High Performance Design of Linear Algebra Operations for Image Processing in FPGA Implementation

Abstract

 Numerical linear algebra operations are key primitives in scientific computing. In this paper, performance optimizations of linear algebraic operations have been extensively investigated. FPGA-based high-performance designs are proposed for dot product, matrix-vector multiplication and matrix multiplication by identifying the parameters for each operation and analyzing the trade-offs. It is proposed to implement dot product in convolution filter, which is useful in noise removal, and Matrix multiplication in boundary tracing, which is useful for shape analysis and calculating geometric features. These high performance designs of linear algebra applications are proposed to be implemented on Xilinx FPGAs.