Fine-Granular Parallel EBCOT and Optimization with CUDA for Digital Cinema Image Compression

Abstract

JPEG2000 has been accepted by The Society of Motion Picture and Television Engineers (SMPTE) as the image compression standard for the digital distribution of motion pictures. In JPEG2000, the biggest contribution to the coding performance comes from the Embedded Block Coding with Optimized Truncation (EBCOT), which is also the most time-consuming module by occupying almost 37% of the encoding time. There have been many research activities in the optimization of EBCOT on platforms like FPGA and VLSI, but on Graphics Processing Unit (GPU), a currently popular parallel computing platform in post-production of motion pictures, still few works have been done. This paper proposes a fine-granular parallel EBCOT by re-designing the highly serialized bit-plane coding to a parallel structure where the coding of all bits in a bit-plane could be performed in parallel, then the bit coding tasks can be distributed to the stream processors in GPU by taking advantage of the programming and memory model of CUDA. Experimental results show that our algorithms reveal 3 to 4 times computational speed improvement on an ordinary GPU compared to that on CPU.