Recent advances in computer vision and image processing using reconfigurable hardware

This ‘Special Issue on FPGAs: Case Studies in Computer Vision and Image Processing’ collects high-quality state-ofthe-art papers to provide a more comprehensive perspective of the great potential of FPGAs in computer vision and image processing fields. Vision has been centre of attention for researchers from the first beginnings of computing since it is the most notorious perception mechanism for human beings. The research into the emulation of the visual capability by means of computers has been developed to such an extent that, at present, vision plays a very important role in many of the application fields, from medical imaging to factory automation, passing for robotics; and it is a key discipline in the most general objective of allowing machines to understand the world. Generally, a computer vision system extracts quantitative information from an image using the following steps: image acquisition, image manipulation, image understanding, and decision making (actuation). The last three steps can be performed by a computer. On the other hand, the image manipulation and understanding steps are carried out by means of image processing techniques. Any situation requiring the enhancement, restoration or analysis of a digital image is a good candidate for these techniques. The main challenge is that computer vision systems are normally used in real-time applications. The use of a generalpurpose computer allows simple verification of an algorithm, but not usually real-time processing. For this reason, very different technologies have been used in order to build computer vision applications. These technologies go from parallel architectures to specific-purpose processors, or even programmable logic devices. Parallel computer vision architectures are based on the fact that computer vision implies the execution of data intensive and computing intensive applications that require the use of parallel mechanisms. The different computer vision systems have used parallel architectures with different interconnection structures: lineal array, mesh, pyramid,.. These structures show a good performance in specific stages of a computervision operation chain; but none of them, by itself, is in general suited for complete applications. Also, despite the large number of researches and the advances in parallel