Logarithmic r-θ Map for Hybrid Optical Neural Network Filter for Object Recognition within Cluttered Scenes

Space-variant imaging sensors can be designed to exhibit in-plane rotation and scale invariance to image data. We combine the complex logarithmic r-θ mapping of a space-variant imaging sensor with the hybrid optical neural network filter to achieve, with a single pass over the input data, simultaneous invariance to: out-of-plane rotation; in-plane rotation; scale; projection and shift invariance. The resulting filter we call a complex logarithmic r-θ mapping for the hybrid optical neural network filter. We include in the L-HONN filter’s design a window based unit for registering the translation invariance of the input objects, initially lost by applying the logarithmic mapping. We test and record the results of the L-HONN filter for single and multiple input objects of the same class within cluttered still images and video frame sequences.