GPU implementation of a modified signed discrete cosine transform

Abstract

Real time imaging is essential for internet multimedia and modern satellite communications. Discrete Cosine Transform (DCT) is the core for image processing such as image compression and coding for its high power compaction property. The Signed DCT (SDCT) and its modifications approximate the DCT while requiring much less number of arithmetic operations which is essential to speed up real time applications. However, carrying out 2-D DCT or SDCT on CPU, takes much time for the high order of computation. This paper employs the Graphics Processing Unit (GPU) based on Compute Unified Device Architecture (CUDA) to reduce the time required to handle such high amount of computations. An efficient and fast modified SDCT (MSDCT) is employed. The essential features of the MSDCT are presented. A flow diagram is provided for the efficient implementation of the MSDCT. Only 17 additions are required for both forward and backward transformations. Computer analysis is provided to illustrate the high speed up achieved by the GPU implementation for the MSDCT.