SEMANTIC OBJECT RECOGNITION USING CLUSTERING AND DECISION TREES

Each object in a digital image is composed of many patches (segments) with different shapes and colors. In order to recognize an object, e.g. a table or a book, it is necessary to find out which segments are typical for which object and in which segment neighborhood they occur. If a typical segment in a characteristic neighborhood is found, this segment will be part of the object to be recognized. Typical adjacent segments for a certain object define the whole object in the image. Following this idea, we introduce a procedure that learns typical segment configurations for a given object class by training with example images of the desired object, which can be found in and downloaded from the Internet. The procedure employs methods from machine learning, namely k-means clustering and decision trees, and from computer vision, e.g. contour signatures.