Select to view content in your preferred language

2009 NAIP Image contrast difference

1101
10
12-12-2010 07:27 PM
JamesFitzgerald
Frequent Contributor
Hey

Need help!

The process is...I am creating a mosaic of three 2009 NAIP images for three counties within a watershed.  For each county the watershed has three parts (northern, central, and southern).For the NAIP images, I am extracting by mask with each separate part of the watershed, then I will do another extract by mask with the county and the northern watershed.  I will also do this same process with the central and southern portion of the watershed. This gives me an image of just the watershed and not the three counties.  The problem is with one of the images that I had extracted by mask. It is different in contrast.  The other two are fine.  How do I correct the contrast so it looks the same as the other images?  Is there a fix?
0 Kudos
10 Replies
RobertBerger
Occasional Contributor
Hi,

Does all your data have statistics or is it only your one image that has statistics? Also, what version of ArcGIS are you using?
Generally, if you need to remove statistics you can create a copy and in the environmental settings of geoprocessing turn calculate statistics off. If that is indeed your issue.

Robert
0 Kudos
JamesFitzgerald
Frequent Contributor
All the images have statistics. I am using arcGIS 9.3x.

Background... I extracted a 2005 NAIP image (2X2 meter) for each of the three counties.  For each county, I extracted by mask (NAIP image) and used the county.shp as the feature.  I did this process for the other two counties.  Then I preformed another extract by mask using the watershed.shp as the feature.  This gave me a image for the entire watershed.  I then classified the image to give me a land use raster.  There was no problem processing the 2005 NAIP image this way.


Problem..For the 2009 NAIP image,  I had problems processing the 2009 NAIP image (1 X 1 meter) as I did for the 2005 NAIP image.  I decided instead of processing three NAIP images for each county I would extract by mask using the watershed.shp as feature.  The problem is the northern NAIP image for the watershed appearing different in contrast, and when I do a classification for land use, it appears different than the other two 2009 NAIP images. 

Possible solutions...I could try to extract by mask the 2009 NAIP image with each county.Then I could do another extract by mask with the watershed.shp.  The file is large but to cut down on the processing, I could use the C:\Drive instead of the external hard drive.

What do you think? Any suggestions appreciated!

Thanks for the help!
James
0 Kudos
RobertBerger
Occasional Contributor
Hi,

When you load your 2009 NAIP imagery that is unprocessed and compare the images, do you already see a difference or is the difference only after your did the clipping to the watershed?

Robert
0 Kudos
JamesFitzgerald
Frequent Contributor
I load it into arcMap and can instantly see a difference before I extract by mask.
James
0 Kudos
RobertBerger
Occasional Contributor
James,

Please see what your display stretch is for these images (and set them to none if needed) - you can do this in ArcMap > Layer Propertyes > Symbology tab.
Assuming that the stretch is set appropriately, then it seems that your images just have a different color stored in them. You can use the color correction (color balancing, matching algorithms) available in the raster catalog to make the images look seamless, but changing the pixel values ahead of time can completely mess up your analysis later.
Is your naip imagery compressed (e.g. jp2, mrsid, ecw, etc format)? If so then I'd be very careful doing analysis on this data (depending on what the purpose behind this analysis is) since the compression already alters pixel values slightly.

Robert
0 Kudos
JamesFitzgerald
Frequent Contributor
Hey Robert

I am now using the R band and have created a transformation factor. In order to create a transformation factor,  I created water, forest, urban, and bare earth shapefiles and extracted by mask using the features. Then,  I only used the forest raster, to look at the mean pixel value in properties for all three NAIP images and took the average value. The transformation factor is divided by the image with the different contrast over the middle image.  The middle image and the bottom image have the same mean value so I only used the middle for the transformation factor calculation. 

The transformation factor (0.91) is multiplied by the top image (the image different in contrast than the other two) in raster calculator.  The same calculation was performed for the middle and bottom images. This allowed me to see the range in values for the upper, middle, and bottom R-band images. 

Once I determined the range, I then went to the stretch group box and changed the statistics to min-max so I was able to adjust the values.  For the upper image, I changed it to 45-238. The middle and lower are very similar in value...24 and 26 to 238.  The images, now, are identical and have no contrasting issues.

What do you think? 

Thanks
James
0 Kudos
RobertBerger
Occasional Contributor
James,

That sounds like a lot of work! Congrats on figuring these things out.
So in the end you changed the stretch with your new values to have the images appear in the same color, do I understand that correct? This might help you when you do your classification for picking regions, but I think the classification will work on the base pixel values and not on the stretched pixel values. Again, changing the pixel values prior to your classification can introduce errors in your analysis.
If you do want to classify your stretched pixel values (rather than the original) then you can do a data export and use "Use Renderer" to change the pixel values in the output.
Hope this helps.

Robert
0 Kudos
JamesFitzgerald
Frequent Contributor
Thanks for the help!
0 Kudos
JohnNiles
Deactivated User
Robert,

I don't have anything to offer as far as a solution, but I do have a question for you!

I am attempting to extract a specific area from a 2009 NAIP raster by using 'extract by mask'. The Environment setting are set so that the 'extent' is the mask polygon and the 'mask' is the mask polygon'.

I attempted to do this with no compression and after six hours, it had created a gigantic dataset that had only processed two bands after that time... I stopped it because it I realized something is screwy.

My question for you is what do you set as compression? I looked at the properties of the original NAIP and they have "wavelet (MG3)" as the compression, which is not a choice in the 'environment settings'. The only "wavelet" option, as far as I can tell' is JPG2000.

Being that you have already successfully extracted NAIP rasters, would be so kind as to share your wisdom as to how to go about this?

Much thanks,
John
0 Kudos