An improved architecture for the adaptive discrete cosine transform

Abstract

This paper presents a new approach to the efficient realisation of the discrete cosine transform for the specific case of interlaced image sequence coding. In such cases, the conventional approach of decomposing each frame or frame difference into 8x8 blocks is often no longer satisfactory and an adaptive architecture capable of processing either 8x8 or two 4x8 blocks is desirable. The approach described is based on the decomposition used by Madisetti, modified to maximise shared hardware resources and to exploit arithmetic redundancy using primitive operator methods. The resulting architecture is compared with alternative implementation options using an area-time metric with savings in excess of 50% having been observed.