2007.04.06 20:22 "Re: [Tiff] Floating point RGBA", by Toby Thain

On 6-Apr-07, at 3:41 PM, Chris Cox wrote:

> LZW will expand floating point data, as it frequently does with 16 > bit data.

Yes & no, depending on data characteristics. My point is that dictionary compressors will work in cases (such as flat colour), where RLE will fail utterly. Without knowing what kind of data they're compressing, it's rather moot. Joris has come up with only one very specific scenario where RLE would be effective...

--Toby

>
> ZIP does a decent job with straight floating point values.
>

> But to get good compression, you need a predictor that knows about the > floating point format. (and I'm still researching better ones)

>
> Chris
>
>
>
> On 4/5/07 1:04 PM, "Toby Thain" <toby@smartgames.ca> wrote:
>

On 5-Apr-07, at 4:58 PM, Craig Bruce wrote:

That seems like a very strange combination. If I remember

>>>> correctly,

packbits is a byte-based RLE scheme. Applied to floating point

>>>> data,

compression ought to be very inefficient.

It may be a sensible option if the image has significant regions of solid black. The IEEE floating-point representation of 0.0 is:

>>> 0x00,

0x00, 0x00, 0x00 (in either endian).

Dictionary compressors are much more likely to win, overall. RLE does seem an odd choice, regardless of the actual data characteristics.

--Toby

>
> --------------------------+----------------------
> +--------------------------

Dr. Craig S. Bruce        | Ph 819-771-8303 x205 |              CubeWerx Inc.
Senior Software Developer |   Fax 819-771-8388   |  Gatineau,  Québec, Canada

> csbruce@cubewerx.com      |  http://csbruce.com/ |  http://

> www.cubewerx.com/
> --------------------------+----------------------
> +--------------------------

"Hard work may not kill you, but why take chances?"

>