A neural network model for object invariant recognition

We apply a biologically inspired model of visual object recognition to the object translation invariant problem. Our model modifies that of Serre, Wolf, and Poggio. This biologically inspired model is hierarchical, formed after the division of the human visual system, which resembles the increasing size of the receptive field in the ventral stream that aid in invariant object recognition. As in this work, we first apply Gaussion filters at all positions; feature complexity and position invariance are built up after the network being setup. We refine the approach in several biologically plausible ways, using lateral inhibition and the trace learning rule. After training, the output of the neural network model becomes tolerant to certain changes about the input thus gradually learning about the different transformation of the object. The network was tested using simple objects such as T, L, and +, achieving excellent performance. The results strengthen the case for using this class of model in computer vision.