@Article{Supelec461,
author = {Hocine Bekkouche and Michel Barret and Jacques Oksman},
title = {{Adapted generalized lifting schemes for scalable lossless image coding}},
journal = {Signal Processing},
year = {2008},
volume = {88},
number = {11},
pages = {2790-2803},
month = {novembre},
url = {http://hal-supelec.archives-ouvertes.fr/hal-00287898/fr/},
abstract = {Still image coding occasionally uses linear predictive coding
together with multi-resolution decompositions, as may be found
in several papers. Those related approaches do not take into
account all the information available at the decoder in the
prediction stage. In this paper, we introduce an adapted
generalized lifting scheme in which the predictor is built upon
two filters, leading to taking advantage of all this available
information. With this structure included in a multi-resolution
decomposition framework, we study two kinds of adaptation based
on least squares estimation, according to different assumptions,
which are either a global or a local second order stationarity
of the image. The efficiency in lossless coding of these
decompositions is shown on synthetic images and their
performances are compared with those of well-known codecs (S+P,
JPEG-LS, JPEG2000, CALIC) on actual images. Four images’
families are distinguished: natural, MRI medical, satellite and
textures associated with fingerprints. On natural and
medical images, the performances of our codecs do not exceed
those of classical codecs. Now for satellite images and
textures, they present a slightly noticeable (about 0.05 to 0.08
bpp) coding gain compared to the others that permit a
progressive coding in resolution, but with a greater coding time.}
}