If both images were not taken with the same camera, or if the camera setting changed (you mentioned one was distorted, not the other), then I would say both images need to be calibrated and corrected before you pass them to the algorithm.
Vision Development Module provides 3 ways of learning a calibration:
1) Simple Calibration, during which you just specify a correspondance between pixels and real-world distances. i.e 10 pixels represent x millimeters. This simple method does not take into account distortion. This would not work in your case. This would work, say if you acquire images with a telecentric lens and you know there are no distortion.
2) If you still have the settings to acquire your images, then yes, printing a calibration grid and using it to acquire an image of it in the same condition your original image was acquired, and using the grid of point to calibrate the image is the method that will give you the most accurate results.
To learn the calibration using this method, use IMAQ Learn Calibration Template, provide the grid of points image (which should be distorted the same way as your image) as the input image.
Provide the Grid Descriptor, which defines the real-world distance between the points, and threshold parameters to threshold the grid.
In addition, the VI lets you define a Calibration Axis.
Once the grid image has been calibrated with that VI, use IMAQ Set Calibration Info to apply that calibration to other images acquired under the same conditions.
3) If you can't reproduce the acquisition conditions in which the distorted image was acquired, you can still calibrate that non linear image, if you can locate multiple points in the image and can associate them with real-world coordinates. You do that by providing "Reference Points" to IMAQ Learn Calibration Template. (In this case, don't connect Grid Descriptor).
I hope this clarifies things.
Christophe
