Using GPUs for JPEG Decoding

Temporary Supervisor

Dr Eric McCreath


The JPEG image format is a lossy image format which is commonly used by digital cameras. Some Webcams and IP cameras will use Motion JPEG in which frames of a video are made up of independent JPEG images. JPEG images use DCT (Discrete Co-Sin Transformations) on patches of 8x8 pixels, once in the frequency domain the values are quantised, these quantised values are finally stored using lossless Huffman encoding. This project will explore how GPUs can be used to accelerate the decoding of baseline JPEG images. Both the DCT and the quantisation are easily parallelizable, however, Huffman encoding is not. This project will involve understand how other researchers have used GPUs for decompressing images and also explore ways of improving on the current state of the art approaches. The interesting part of this problem is how to accelerate the decoding of the Huffman bit sequence.


A strong background in programming.

Updated:  1 June 2019/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing