OpenCV Python功能检测和匹配

Question

I don't understand the feature matching API in FLANN. I found this code and I am using it as a base line for my development: Image stitching Python

But I can't figure out what the flann.knnMatch() function is returning. It's a list of tuples of the DMatch type. I don't understand why there are two matches in each tuple. I also don't understand why Lowe's ratio is applied to compare the two items in the tuple and then we only use the first item in the tuple.

I can't seem to find any information on the flann python bindings in the OpenCV doc, so I've been struggling to figure out what all of this is.

Any explanation of this would be greatly appreciated.

Thanks.

Answer 1

1. The function flann.knnMatch() method (with k=2) is returning two best matching points.
2. For image stitching you might have to do matching in two directions ie for each point in first image we need to find two best matches in second image and we do the same for the feature points in second image finding their two best matches in the first image.
3. The two best matches are returned based on distance between the descriptors. If the measured distance is very low for the best match and much larger for the second best match then first match is unambiguously the best choice.
4. If the two best matches are relatively close in distance to each other then you perform ratio test of the distance of best match over the distance of the second match is not greater than a given threshold(This generally is what ratio test is done for)