I understand how overlap-add works for 1-dimensional signals, but I need to do something similar in 2 dimensions. The paper at this link covers the 1D case pretty well (see p29):
The right-hand columns indicate overlap amount and expected accuracy.
Is there anything similar for 2D? I'd like to break up a large image, process each sub-square separately, then merge the results without generating a lot of seams and square-looking artifacts. The end-result is not super-critical, but I would like to find some type of overlapping windowed system to avoid sharp transitions between pixels on boundaries.
BTW, I'm not doing FFTs. I'm trying to average the luminance in each square of pixels, then multiply a 'weighted window' by that average, then re-add all the squares. There's more to it, but the intent is a sort of filter process.