Building an Artificial Retina for Distance- and Orientation-Invariant Pattern Recognition

The present paper demonstrates how — due to a special construction principle for an artificial etina — orientation- and distance-invariant pattern recognition can be realized. Logarithmic retinae in their pure form can support invariant recognition, but they suffer from difficulties concerning an inhomogeneous representation of objects. Selection of ring shaped representations from a superposition of differently scaled logarithmic retinae helps to overcome these difficulties. On the basis of this retinal structure the orientation of objects can be identified with an angular resolution of ± 1.4°. This orientation information combined with available distance information serves as input for mapping processes producing normalized object representations.