Programmable Opto-electronic CNN implementation provides a new and powerful tool for image processing applications

An opto-electronic CNN implementation, as it combines the huge inherent computational capabilities of a massively parallel optical system with programmability of the CNN-UM systems, provides an efficient frame for diverse image processing applications. Our optical CNN implementation is based on a new, semi incoherent optical correlator architecture that is superior from several points of view to any alternative optical correlator. This architecture provides the possibility of real time reprogramming of the correlator like in the joint Fourier transform correlators, but preserves the speed of VanderLugt type of systems. Programming is done through fast holographic read-out by B-templates. With combination of a high speed CNN-UM sensor chip it forms a programmable optoelectronic analogic computer (POAC) architecture. This device is especially useful in those image-processing algorithms, which requires a great number of pattern