A Pyramidal Neural Network Based on Nonclassical Receptive Field Inhibition

This paper presents a new artificial neural network, called I-PyraNet. This new architecture is based on the combination between concepts of the recently described PyraNet and the nonclassical receptive fields inhibition, integrating the feature extraction and the classification stages into the same structure which is formed by 2-D and 1-D layers. The main difference between the PyraNet and the I-PyraNet is that while in the first a 2-D neuron always provide the same output, in the I-PyraNet the signal of the output of a 2-D neuron will invert when it appears inside a inhibitory field. Furthermore, the I-PyraNet is applied over a face detection task where different configurations of the network are tested.