A fast FPGA-based clustering algorithm for real time image processing

Real time image analysis has undergone a rapid development in the last few years, due to the increasing availability of low cost computational power. However computing power is still a limit for some high quality applications. Highresolution medical image processing, for example, are strongly demanding for both memory (~250 MB) and computational capabilities allowing for 3D processing in affordable time. We propose the reduction of execution time of image processing exploiting the computing power of parallel arrays of Field Programmable Gate Arrays (FPGAs). We apply this idea to an algorithm that finds clusters of contiguous pixels above a certain programmable threshold and process them to produce measurements that characterize their shape. It is a fast general-purpose algorithm for high-throughput clustering of data "with a two dimensional organization". The two-dimensional problem is well processed by FPGAs since their available logic is naturally organized into a 2-dimensional array. The algorithm is designed to be implemented with FPGAs but it can also profit of cheaper custom electronics. The key feature is a very short processing time that scales linearly with the amount of data to be processed. This means that clustering can be performed in pipeline with the image acquisition, without suffering from combinatorial delays due to looping multiple times through the whole amount of data.