Ed(S) is a function of all pixels in image S and encodes our first
requirement of including all semantically relevant object parts. Intuitively, Ed(S) should return a small value if all important parts
are included in the compositon and a high value if semantically important parts are missing.