Problem loading rasters into ArcSDE 9.3.1 Oracle

[haifaAl_Yazeedi]

Hello,

We are having problems loading raster images into ArcSDE.

We successfully imported a land sat image (MrSID format, 3 bands, 230 MB size) into ArcSDE (with pyramids, statics and 128x128 block size). The image appears and is well georeferenced in ArcMap. BUT the quality of the image is not good. It seems it has been damaged by the loading process.

Any idea?

Thanks

[haifaAl_Yazeedi]

I have tried the same workflow with a different MrSID raster - same problem. I then tried with a few TIF rasters - no problem.

Could there be an issue with MrSID format? It loads the raster but I get a kind of snowing effect. Image distortion?

[VinceAngelo]

It is generally not recommended to use lossy-compressed image sources for loading into ArcSDE.
You can do so, but you must be careful on which compression algorithm you use in the database,
and which algorithm is used in the pyramid (they can be different).

What algorithms are you using? With what parameters? What pyramid interpolation method
did you choose?

Is the image quality issue only with pyramid tiles, or at the base resolution as well.

- V

[haifaAl_Yazeedi]

I used the defaults parameters/algorithms - including 'build pyramids' (it created level 8), Nearest Neighbor resampling, LZ77 compression, 128x128 tile size.

I indeed noticed that when I "zoom to raster resolution" the quality is ok. Then if I zoom out a few times, the "snowing effect" arrives and increases as the area increases.

I also found the below bug. This is for SQL Server but I wonder if my issue could be related to this bug as well. I tried to load my MrSID raster into a raster catalog - but the quality issue is the same...
http://resources.arcgis.com/content/nimbus-bug?bugID=TklNMDM3MTAz

[VinceAngelo]

Never, never, never use LZ77 compression for a lossy source image.  If the source was
MrSID or JPEG2000 use JP2 compression (otherwise you increase image size by a factor
of 10-40 [yes, 1000% to 4000% storage increase], with no benefit). 

NEAREST is most likely to increase "snow" since it doesn't try for a best fit (as a rule,
it should only be used with colormapped rasters).  You need to experiment with BILINEAR
and BICUBIC (which use weighted averages of neighbors to better blend the resampled
pixels), to see which produces better pyramids with your images.

- V

[haifaAl_Yazeedi]

Yes. This solves the issue. Quality is perfect with JP2000 compression and BILINEAR resampling.

Great support. Thanks.