Matrix image regression or regression of subscales A person is given 10 sample images (based on the same image) with corresponding techniques on how the image was achieved. I want to regress the matrix bitmap of the image of the 10 images on the image created by the person as a gross way of knowing which images the person copied from. I am assuming that if the person copied one particular person, they will have more corresponding pixels than other people.
Is there such a thing as matrix regression? Can anyone point me to other possible techniques?
I also have a rating scale of the features present in each image. With a similar intent, I would like to regress the subscores of the rating scale of the sample images to the subscore of the particular person's image. I imagine the same regression technique could be used.
Hoping someone out there can point me to a technique that would let me do this.
 A: One image becomes another, different image by a transformation.  There are a large number of transformations possible.  They are in fact infinite given they are real valued.  There are a large number of general forms.  
You could add a constant.  That would displace the intensities.
You could translate in x and y.
You could rotate.  Although it is a 2d image you can treat the intensities as heights and thus treat the image as a surface in 3-space.  You can then rotate it on any of the 3 axes.
These are "rigid" transforms.  You could bend or warp the image in any number of ways.
For some convenient and intuitive transforms I like to use 3d homogenous affine transforms. In the homogenous transform there is an additional dimension whose coordinates are unity.  This allows translations to work easier.
I have a presentation on the topic of bottom-up exploration of transformations.  If there was a way to make it accessible on the internet, I would like to do that.  
Bottom line: 3d homogenous affine transforms might do very well.  You can treat it like a least squares problem with pseudo-inverse solution.
