Locate fiducial marks in old aerial photos
I need to locate the fiducial marks in a huge set of old aerial photos. The fiducial marks are located along the borders of the photo and I know very roughly where to look for them:
The search area is thresholded to remove any actual image content (noise). Each photo contains 4 fiducial markers and the following shows all 4 of them joined together in a small image:
As seen in the image the markers vary slightly in shape and my initial tests with a SURF detector and a BruteForceMatcher didn't go well. This might-be/is-probably due to my lack of knowledge about which detectors/matchers to use and their corresponding parameters.
I have thought of using ICP in 2D on contours but I don't know how to combine those and my reading suggests that ICP might no be the right way to go. I have gone away from template matching due to the resolution of the images and the computational time required to find match the template.
I am fairly inexperienced in this field but can't help thinking that these shapes are so distinct that it should be fairly easy somehow to locate them. Can you help me along the way?
your image does not show. could you try again ?
I see the image and I have tried from another PC and it displayes fine there too.
I see the image, too; but have no idea, maybe shape matcher?
probably just my stupid brower.
@Chau, could you clarify, if the 4 small dots close to the center are your actual markers ?
@berak, the four small white dots are the points I'm ultimately looking for. The black areas are the markers and my thought is that if I can locate the black marker then I hope it will be easier to find the small white dot. Does that make sense?
sure makes sense.
@Chau would be easy to add the original image as well?
@theodore, I have updated my question to include a thresholded version of part of the original photo.
@Chau I would be more interested to check the original colour image before you apply any procedure on it (e.g. grayscale/thresholding, etc...), if there is not any problem of course. Moreover, did you apply the feature detection algorithm on the colour or in the binary image?
@theodore, the original image is grayscale and I would think that any actual photo contents would lead to more confusion for the algorithms. Thus I have only tried the algorithms on the thresholded image.