Image Compression Performance

Engineering
1

I have encountered a situations where using gzip level 9 to compress an image is taking ~40x longer (depending on the image content) than writing the image without compression

@dzenanz I recall you did some performance analysis of various compression algorithms and level. Is this analysis still available on line some place?

Thanks!

2

I can’t find it now, but I do have some spreadsheets with the benchmarks.
zlib-ng-announcement.xlsx (22.8 KB)
zlib-comparison.xlsx (24.6 KB)

1 Like
3

Also, here are some older experiments I conducted. They compare Blosc+LZ (blosc’s default), blosc+zstd, zstd used directly, and zlib.
CBCT.tsv (1.2 KB)
mra.tsv (1.2 KB)
mraSingle.tsv (1.2 KB)
us1.tsv (1.2 KB)
us2.tsv (1.2 KB)
wb.tsv (1.2 KB)
wbPET1.tsv (1.2 KB)
wbPET2.tsv (1.2 KB)
wbPET8.tsv (1.2 KB)
wb-seg.tsv (1.2 KB)

1 Like
4

Thank you very much for sharing that. It is very useful! I see one case where zlib level 9 takes ~10x longer than level 1, but does not yield even 1% more compression ratio.

5

That’s why we chose a relatively low default compression level.

1 Like