Feature detection of edges

edit

Align the images by using local feature correspondences on the original image. I doubt that an automatic alignment of the canny edge images is possible.

Avoid canny where possible as it does not yield stable edges. If you need an edge image then just display/use the plain Sobel filter output.

What I would suggest is to apply one of the feature detectors on the image that can be found inside the feature2d library: http://docs.opencv.org/modules/features2d/doc/features2d.html

Basicly use the feature points extracted by SURF, SIFT, ORB, FREAK, BRISK, ... to match both image by searching for matching feature points. The discriptor matching functionality of openCV will help out there. These descriptors create a unique representation of the area around the feature points and try to retrieve them.

By using these feature points, and looking at local features, you do not have to align both images.