Tangent Vectors and Local Representations Combination of Tangent Vectors and Local Representations for Handwritten Digit Recognition Keysers, Paredes, Ney, Vidal Daniel Keysers (RWTH Aachen), Roberto Paredes (ITI, UP Valencia), Hermann Ney (RWTH Aachen), Enrique Vidal (ITI, UP Valencia)

Combination of Tangent Vectors
and Local Representations for
Handwritten Digit Recognition

Daniel Keysers, Roberto Paredes, Hermann Ney, and Enrique Vidal

Abstract:

Statistical classification using tangent vectors and classification based on local features are two successful methods for various image recognition problems. These two approaches tolerate global and local transformations of the images, respectively. Tangent vectors can be used to obtain global invariance with respect to small affine transformations and line thickness, for example. On the other hand, a classifier based on local representations admits the distortion of parts of the image. From these properties, a combination of the two approaches seems very likely to improve on the results of the individual approaches. In this paper, we show the benefits of this combination by applying it to the well known USPS handwritten digits recognition task. An error rate of 2.0% is obtained, which is the best result published so far for this dataset.