A stream processing architecture for real-time implementation ofperspective spatial transformations

摘要

A system is presented to perform perspective spatialtransformations at video-rate using a mapping function which can vary atframe rate. The mapping is implemented by two successive applications ofa one-dimensional algorithm operating in scan-line order using aframestore to store an intermediate image. A novel one-dimensionalfiltering algorithm provides a continuous mapping from source todestination co-ordinates, preventing aliasing. These algorithms areimplemented using a stream processing architecture, an efficientmechanism for video-rate operations. Emphasis is placed on the design ofa novel stream processing architecture which, coupled with an efficientpipelined co-ordinate generation system, provides single-cycle peroutput pixel performance