Object recognition based on projection

The authors propose a new connectionist model for invariant object recognition. The model has four stages. The first stage obtains projections from the input image plane. The projection features can separate translation and rotation into two independent translations in s and θ, where s and θ are the axes of the transformed domain directions. The first stage of the recognition network is a discrete implementation of the Radon transform. The second stage adopts the Mellin transform to provide scale invariance. The outputs of this stage are translation and scale invariant. To obtain rotation invariance, the Rapid transform is used. Finally, a pattern classification network is used for recognition. In experiments, five exemplars were used to recognize 240 transformed test objects, and very good results were obtained