Bounding Fundamental Performance of Feature-Based Object Recognition

Performance prediction is a crucial step for transforming the eld of object recognition from an art to a science. In this paper, we address this problem in the context of a vote-based approach for object recognition using 2-D point features. A method is presented for predicting tight lower and upper bounds on fundamental performance of the selected recognition approach. Performance bounds are predicted by considering data-distortion factors, which are uncertainty, occlusion and clutter, in addition to model structural similarity. Given a statistical model of data uncertainty, the structural similarity between every pair of model objects is computed as a function of the relative transformation between them. Model-similarity information is then used along with statistical data-distortion models to predict bounds on the probability of correct recognition. Validity of the method is experimentally demonstrated using MSTAR public SAR data.