Automatische Merkmalsynthese : ein Lern- und Klassifikationssystem zur Erkennung komplexer und deformierter, natürlicher und künstlicher Objekte

This work deals with the recognition of isolated visual objects. These objects stem from the real world and in general they are complex and deformed.
The outer contours of 3d-Objects, that are projected on a plane, are analyzed
with regard to local significant contour sections. These sections are sufficient in order to distinguish the contours from different classes.
It is emphasized that the subject automatic feature generation is to the fore in
this work. Up to know there is paid little attention to this subject. The automatic feature generation must not be confused with the feature selection. Feature selection deals with the selection of an as much as possible optimal subset from a givenfeature set. Also automatic feature generation must not confused with the unsystematic conventional feature extraction methods. In the field of traditional feature
extraction predefined features like the length-width-ratio are extracted from an object in the image. In the field of automatic feature generation abstract features like the top of a bottle or the fin of a fish are generated systematically. This is done by synthesizing more complex features out of basic features of the same kind like successive contour points. The main attention is not directed to the construction of a classifier, but to a systematic innovative feature generation method based on automatic feature synthesis.