Design of multi-dimensional DCT array processors for video applications

Abstract

In this paper, we propose array processors for computing the 2-D and 3-D DCTs. We first introduce a new method, called dimensional splitting method, for the design of array processors for multi-dimensional image transforms. The method can be applied to any multi-dimensional image transforms with separable kernels such as DFT and DCT. Then, we propose a new coding scheme for the 1-D DCT in which the need for generating the kernel matrix in advance is eliminated. Finally, we show array processors for computing r-D DCT (r ≥ 2) which are scalable, regular, locally-connected, and fully-pipelined.