An interesting problem

[Gareth Kirwan]

>> I think that the previous mention of a 2D autocorrelation will be  
>> good to explore, but you can use the "moving average" trick 
>to avoid  
>> lots of extra cycles while summing neighborhoods.  I've implemented  
>> this in C before and gotten speedups of 1x to 50x, depending on the  
>> neighborhood size.
>
>Aye, That's the trick. Run time from 10 mins to 16secs for the 30x100
>case. Great fun!
>
>Code available if anyone else was interested.

I certainly am, and I'd imagine most others who read the thread.