Abstract—Future multi-media technologies are expected to support efficient on-line processing of huge amounts of high-dimensional data without any special pre-processing. In this paper, we will introduce a new model of the so-called Growing Hierarchical Neural Networks (GHNN) applicable to image classification without requiring advanced domain-specific feature extraction techniques. It can be, moreover, supposed that the involved dynamic data-dependent adjustment of both the number and position of the neurons improves generalization. Experimental results obtained so far for two case studies on face and hand-written digit recognition show that local features detected automatically by GHNN-networks impact a transparent and compact representation of the extracted knowledge.
Index Terms—Convolutional neural networks, image classification, face recognition, self-organization.
The authors are with the Department of Theoretical Computer Science and Mathematical Logic, Faculty of Mathematics and Physics, Charles University, Malostranske nam. 25, 118 00 Praha, Czech Republic (e-mail: iveta.mrazova@mff.cuni.cz, mkukacka@gmail.com).
[PDF]
Cite:Iveta Mrazova and Marek Kukacka, "Image Classification with Growing Neural Networks," International Journal of Computer Theory and Engineering vol. 5, no. 3, pp. 422-427, 2013.
