Recognition of Handwritten Digits and Human Faces by Convolutional Neural Networks

TitleRecognition of Handwritten Digits and Human Faces by Convolutional Neural Networks
Publication TypeTechnical Report
Year of Publication1996
AuthorsNeubauer, C.
Other Numbers1066
Abstract

Convolutional neural networks provide an efficient method to constrain the complexity of feedforward neural networks by weightsharing. In this paper two variations of convolutional networks - Neocognitron and Neoperceptron - are compared with classifiers based on fully connected feedforward layers (i.e. Multilayerperceptron, Nearest Neighbor Classifier, Autoencoding network). Beside the original Neocognitron a modification called Neoperceptron is proposed which combines neurons from Perceptron with the localized network structure of Neocognitron. Instead error backpropagation in this work a modular training procedure is applied, whereby layers are trained sequentially from the input to the output layer in order to recognize features of increasing complexity.For a quantitative experimental comparison with standard classifiers two recognition tasks have been chosen: handwritten digit recognition and face recognition. In the first example on handwritten digit recognition the generalization of convolutional networks is compared to fully connected networks. In several experiments the influence of variations of position, size and orientation of digits is determined and the relation between training sample size and validation error is observed. In the second example recognition of human faces is investigated under constrained and variable conditions with respect to face orientation and illumination and the limitations of convolutional networks are discussed.

URLhttp://www.icsi.berkeley.edu/ftp/global/pub/techreports/1996/tr-96-058.pdf
Bibliographic Notes

ICSI Technical Report TR-96-058

Abbreviated Authors

C. Neubauer

ICSI Publication Type

Technical Report