Statistical 3-D object localization without segmentation using wavelet analysis

Abstract

This paper presents a new approach for statistical object localization. The localization scheme is directely based on local features, which are extracted for all image positions, in contrast to segmentation in classical schemes. Hierarchical Gabor filters are used to extract local features. With these features statistical object models are built for the different scale levels of the Gabor filters. The localization is then performed by a maximum likelihood estimation on the different scales successively. Results for the localization of real images of 2-D and 3-D objects are shown.