Feature based parametric eigenspace method for object extraction

This study is done in the context of a navigation support system in which image understanding is used to guide the navigating agent in an unfamiliar environment. We are concerned with modeling which allows recognition under varying viewing angles and/or when occlusions are present. We present an approach to object extraction based on an extension of the parametric eigenspace method using local features from the image of the object of interest. Spatial constraints for the local features used to model the object are also recorded. Based on the detection of local features, and their location constraints several object regions are hypothesized in an input image. Fuzzy logic is used to grade the goodness of these regions allowing to discriminate between them in order to achieve accurate object extraction. The approach is illustrated for an object in an indoor scene environment