ArcMap 10.2.2 Mosaic Dataset -- Compression/Resampling Question

[PhilipGriffith]

I've created a NATURAL_COLOR_RGB mosaic dataset using all of the default options for adding rasters, as well as for defining and building overviews. The source rasters are 8-bit unsigned MrSIDs in one of two projections and range in cell size from 0.32 ft to 1 ft; the mosaic dataset is in yet a third projection.

When I view the imagery at a scale of 1:150, I see the image below:

When I identify the image, ArcMap tells me that I'm looking at the source imagery.

However, when I load the source imagery, I see the image below:

Clearly these two images are not the same.

The first image has been smoothed to the point that I can't differentiate pixels even at a scale of 1:1. The second has clearly delineated pixels that measure 1ft x 1ft, which is the image resolution.

What is the mosaic dataset doing to the image?

My intuition is that the mosaic dataset is using the default resampling method (in this instance, Bilinear Interpolation) to smooth the image, but when I look at the help it mentions that the resampling method is used to display the mosaicked image. Does the source image count as part of the mosaicked image? Changing this option does not alter the display in any noticeable way.

A second thought is that it has something to do with the cell size tolerance factor, which is currently set at the 0.8 default level.

The original images have a MinPS value of 0. Could this be causing the alteration? If so, how should I alter the value to stop this behavior?

I previously set the cell size tolerance factor to 999 as mentioned in the best practices here, but it resulted in unexpected behavior, so I've shied away from this setting.

Is this actually another image?

Disk space is a concern for me, so I would like to create a slim mosaic dataset, if possible. I'm willing to give up pretty, smooth imagery in return for a smaller memory footprint and (potentially) faster transmission speeds. If it is another image, how can I get the mosaic dataset to stop this behavior?

Thanks for any and all help in this matter. Of course, if the smooth imagery actually consumes less memory and allows the mosaic dataset to run faster than it would with the unsmoothed imagery, I'm all for it. I just don't understand how it couldn't increase processing time, which is something I'd like to keep to a minimum.

[CodyBenkelman · Accepted solution]

Philip

Check the properties of the image layer in the Table of Contents (e.g. under Mosaic you'll have boundary, footprint, image - right click image --> Properties --> Display tab).  Your first view above (of the Mosaic Dataset) is using bilinear interpolation resampling, the second screenshot (the Sid image) is using nearest neighbor resampling.

If you change the properties of the MD in Catalog such that default resampling is nearest neighbor, you'll have to remove the mosaic from the TOC and then reload to see the change.  Most users prefer the smoother appearance, so bilinear interpolation is the normal default.

Regarding performance, you should not be able to discern any difference between the two settings, although if you have a VERY large display (or in a few other scenarios) the nearest neighbor resampling is slightly faster.  Neither will affect your disk space - you're doing the right thing by leaving the images as-is and accessing them via the Mosaic Dataset - it is a tiny data volume compared to the imagery.

The MinPS value controls when the imagery appears or disappears.   0 means "no minimum pixel size" i.e. you can zoom in to cm resolution to see really big pixels.  If you set minPS=2, the imagery will disappear if you zoom in past 1:8000 (pixel size ~2 meters).

I'm glad you found the Image Management Guidebook!  It's dense reading but I hope you find it helpful.  If you start scaling up to larger image collections, be sure to check our landing page for new workflows, and follow the link from that page to this Group on ArcGIS Online for workflow-specific example scripts

Cody B.

[PhilipGriffith]

Hi Cody,

Thanks for the quick response! The dataset is for the entire state of Florida. I'm not sure that this counts as very large, but I'm more worried about some of our user's slow network connections when they're out at a research station or state park, so any advantage in speed is a plus.

The guidebooks have indeed been useful. I'm interested in time-enabling our imagery, as well as creating a statewide DEM layer, and am sure they'll continue to be helpful.

Thanks again,

Philip

[CodyBenkelman]

Philip

I'm glad my response was helpful. 

Before you build the statewide DEM, be sure to look at the AGOL World Elevation service (start at http://www.esri.com/elevation  or go to AGOL and search for GROUPS using text "elevation layers").  That should provide a good base that you could build on; the Guidebook and AGOL sample group mentioned in my post above include details specific to elevation and lidar...

Best wishes

Cody B