Fix canny

[rolalaro]

Improved the ViSP Canny implementation by :

• changing the edge thinning algorithm
• considering only the gradient points having a value greater than the lower threshold

To test with the result of Canny by OpenCV beside the ViSP implementation result:

./tutorial-canny -t -1 -1 -r 0.4 0.95 -b opencv-backend --image

Here is a dataset that can be used to have a look at the capabilities of the new implementation:
testCanny.zip

[fspindle]

@rolalaro Seems ok to me. I have just the feeling that using -r 0.4 0.85 gives results much closer to the one obtained with OpenCV backend.

Could be seen especially on cube.ppm image. On mire.png the values are more around -r 0.4 0.9.

Anyway, I will merge.