Far and away the most common way of "automatically" identifying region(s) of interest is by thresholding. It's hard for me to think of a way that doesn't do that. Basically you start with an image, then you go through a series of image processing steps (color segmentation, texture segmentation, motion detection, morphology, cluster analysis, modeling, or whatever might be appropriate), if necessary, until you finally get to an image that can be thresholded. You then get a binary image that represents your regions of interest. If you want you can then alter the regions of interest via other methods to change their size or shape or get rid of blobs in a certain size range or whatever. Then you'll have your "final" region(s) of interest and you can get measurements from it using regionprops() or other operations.
See my Image Segmentation Tutorial for a full demo: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862&sort=downloads_desc
