Parallel Image Processing Using Incoherent Optical General Matrix Multiplication

Image filtering is essential to image processing and machine vision systems. To enhance the throughput, multiple filters are applied simultaneously using matrix-matrix multiplication. Here we experimentally demonstrate the use of a novel optical general matrix multiplication (OGeMM) scheme to apply several image kernels, namely Laplace, Gaussian, and Sobel horizontal and vertical, in parallel. The OGeMM uses incoherent laser and wavelength multiplexing, handles signed numbers, and enables single simultaneous data write and read for single-shot parallel image processing.