A Highly Efficient and Comprehensive Image Processing Library for C++-based High-Level Synthesis

Field Programmable Gate Arrays (FPGAs) are proved to be among the most suitable architectures for image processing applications. However, accelerating algorithms using FPGAs is a time-consuming task and needs expertise. Whereas the recent advancements in High-Level Synthesis (HLS) promise to solve this problem, today's HLS tools require apt hardware descriptions of algorithms to be able to provide favorable implementations. A solution is developing highly parameterizable and optimized HLS libraries for the fundamental image processing components. Another solution is providing a higher level of abstraction in the form of a Domain-Specific Language (DSL) and a corresponding efficient back end for hardware design. In this paper, we provide a highly efficient and parameterizable C++ library for image processing applications, which would be the cornerstone for both approaches. In our library, nodes of a stream-based data flow graph can be described as C++ objects for specified functions, and the whole application can be efficiently parallelized just by defining a global constant as the parallelization factor. Moreover, the key hardware design elements, i. e., line buffers and sliding windows with different border handling patterns, can be utilized individually to ease the design of more complicated applications.