Method for overlapping images added to Cross-Correlation (Phase Correlation) example.

[paalge]

This adds to the Cross-Correlation (Phase Correlation) example by showing how to overlap two images once the shift has been calculated.

[emmanuelle]

You should convert to float before taking the difference, using img_as_float, otherwise overflows can happen.

[emmanuelle]

Thanks for this addition to the example. If you're plotting the difference between the orginal and "back shifted" image, then it would be interesting to use a different shifting method to build the shifted array. Indeed, here we use both Fourier transform to shift forward and back, resulting in almost no difference. If you're using another method, like spline interpolation with ndimage.shift, then you can see that the error is greater for edges, which is quite interesting I think.

[emmanuelle]

Also, figures showing the grayscale images should use cmap='gray'. Could you please change this?

[paalge]

Thank you for the comments. I've changed the cmap and the overflow issue. As for using another method I'm not sure if it should be added here or to a new example comparing the different methods for shifting?

[emmanuelle]

@paalge I'm really not an expert on methods for shifting, so I don't know if it's worth an example (if you think so, go for it!).

[paalge]

I'm no expert on shifting algorithms either. The addition to this example added in order to clarify how you can use the calculated shift.