Improved Appearance-Based 3-D Object Recognition Using Wavelet Features
Michael P. Reinhold, Dietrich W. R. Paulus, Hendrik J. Niemann · 2001
In this paper we present an improved appearancebased approach for the localization and classification of 3-D objects in 2-D gray level images. Thereby we calculate local feature vectors by the coefficients of the wavelet multiresolution analysis and model them statistically. Since the appearance of the objects, i. e. also the size of the objects in the image, vary due to out-of-image-plane transformations, the features themselves as well as the region of interest are modelled as function of the external transformations. Further, we present and test different measurements for the recognition of objects that have different sizes. The experiments on a large dataset with more than 40000 images show that the approach is well suited for this recognition task. 1