For the first, eliminating the current cell:
A = 1:7;
B1 = movsum(A, [0 2]);
B2 = A; % Technically movsum(A, [0 0]) but that's just A itself
C = B1-B2
For the second, eliminating the current and next cell:
D = 1:7;
E1 = movsum(D, [1 2]);
E2 = movsum(D, [0 1]);
F = E1-E2
If you want more complicated patterns, treat the windows as sets and use the inclusion-exclusion principle to determine which movsum calls you need to add or subtract. For instance, to get a window like:
pattern = [1 0 1 0 1]
where 1 means it's counted and 0 means it's not, you can write this as:
[1 1 1 0 0] ... % movsum(A, [2 0]), two previous points plus current point
- [0 1 1 1 0] ... % movsum(A, [1 1]), one previous point, current point, one next point
+ [0 0 1 1 1] % movsum(A, [0 2]), current point plus two next points
This is
A = (1:10).^2;
B = movsum(A, [2 0]) - movsum(A, [1 1]) + movsum(A, [0 2])
As a spot check, according to the pattern B(5) should be A(3)+A(5)+A(7) and it is.
