The BCIF image compression algorithm

Links

BCIF whitepaper

Imagemagick converter (integrates with the BCIF GUI to convert from other formats than BMP)

Homepage of the author

Other image compression algorithms

The PCIF algorithm

The Jpeg2000 format

Kakadu: a Jpeg2000 implementation

The LOCO (Jpeg-LS) algorithm

Char-LS: a Jpeg-LS implementation

The PNG standard

Benchmarks

The Waterloo image set

The Kodak image set

I.C.I. image set

Squeezechart: compression benchmarking

The BCIF compression algorithm has been compared with the main lossless compression standards: JPEG 2000 (implementation from Kakadusoftware), JPEG-LS (reference implementation, after color transform R - G, G, B - G) and the PNG standard (files created with an optimized PNG encoder). The compression results of the PCIF algorithm have also been included.

The benchmarking sets are formed by some well known images that have been proposed for the evaluation of image compression algorithms. The Waterloo image set is the oldest one, and it contains three computer generated images and five photographical ones, comprehensive also of the famous Lena image; the filesize of the contained images varies from 512x512 to 1118x1105 pixels. The Kodak image set contains 24 photographical images of size 768x512 or 512x768. Update: I've been recently informed that the Kodak image set, contrarily to what affirmed in some websites, originate from a lossy compression: for further benchmarks, it would be better to use other sets. Finally, the imagecompression.info image set is the most recent at it has been proposed with the goal to test compression algorithms on high resolution images, large up to 7216x5412 pixels. All images are publicly available on the respective websites.

The comparison here takes into account only some well known standards, and there are more recent algorithm that may give better results: for a comparison with a broader range of research algorithms, the author recommends to check some third party benchmarking charts, of which Squeezechart is an excellent example.

The sum of the size of the compressed files produced by the various algorithms has been compared to the one obtained with BCIF, and the results can be seen in the following figure. BCIF outperforms all of the other competing standards, resulting slightly worse than another algorithm (JPEG 2000) only in one image set (the Kodak image set). The major advantages derive mainly, but not only, from computer generated or high edged images where the BCIF algorithm performs very well. Details on the compression results on every single file can be found below. Those who desire to test the BCIF algorithm on other images are encouraged to download BCIF to give it a try.

BCIF lossless image compression benchmarks

In the following table, the same data is shown in numbers, and it is also compared to the uncompressed filesize of the original images (BMP column).

All image sets
Image set BCIF PCIF JPEG-LS JPEG2000 PNG BMP
Waterloo 100% 119.0% 125.7% 180.4% 126.8% 383.1%
Kodak 100% 103.3% 100.6% 98.6% 135.1% 248.4%
I.C.I. 100% 104.6% 104.5% 104.5% 122.9% 236.1%
Average 100% 109.0% 110.3% 127.8% 128.3% 289.2%

Below, the resulting filesizes of the compressed images for the Waterloo image set. BCIF results to be a very efficient algorithm thanks to the presence of several artificial sharp-edged images. For the same reason JPEG 2000 obtains a very bad global filesize, as it is not suited for such images.

The Waterloo color image set
Filename BMP BCIF PCIF JPEG-LS JPEG2000 PNG
clegg21003785716461369490
frymire36214136417961560361
lena768414431442434500
monarch1152450468448431621
peppers768331343328327441
sail1152541554541511777
serrano1463152242286623147
tulips1152500527507477691
Total 12178 3179 3782 3997 5736 4031

In the following table, the compression results for the Kodak image set. Since images are very similar in nature, origin and size, there is no big difference between the results on the various images. The best algorithm results to be JPEG 2000, with an advantage of a 1.5% over the second best algorithm, BCIF. Very close results are also obtained by the JPEG-LS format.

The Kodak image set
Filename BMP BCIF PCIF JPEG-LS JPEG2000 PNG
Kodak011152506516511498719
Kodak021152452467449439603
Kodak031152380398374388491
Kodak041152453463455437622
Kodak051152535559547519767
Kodak061152462477464460604
Kodak071152404426405408553
Kodak081152534553554534769
Kodak091152431443437427569
Kodak101152438450444434579
Kodak111152452464448446606
Kodak121152412420401415518
Kodak131152591604601569803
Kodak141152506520503487675
Kodak151152424443426431598
Kodak161152421432417421521
Kodak171152442454437435587
Kodak181152558565562516762
Kodak191152479490482463655
Kodak201152350375362387480
Kodak211152484495480468622
Kodak221152513524517483685
Kodak231152419434416407544
Kodak241152483508494488689
Total 27649 11129 11491 11199 10970 15033

In this last table, the considered compression algorithms are compared on the imagecompression.info image set. In this case BCIF obtains the best result for almost every image, proving itself to be a very efficient algorithm also for high resolution images.

The imagecompression.info image set
Filename BMP BCIF PCIF JPEG-LS JPEG2000 PNG
artificial1843211711550178322521611
big_building1144135011053343545105088564369
big_tree811534288444287437184450651419
bridge326131827718637184551847620962
cathedral1762578738280831582539840
deer312891994520244204232079521732
fireworks2160939844434410748065670
flower_foveon1004622812355255126823134
hdr1843255035724566258317072
leaves_16001762596291012610210989711805
leaves_20017625773582108630813010331
nightshot_1002160958166185592965657701
nightshot_1600216091131611879113501185413275
spider_web35511810482348050843910228
Total 459593 194628 203494 203698 203377 239156