A trainable feature extractor for handwritten digit recognition

Abstract : This article focuses on a particular task among pattern recognition, the handwritten digit recognition. More precisely, the problems of feature extraction and classification are explored. A trainable feature extractor based on the LeNet5 convolutional neural network architecture is introduced to solve the first problem in a black box scheme without prior knowledge on the data. The classification task is performed by Support Vector Machines to enhance the generalization ability of LeNet5. In order to increase the recognition rate, new training samples are generated by affine transformations and elastic distortions. Experiments are performed on the well known MNIST database to validate the method and the results show that the system can outperfom both SVMs and LeNet5 while providing performances comparable to the best performance on this database. Moreover, an analysis of the errors is conducted to discuss possible means of enhancement and their limitations.
Document type :
Journal articles
Pattern Recognition, Elsevier, 2007, 40 (6), pp.1816-1824. <10.1016/j.patcog.2006.10.011>

Contributor : Fabien Lauer <>
Submitted on : Thursday, February 2, 2006 - 4:17:57 PM
Last modification on : Thursday, May 19, 2016 - 1:05:53 AM
Document(s) archivé(s) le : Saturday, April 3, 2010 - 10:04:47 PM




Fabien Lauer, Ching Y. Suen, Gérard Bloch. A trainable feature extractor for handwritten digit recognition. Pattern Recognition, Elsevier, 2007, 40 (6), pp.1816-1824. <10.1016/j.patcog.2006.10.011>. <hal-00018426>



Record views


Document downloads