BIM (a new lossless image compressor) is here!

[kampaster]

Matt, thanks for testing!
New results
I used preprocessing:
No code has to be inserted here.

Once again - BIM was not designed for such images. BIM is for true color pictures from your camera...

Badly.
The model for artificial images is necessary!
Update 31,08,13: rev+BWT

[kampaster]

Update 31,08,13 21:32 +MTF

[encode]

BTW, try BIM v0.01 on this file!
Keep in mind, BIM was deigned to be simple and fast - and multiple CM models is out of this conception. Anyway, BIM v0.03 is much more complex, yet is a fast enough - will check later how it performs on such files...

[Piotr Tarsa]

LZ4 -c0 compresses the file to about 1 MB. Some very low complexity model for repetitions could be added to an image codec but overall I think a codec can't be (ie it is infeasible to meke it like that in near future) really fast, efficient and universal at the same time. If we try to optimize for repetitive images then what's next? QR codes? Screenshots?

[encode]

BIM v0.03 is here!

Slightly improved predictor + SLOW counters. Enjoy anyway!

http://compressme.net

http://compressme.net/bim003.zip

[kampaster]

BTW, try BIM v0.01 on this file!
Keep in mind, BIM was deigned to be simple and fast - and multiple CM models is out of this conception. Anyway, BIM v0.03 is much more complex, yet is a fast enough - will check later how it performs on such files...

OK
Results on AIMP_free.tga:

No code has to be inserted here.

New file: http://nishi.dreamhosters.com/v/Face_mosaic.7z
No code has to be inserted here.
_
No code has to be inserted here.

[encode]

Now please test BIM on text files! It is possible, just add a fake PNM header to a text file.

[Bloax]

Alright then. ([spoiler]s would be a godsend at this point)

PNG: 90,039 bytes
BIM (v0.03): 67,438 bytes


PNG: 453,842 bytes
BIM (v0.03): 384,943 bytes


PNG: 98,207 bytes
BIM (v0.03): 299,607 bytes


PNG: 69,948 bytes
BIM (v0.03): 124,028 bytes


PNG: 39,746 bytes
BIM (v0.03): 303,106 bytes

BIM seems to win over what little truecolor PNGs I have without breaking a sweat (but that's no surprise considering PNG is pretty bad for such things), but the losses on paletted PNGs are pretty colossal.
oh and here's an archive with the bim/png/pnm files, along with bim.exe v0.03

[Skymmer]

New file: http://nishi.dreamhosters.com/v/Face_mosaic.7z

Result of UHARC on this file can be much better.
17 361 742 srep64 3.92 -m4o + uharc -mx -mm+ -md32768

[encode]

Face_mosaic.tga
ULZ v0.03 -> 128,350,179 bytes
CRUSH v1.00 -> 74,538,342 bytes
BCM v0.14 -> 27,383,994 bytes

And this is the "world95.txt" file compressed with BIM v0.03:

[kampaster]

Result of UHARC on this file can be much better.
17 361 742 srep64 3.92 -m4o + uharc -mx -mm+ -md32768

17 177 903 mm+rep+uharc

Now please test BIM on text files! It is possible, just add a fake PNM header to a text file.

If everything is correct...
enwik8
BIM v0.03 70 801 674 bytes

And this is the "world95.txt" file compressed with BIM v0.03:

I am sure, BIM has the hidden key

[Piotr Tarsa]

If everything is correct...
enwik8
BIM v0.03 70 801 674 bytes

Not that bad. Now we can test FLAC on executables.

[m^2]

And this is the "world95.txt" file compressed with BIM v0.03:

LZRW - like result. After you decided to show it to us I expected more...

[encode]

No surprises. Despite CM-backend, BIM has relatively simple, non-adaptive linear predictor, that simply transforms a text file into noise...