pattern detection

[Alain_S]

Hi Zwired1,

I run the "gradient histogram.vi" code with the data sets I showed previously and put a picture of the result under each data set, see attached document. Now I just need to understand what the results mean.

After reading  this Wiki page some things getting clear, but still not 100% 😞

Since in all three examples the maximum of the histogram is around 133/135 and the only constant in all data sets are the vertical lines, I assume these two belongs together.

But then I'm not clear about the location of the minimum 92, 177 and 88.

Am I right when I say that when using 180 "bucket's" the interval between 2 histogram values is 2°?

Anyway, thanks for assistance so far.

Best regards

Alain

[Zwired1]

<hand waving>

convolution basically takes one function and "sweeps" it across another.  It multiplies the first function by the second and adds up all of the results, shifts one of the functions one time step (pixel) and repeats the process.  You'll get a big resultant wherever the two functions are similar (imagine constructive interference even though that's not exactly correct). If you look at the Sobel operators columns they are low-med-high values and so after convolving these with your image the areas in your image with similar characteristics (i.e. edges) are highlighted. 

</hand waving>

You are correct, with 180 bins, the x-axis step size is 2 degrees. I usually use 360 bins so I don't have to think.  The peaks at n*90 degrees are coming from your highly oriented noise.  I'd be looking at the non-90 degree peaks and I expect them to be much lower in amplitude.

The VI applies the Sobel operators separately in X and Y.  It then takes the ATAN2 of the ratio of these magnitudes for each point in the image and makes a histogram of the result.  Because there's no magnitude information in the ATAN2 output, every point has an equal contribution to the histogram, regardless of the "strength" of the edge feature in that location and that may be a factor in your case.

I played around a bit with your data (screen shots of the screen shots in your Word doc) and I had some promising results with inverting the signals.  " max(signal)-signal  ".  Because the noise is so highly oriented, I might try to do a low pass filter in the x direction before finding the gradient.  I'll look at this some more when the boss isn't lurking.