Title :
Robust still image coding using lapped transforms with block classification
Author :
da Cunha, Arthur L A ; Finamore, Weiler A. ; da Silva, Eduardo A B
Author_Institution :
Center for Telecommun. Studies, Pontificia Univ. Catolica do Rio de Janeiro, Brazil
Abstract :
Very efficient still image compression methods exist today, but most of them are not fit for transmission on error prone channels. One example of state-of-the-art robust image coding is the use of wavelet compression followed by forward error correction, as proposed by P.G. Sherwood and K. Zeger (see Proc. IEEE Data Compress. Conf, p.72-81, 1997); that method gives high coding efficiency but suffers from the incomplete decoding problem. An alternative to this is the joint source/channel coding scheme by Qing Chen and T.R. Fischer (see IEEE Trans. on Img. Proc., vol.7, no.4, p.496-505, 1998), which overcomes this difficulty at the expense of a lower PSNR performance. We propose a joint source/channel coding scheme with increased coding efficiency based on lapped transforms with block classification. With an efficient bit allocation strategy and the high coding gain of lapped transforms, a significant increase, of the order of 2 dB, in the PSNR versus BER performance has been obtained over the classical joint source/channel scheme on binary symmetric channels.
Keywords :
combined source-channel coding; error statistics; forward error correction; image classification; image coding; quantisation (signal); telecommunication channels; transform coding; transforms; BER; FEC; PSNR; bit allocation strategy; block classification; coding efficiency; error prone channels; forward error correction; image compression; incomplete decoding; joint source/channel coding; lapped transforms; robust quantization; still image coding; wavelet compression; Bit error rate; Bit rate; Decoding; Digital images; Forward error correction; Image coding; Information theory; PSNR; Phase noise; Robustness;
Conference_Titel :
Image Processing. 2002. Proceedings. 2002 International Conference on
Print_ISBN :
0-7803-7622-6
DOI :
10.1109/ICIP.2002.1038006