Transinformation of object recognition and its application to viewpoint planning

✍ Scribed by Bernt Schiele; James L. Crowley

• Publisher: Elsevier Science
• Year: 1997
• Tongue: English
• Weight: 952 KB
• Volume: 21
• Category: Article
• ISSN: 0921-8890
• DOI: 10.1016/s0921-8890(97)00009-2

✦ Synopsis

This article develops an analogy between the transmission of information through a channel and an object recognition process. This analogy is based on the statistical representation of 3D objects by several multidimensional receptive field histograms. Each histogram corresponds to a particular appearance of a 3D object. The analogy between transmission of information and object recognition allows to evaluate quantitatively the employed measurement (or feature) sets and therefore to predict the performance of the object recognition process. Furthermore one can determine the most discriminant viewpoints of objects by calculating the transinformation of each viewpoint. As an application the article develops an active object recognition algorithm which is able to resolve ambiguities inherent in a single-view recognition algorithm. Most interestingly the algorithm incorporates 3D information of the objects entirely based on 2D measurements in images of the object.