In my quest for optimization I have looked at convolution in matlab. There is a complete function "conv2(matrix,kernel,'same')" that does exactly what imaq convolute does. This works a lot slower than labview however - congratulations ni, you are faster at matrices than matlab!!!
There is one thing though. The convolution should be much faster if the matrix consists of many zeros - a zero basically means that a convolution "around" that specific element is not needed. This is clearly seen in the computation time in matlab where the convolution time is greatly reduced when many elements are zero (at least a factor of 2).
In Labview however the imaq convolute takes the same time no matter what image runs through - even if it is an image of only zeros!
I would therefore expect that a the convolution i labview could be greatly optimized by taking advantage of elements with value zero.
In my images of size 575*575 = 330,625 pixels, there are usually 100 particles, taking up about 10,000 pixels. The remaining 320,000 pixels usually have values below 5 (on a U8 image), a value that come from random noise. That noise I can easily remove by applying a threshold. As you can see im greatly interested in a vi that uses this advantage in a convolution.
So my question is: is there a vi that can take advantage of zeros in the input matrix, or is it possible to modify the convolute vi?
Thanks
Simon
