Horn-Schunck is an algorithm; the size of the output is going to depend upon the implementation.
If you use http://www.mathworks.com/help/vision/ref/opticalflowhs.estimateflow.html then what is returned is an "optical flow object". The size of "objects" is not specified. By examining some of the surrounding documentation I can see that optical flow objects have some fields to hold the options that you selected, and possibly some information describing the system. They also hold the u and v arrays that indicate direction and magnitude for each point of the input image; those are each going to be double() with the same number of rows and columns as the input image.
