Direct implementation of 2-D DCT on a low-cost linear-array architecture without intermediate transpose memory

A direct method for the computation of 2-D DCT on a linear-array architecture is presented. The original 2-D DCT is converted into 1-D problem with representation of matrix-vector product. Then, we propose a fast algorithm with low computation complexity, and exploit an efficient mapping technique to generate from the algorithm a hardware-efficient architecture. Unlike other 2-D DCT processors that usually require transpose memory, our new architecture is easily pipelined for purpose of high throughput rate and is easily scalable for the computation of longer-length DCT.