A lot of people have requested a copy of my PhD dissertation, which is a hassle to obtain throught the old fashioned channels. The interest in this work stems partly from the fact that it is one of the earliest comprehensive treatments of highly scalable compression, for both images and video, including work on motion compensated scalable video coding and scalable content. Quite a few of the ideas in JPEG2000 were borrowed (not without enhancement) from this earlier work. In particular, the idea of coding subband data in independent blocks, having customizable dimensions (in space and time where video is concerned), each with its own embedded bit-stream, generated by context-adaptive arithmetic coding of bit-plane data finds strong expression in JPEG2000.
In the dissertation, you may also find a description and analysis of incremental (sometimes called rolling) subband transform implementations. The context of the discussion is temporal transformation of video, but incremental implementations of spatial wavelet transforms are clearly a special case of the more general three dimensional problem. The discussion is very limited, since the ideas were considered obvious at the time, but there has been some discussion of intellectual property in this area.
Implementations of most of the ideas presented in the dissertation were published to public news groups in mid-1994 and the implementations contain some details which are not obvious from the dissertation itself. These include context-adaptive coding of sign information during bit-plane coding of the subband samples, as well as efficient representation of code-blocks which have no significant samples at some quality layer.
Download a PDF version (about 3 MBytes)
Download scalable1.zip (2.4 MBytes)
First distribution, released to
public news groups in mid-1994
Fully scalable image and video
coder, with fractional pel invertible pan compensation for video, and fully
customizable spatio-temporal transform and code-block structure
Download scalable2.zip (3.8 MBytes)
Second distribution, released
about 1 month later
Includes 3D context modeling of
video subbands