High speed low level image processing on FPGAs using distributed arithmetic

Abstract

A method of calculating the 3x3 mask convolution required for many low level image processing algorithms is presented, which is well suited for implementing on reconfigurable FPGAs. The approach is based on distributed arithmetic, and uses the symmetry of weights present in many filter masks to save on circuit space, at the expense of requiring custom circuitry for each set of coefficients. A four product example system has been implemented on a Xilinx XC6216 device to demonstrate the suitability of this reconfigurable architecture for image processing systems.