Optimizing conditional code

I'm currently trying to wrestle my brain into the correct configuration for producing properly optimized assembly code, and it's really quite difficult ;). For the last couple of days I've been attempting to come up with an assembler version (vanilla, no extensions) of the following: Code:

Here's what I've come up with so far. This verion, I believe, causes the fewest instructions to be executed given that a slight majority of pixels are copied rather than processed. esi = start of row, eax = index into row, pp = start of row in previous frame, np = start of row in next frame. Code:

And here's another version: Code:

Not sure about "(vanilla, no extensions)" but how about something like: Code: