linear regression
ERDAS Communities - Automated Email
linear regression
By John in ERDAS Software Forum
Hi,
I have been trying to do a linear regression analysis between 2 different layers in 2 images. I found a linear regression model (.gmd) from the ERDAS help pages (http://www.erdas.com/search.aspx?IndexCatalogue=Entire+Site&SearchQuery=linear%20regression), but it only allows for the uploading of 1 file. I'm a little confused about it.
I am ultimately trying to do a linear regression analysis to solve the k values (Minnaert constants) for a Minnaert topographic normalization. I took advice from someone on how to do the process;
The Minnaert law has only a single parameter, so
the way I like to use is to compute simulated images from the shape
model with different Minnaert k values (0.0, 0.1, 0.2, ...). Then I
do a linear regression of the image on the simulated image at each
k value and look at the residuals. The result should be a curve with
a clear minimum at the best estimate of k. A nice feature of this approach is that it can take care of atmospheric haze. The additive
term in the linear regression equation corresponds to light added
uniformly to every pixel by haze, regardless of what the surface is
doing. The slope coefficient corresponds to the brightness of the
surface compared to your model. If you are working with a body
with no atmospere then you need to do a regression with a constant
term of zero, corresponding to no haze contribution.
Is anyone familiar with this process? Or, could anyone provide me with how to do a linear regression analysis between layers in two different images? Thanks a lot!
John
the way I like to use is to compute simulated images from the shape
model with different Minnaert k values (0.0, 0.1, 0.2, ...). Then I
do a linear regression of the image on the simulated image at each
k value and look at the residuals. The result should be a curve with
a clear minimum at the best estimate of k. A nice feature of this
----------------------------------
You were sent this email because you opted to receive email notifications when someone created a new thread, or replied to a topic within this forum, ERDAS Software Forum.
To unsubscribe:
- To unsubscribe from this item only, visit the item link and turn off notifications.
- To stop all notifications of any updated items from ERDAS Communities, go to Edit Profile, click on the Site Options tab, then under Email Configuration, set Send Notifications to "No".
ERDAS Communities - Automated Email
Re: linear regression
By Al Kirschbaum in ERDAS Software Forum
Hi John-
After reading the description on the model I see that you will have to take each of the layers of interest out of your images, then stack them together into one image. Then you'll just type the layer number into the X and Y scalar objects, which are n46_Integer and n47_Integer, respectively. I've copied the text from the model description which led me to this conclusion below.
"The model regression_tables.gmd does this with two tables (lists) of variable values. Based on the input tables, it will calculate the a and b coefficients and The model regression_imagery.gmd does something very similar, but uses two bands of imagery as the inputs. The way the model is written, the two bands must come from a single image, so the bands may have to be layer-stacked into a single image first, but the model could be modified to take multiple input images. Which bands to use are specified by typing a layer number into the X and Y Scalar objects. The regression coefficients a and b are written out to the Session Log again."
Cheers
-al
----------------------------------
You were sent this email because you opted to receive email notifications when someone responded to this thread.