Local context in non-linear deformation models for handwritten character recognition

We evaluate different two-dimensional non-linear deformation models for handwritten character recognition. Starting from a true two-dimensional model, we derive pseudo-two-dimensional and zero-order deformation models. Experiments show that it is most important to include suitable representations of the local image context of each pixel to increase performance. With these methods, we achieve very competitive results across five different tasks, in particular 0.5% error rate on the MNIST task.

[1]  Hyun-Chul Kim,et al.  A numeral character recognition using the PCA mixture model , 2002, Pattern Recognit. Lett..

[2]  Jitendra Malik,et al.  Shape Context: A New Descriptor for Shape Matching and Object Recognition , 2000, NIPS.

[3]  Seiichi Uchida,et al.  A monotonic and continuous two-dimensional warping based on dynamic programming , 1998, Proceedings. Fourteenth International Conference on Pattern Recognition (Cat. No.98EX170).

[4]  Patrice Y. Simard,et al.  Best practices for convolutional neural networks applied to visual document analysis , 2003, Seventh International Conference on Document Analysis and Recognition, 2003. Proceedings..

[5]  Hermann Ney,et al.  Classification of Medical Images using Non-linear Distortion Models , 2004, Bildverarbeitung für die Medizin.

[6]  Geoffrey E. Hinton,et al.  Modeling the manifolds of images of handwritten digits , 1997, IEEE Trans. Neural Networks.

[7]  Bernhard Schölkopf,et al.  Training Invariant Support Vector Machines , 2002, Machine Learning.

[8]  Seiichi Uchida,et al.  Handwritten character recognition using elastic matching based on a class-dependent deformation model , 2003, Seventh International Conference on Document Analysis and Recognition, 2003. Proceedings..

[9]  Seiichi Uchida,et al.  Eigen-deformations for elastic matching based handwritten character recognition , 2003, Pattern Recognit..

[10]  Oscar E. Agazzi,et al.  Keyword Spotting in Poorly Printed Documents using Pseudo 2-D Hidden Markov Models , 1994, IEEE Trans. Pattern Anal. Mach. Intell..

[11]  Hermann Ney,et al.  Experiments with an extended tangent distance , 2000, Proceedings 15th International Conference on Pattern Recognition. ICPR-2000.

[12]  Michael E. Tipping,et al.  Probabilistic Principal Component Analysis , 1999 .

[13]  Hermann Ney,et al.  Combination of Tangent Vectors and Local Representations for Handwritten Digit Recognition , 2002, SSPR/SPR.

[14]  Daniel Keysers,et al.  Elastic image matching is NP-complete , 2003, Pattern Recognit. Lett..

[15]  Yoshua Bengio,et al.  Gradient-based learning applied to document recognition , 1998, Proc. IEEE.