shrug-l: SOLUTION: Orthoimage quality evaluation references?

Jay Johnson johnsonj-gis at
Fri Jan 18 13:18:51 EST 2008

Thanks to James Klugh of WilsonMiller, Inc for the specific solution to this issue.  
ISSUE:  Poor quality display of TIFs viewed at small scales.
PROBLEM:  When building raster pyramids ArcGIS uses nearest neighbor resampling by default.  This is the fastest method, but can create serious quality problems in images.  These problems seem particularly noticeable in highly urbanized images.  In the 1-ft resolution color images I am dealing with the problems seem specifically exacerbated in and around areas of 255,255,255 (white) values, such as rooftops and road paint lines.  The effect is much less noticeable in non-urban areas. SOLUTION:  Don't allow ArcGIS to build the pyramids by default (i.e. when you add a raster that doesn't already have a pyramid to your MXD you are prompted to build pyramids - don't build them this way).  Instead, build your image pyramids using the Toolbox, Data Management Tools, Raster, Build Pyramids Tool (ArcMap or ArcCatalog).  Within the Tool dialog, use the Environments button to set the Raster Storage Settings to Bilinear or Cubic (for my images the visual difference between these two options is minimal - YMMV).  RELATED ISSUES:  For best display quality, also make sure that your Layer Properties, Display is set to Bilinear or Cubic Convolution for the image layer.  Interestingly, if NO pyramids are built the image will still display poorly, even with CC display resampling.Changing the ArcMap Advanced Settings, Raster, Default Resampling Mode to Cubic Convolution appears to only affect the display resampling, NOT the default pyramid-building resampling.  
Changing the Environment, Raster Storage Settings (via right clicking in the Toolbox whitespace) doesn't appear to override the default behavior of building Nearest Neighbor pyramids -- thus the importance of using a Tool to do the building.
For some reason, regardless of the resampling method used in building the pyramids, the Layer Properties, Source tab of the raster layer always shows pyramids as Nearest Neighbor, even when clearly they are some other method.  I'm running an evaluation copy of ArcEditor 9.2 with no service patches installed, so YMMV on this one...  
 Have a great 3-day weekend,
Jay Johnson, GISP________________________________From: johnsonj-gis at hotmail.comTo: shrug-l at lists.dep.state.fl.usSubject: Orthoimage quality evaluation references?Date: Thu, 10 Jan 2008 09:55:05 -0800SHRUGEES,Can anyone point me to a book, article, or online reference that gets into the nitty-gritty of evaluating orthophoto quality? I'm specifically interested in looking at the histogram characteristics of an image as a guide to assessing the quality of the image or 'troubleshooting' it.The particular issue I'm looking at is some color TIFs of an urban area that look okay in ArcMap at the raster resolution (1:1200), but display very poorly when zoomed to smaller scales (1:5000). Pyramiding makes no difference. Display settings make no appreciable difference. What APPEARS to be the issue is that the white areas in the photos are VERY white. The histogram for each of the three bands in the image show very high numbers of 255,255,255 pixels. I theorize that this is throwing a wrench in ESRI's algorithm for resampling the image for display (cubic convolution). The really ODD thing is that the images look fine in PaintShop no matter what scale they are shown at...In a more general sense I'm interested in how one would quantitatively assess an orthophoto vendor's product. Surely there must be guidelines of some sort that can be applied other than the 'eyeball test'?Not sure that understanding this problem will lead to its solution, but any insights would be welcome.Jay Johnson, GISP________________________________Share life as it happens with the new Windows Live. Start sharing!<>
Need to know the score, the latest news, or you need your Hotmail®-get your "fix".
-------------- next part --------------
An HTML attachment was scrubbed...

More information about the SHRUG-L mailing list