Processing an image for analysis

[Rule]

I'm wanting to process images of textures (e.g. repetitive patterns, like brick walls), for a regression problem.  The issue, most of the images are at least on the order of 2000 x 2000 pixels, or 4 million datapoints.  If significantly cropped, maybe 400x400, 160000 datapoints.  I need something more on the order of 1000 datapoints for the regression method I'm using. 

So I'm not sure if there might be some nice compression algorithms I can use, to decrease the number of pixels.

Let me know if you have any ideas, ASAP. (PM or reply to this thread).  If you have experience with Matlab, that's a plus.

Thanks