedings of COGNITIVA 87, 1987.Google Scholar
- R. Haddad and A. Akansu. A class of fast gaussian binomial filters for speech and image processing. Trans. Sig. Proc., 39(3):723–727, Mar. 1991. Google ScholarDigital Library
- G. E. Hinton, S. Osindero, and Y.-W. Teh. A fast learning algorithm for deep belief nets. Neural Comput., 18(7):1527–1554, July 2006. Google ScholarDigital Library
- G. E. Hinton and R. R. Salakhutdinov. Reducing the dimensionality of data with neural networks. Science, 313(5786):504–507, July 2006.Google ScholarCross Ref
- L. Holmström and P. Koistinen. Using additive noise in back-propagation training. Research Reports A3, Rolf Nevanlinna Institute, 1990.Google Scholar
- J. J. Hopfield. Neurocomputing: Foundations of research. chapter Neural Networks and Physical Systems with Emergent Collective Computational Abilities, pages 457–464. MIT Press, Cambridge, MA, USA, 1988. Google ScholarDigital Library
- V. Jain and H. S. Seung. Natural image denoising with convolutional networks. In D. Koller, D. Schuurmans, Y. Bengio, and L. Bottou, editors, NIPS, pages 769–776. Curran Associates, Inc., 2008.Google Scholar
- P. Perona and J. Malik. Scale-space and edge detection using anisotropic diffusion. IEEE PAMI, 12(7):629–639, July 1990. Google ScholarDigital Library
- J. Portilla, V. Strela, M. J. Wainwright, and E. P. Simoncelli. Image denoising using scale mixtures of gaussians in the wavelet domain. IEEE Trans. Image Process, 12:1338–1351, 2003. Google ScholarDigital Library
- M. Puri, S. N. Srihari, and Y. Tang. Bayesian network structure learning and inference methods for handwriting. In ICDAR, pages 1320–1324, 2013. Google ScholarDigital Library
- L. I. Rudin and S. Osher. Total variation based image restoration with free local constraints. In ICIP, pages 31–35. IEEE, 1994.Google Scholar
- L. Sendur and I. W. Selesnick. Bivariate Shrinkage With Local Variance Estimation. IEEE SIGNAL PROCESSING LETTERS, 9(12):438–441, Dec. 2002.Google ScholarCross Ref
- H. S. Seung. Learning continuous attractors in recurrent networks. In NIPS, pages 654–660. MIT Press, 1998. Google ScholarDigital Library
- C. E. Shannon. A mathematical theory of communication. SIGMOBILE Mob. Comput. Commun. Rev., 5(1):3–55, Jan. 2001. Google ScholarDigital Library
- H. Takeda, S. Farsiu, and P. Milanfar. Kernel regression for image processing and reconstruction. IEEE TRANSACTIONS ON IMAGE PROCESSING, 16(2):349–366, 2007. Google ScholarDigital Library
- Y. Tang and C. Eliasmith. Deep networks for robust visual recognition. In J. F§rnkranz and T. Joachims, editors, ICML, pages 1055–1062. Omnipress, 2010.Google Scholar
- Y. Tang, R. Salakhutdinov, and G. Hinton. Robust boltzmann machines for recognition and denoising. In CVPR, pages 2264–2271, Washington, DC, USA, 2012. IEEE Computer Society. Google ScholarDigital Library
- C. Tomasi and R. Manduchi. Bilateral filtering for gray and color images. In ICCV, pages 839–, Washington, DC, USA, 1998. IEEE Computer Society. Google ScholarDigital Library
- P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, and P.-A. Manzagol. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. JMLR, 2010. Google ScholarDigital Library
- N. Wiener. Extrapolation, Interpolation, and Smoothing of Stationary Time Series. The MIT Press, 1964. Google ScholarDigital Library
- J. Xie, L. Xu, and E. Chen. Image denoising and inpainting with deep neural networks. In NIPS, 2012.Google ScholarDigital Library
Index Terms
- Removing Structural Noise in Handwriting Images using Deep Learning
Recommendations
A deep convolutional neural network for salt-and-pepper noise removal using selective convolutional blocks▪
AbstractIn recent years, there has been an unprecedented upsurge in applying deep learning approaches, specifically convolutional neural networks (CNNs), to solve image denoising problems, owing to their superior performance. However, CNNs ...
Highlights- A convolutional neural network (SeConvNet) is proposed to denoise salt-and-pepper.
Deep Learning for OFDM Channel Estimation in Impulsive Noise Environments
AbstractImpulsive noise suppression is essential in orthogonal frequency division multiplexing (OFDM) systems, since impulsive noise may cause a serious decline in channel estimation performance. To solve this problem, a channel estimator based on ...
Triple discriminators - equipped GAN for Denoising of Chinese calligraphic tablet images
AbstractDenoising of Chinese calligraphic tablet images is of great importance in regard to the study of both content and character shapes in these images. Formerly GAN (generative adversarial network) based image denoising methods model the noise in the ...
Comments