An Analysis of the Road Signs Classification Based on the Higher-Order Singular Value Decomposition of the Deformable Pattern Tensors - Advanced Concepts for Intelligent Vision Systems - page 190

An Analysis of the Road Signs Classification Based on the
Higher-Order Singular Value Decomposition of the
Deformable Pattern Tensors
Bogusław Cyganek
AGH University of Science and Technology
Al. Mickiewicza 30, 30-059 Kraków, Poland
[e-mail]
Abstract. The paper presents a framework for classification of rigid objects in
digital images. It consists of a generator of the geometrically deformed
prototypes and an ensemble of classifiers. The role of the former is to provide a
sufficient training set for subsequent classification of deformed objects in real
conditions. This is especially important in cases of a limited number of
available prototype exemplars. Classification is based on the Higher-Order
Singular Value Decomposition of tensors composed from the sets of deformed
prototypes. Construction of such deformable tensors is flexible and can be done
independently for each object. They can be obtained either from a single
prototype, which is then affinely deformed, or from many real exemplars, if
available. The method was tested in the task of recognition of the prohibition
road signs. Experiments with real traffic scenes show that the method is
characteristic of high speed and accuracy for objects seen under different
viewpoints. Implementation issues of tensor decompositions are also discussed.
1 Introduction
Recognition of objects in digital images is a key task of Computer Vision. However, the
problem is complicated due to a great diversity of objects of interest, on the one hand,
and limited information provided in digital images, on the other. Nevertheless, due to
development of new classification methods and computational techniques, it is possible
to construct some frameworks for fast and reliable recognition of at least some groups
of well defined objects. In this paper we present one of such software frameworks. It
can classify rigid objects detected in images which views are subject to a subgroup of
projective transformations and noise. These unavoidable distortions are due to the
geometrical and physical properties of the observed objects and conditions of image
acquisition. The presented system relies mostly on the set of classifiers which perform a
multilinear analysis of tensors which are composed of the prototype exemplars of
objects. However, frequently the latter are not available in a sufficient number to allow
recognition of geometrically transformed views of objects. Therefore to remedy this
constraint the second important module of our framework is a generator of affinely
deformed and noise conditioned artificial prototypes. Thus, the system can still reliably
recognize an object even if only its single prototype is provided. Obviously, the more
input prototypes, the better results can be obtained. The method is able to cope with
different number of these for each object it is trained to recognize.