Andr'e Wolfram

Fuzzy invariant indexing: a general indexing scheme for occluded object recognition

In this paper we present a general indexing scheme for model-based occluded object recognition. This new technique is called fuzzy invariant indexing (FII). It is based on fuzzy if-then-classification rules and fuzzified invariant object description, which can be either invariant values or invariant signatures. We describe the framework of the method and its general applicability: if invariant object descriptions can be constructed, the fuzzy invariant indexing technique can be utilized for recognizing the objects. The recognition performance of the FII-technique is demonstrated for partially occluded (quasi-) planar objects as well as for rotationally symmetric objects in real scenes.

Generation of dense range maps by data fusion from active and passive colour stereo vision

The problem of correspondence search in a stereo system may be alleviated if an active light source projects colour codes onto the scene. To reduce the amount of ambiguous or false matches, additional information drawn from feature-based passive stereo techniques can be used In this paper we discuss the application of fuzzy data fusion techniques to a dual-mode stereo system. Information from passive stereo is used whenever the active procedure cannot find an unambiguous match. The algorithm presented in the paper uses a fuzzy rule-based approach to integrate crisp as well as fuzzy input data. It is shown that the combination of data obtained in both modes of operation may lead to a significant increase in the number of matches meeting a certain credibility criterion. We present some of our experimental results based on real stereo image data and conclude the paper with a discussion of perspectives for possible future research.

Recognition of partially occluded objects through fuzzy invariant indexing

We present an approach to the recognition of partially occluded objects employing fuzzy invariant values and fuzzy if-then-rules, called fuzzy invariant indexing (FII). Compared with traditional invariant indexing, the fuzzy method proposed here offers the following advantages: firstly, as shown in the experimental results of the paper the recognition quality may be considerably increased in the case of similar objects; secondly, the ability is provided to control the recognition process during the hypothesis evaluation stage, and thirdly, a FII-based recognition system can be simply extended in a closed form, i.e. new attributes may be added to the fuzzy classification rules resulting in only minor changes to the original structure of the system. We demonstrate the recognition performance of the new FII-technique for partially occluded (quasi-)planar objects in real image scenes taken from different camera viewpoints and conclude the paper with a discussion of the potential of the method and directions of possible future research.