Affine invariant recognition of 2D occluded objects using geometric hashing and distance transformation

An efficient approach for recognition of partially occluded objects from 2-D grey level images is presented. It can be divided into three stages. The pre-processing stage includes local feature extraction from 2-D grey level images and the formation of a hash table. In the recognition stage, a geometric hashing technique is used to vote for the point correspondences between the scene and the models. Finally, distance transformation is employed for verification. An average mismatch distance is defined to measure the goodness of the match quantitatively. The approach has been successfully tested on recognising a number of industrial handtools overlapping each other