This paper presents a hardware architecture for calculating the city-block and chessboard distance transform on binary images. It is based on applying multiple morphological erosions and adding the result, enabling both processing pixels in raster scan order and a deterministic execution time. Which distance metric to be calculated is determined by the shape of the structuring element, i.e. diamonds for the city-block and squares for the chessboard. These properties together with a low memory requirement make the architecture applicable in any streaming data real-time embedded system environment with hard timing constraints, e.g. set by the frame rate. Depending on the application, a priori knowledge of the maximum size of the clusters may be used to reduce execution time and memory requirement even further. The architecture has been implemented for both FPGA and ASIC in an embedded system environment with an image resolution of 320 × 240 at a frame rate of 25 fps, running at 100 MHz and 454 MHz, respectively.
展开▼
