Tricks I learned from imageanalyst:
Threshold the image if there is any grey-level noise. imlabel() 8-connected objects in the image. regionprops() the labeled image and go through the stats and discard any small objects (i.e., noise.) Then for each remaining region, the BoundingBox value will tell you the minimum and maximum X and Y coordinates of the region. You can extract directly from the original image using those coordinates. If you are required to slice into rectangles, you could sort the lower and upper X bounds, discard the first lower and last upper, and then average each upper with its corresponding next lower to determine the X that would slice half-way between the objects.
