Implementation of DCT-domain motion estimation and compensation

Abstract

Hybrid video compression schemes store an entire image for predictive coding. Traditionally, this image is stored in the time domain needing almost 5 Mbit of memory for main-level image format. The amount of storage space can be reduced if the data is stored in the runlength encoding-discrete cosine transform (RLE-DCT) domain, even using the available compression and buffer-control algorithm to guarantee a storage amount. We show that a hardware implementation is feasible and worthwhile in relation to traditional encoders. Motion estimation is performed by recursive block-matching, at sub-pixel accuracy.