We present a complete object recognition system for 3-D objects using a viewercentered object description, so-called surface normal images (SNIs), recently introduced by Park et al. . Based on this representation we utilize a weak active technique (the Photometric Stereo Method (PSM)) to extract 3-D features from the objects. We combine surface orientations with an approximated line drawing to build 2.5-D models. Furthermore we develop an accumulator based matching method, which is adaptive and tolerant regarding the measurement errors. This includes a module to analyze the composition of the actual object library, that supports the construction of the index hierarchy. An effective technique is proposed that combines the results of the sequential feature matching of the rotated 2.5-D scene model set. Both the reconstruction level and the matching level of the object recognition system were tested successfully with synthetic and real object data bases.