I repeated the experiment with larger files. Very interesting. At the
500 MB or larger size, JPEG2000 rivals ECW in speed, and the speed
difference from TIFF diminishes.
This seems to bear out a theory I had that TIFF is best for images up to
1 GB, but above that, a wavelet-based compressor is more appropriate as
the disk seeks become more expensive than the CPU time. I can't test
with input larger than 500 MB, due to ECW's licensing restrictions, so
if anyone does have a licensed unlimited-size ECW compressor, and cares
to repeat this experiment, I would love to hear about it.
*** SETUP
The data is USGS DOQQs (black-n-white) of San Francisco, California.
Several DOQQs were downloaded and merged into a single GeoTIFF using
gdal_merge.py. Then gdal_translate was used to create ECW and JPEG2000
equivalents.
All images use the same projection: EPSG 26910, aka UTM zone 10N with
NAD83 datum.
The spatial extent observed is:
543577.000 4150151.000 555545.000 4192431.000
A mapfile is created specifying three layers, named TIFF, ECW, and
JPEG2000, each one reading from the corresponding raster. No
reprojection is being done in the mapfile.
*** STORAGE SPACE
TIFF, 483 MB
ECW, 98 MB
JPEG2000, 114 MB
Obviously, both ECW and JPEG2000 provide significant benefits over TIFF
as far as storage space.
*** RUNTIME PERFORMANCE
# full view of the region
shp2img -m mapfile.map -l ECW -o ecw-fullview.png
Time: 3.076s
shp2img -m mapfile.map -l JPEG2000 -o jpeg2000-fullview.png
Time: 3.075s
shp2img -m mapfile.map -l TIFF -o tiff-fullview.png
Time: 0.480s
# one square kilometer. that's 1/480 of the full square area
shp2img -m mapfile.map -l ECW -e 549561 4150150 550561 4151150 -o
ecw-1km.png
Time: 3.077s
shp2img -m mapfile.map -l JPEG2000 -e 549561 4150150 550561 4151150 -o
jpeg2000-1km.png
Time: 3.073s
shp2img -m mapfile.map -l TIFF -e 549561 4150150 550561 4151150 -o
tiff-1km.png
Time: 1.065s
As with the experiment with smaller files, we see that TIFF is a
significantly faster format than the other two. However, the time
difference has shrunk from 4:1 to about 1.7:1
Strangely, JPEG2000's speed is not the same as it was with the other
experiment, despite repeated checks and re-runs. At the 500 MB file size
range, JPEG2000's speed closely rivals that of ECW -- JPEG2000 seems to
run faster on this trial, by some margin. (no reprojection in either
experiment, very odd)
*** IMAGE QUALITY
https://www.cartograph.com/~gregor/formats2/ecw-fullview.png
https://www.cartograph.com/~gregor/formats2/jpeg2000-fullview.png
https://www.cartograph.com/~gregor/formats2/tiff-fullview.png
https://www.cartograph.com/~gregor/formats2/ecw-1km.png
https://www.cartograph.com/~gregor/formats2/jpeg2000-1km.png
https://www.cartograph.com/~gregor/formats2/tiff-1km.png
As with the other experiment, the wavelet-based ECW and JPEG2000 formats
lost some degree of fine detail. However, this loss of detail manifested
gracefully as a "smoothing" effect of otherwise grainy features.
