knnMatch does not work with K != 1

Question

I have some python code to compare two images:

 detector_FeatureDetector_1 = cv2.FastFeatureDetector_create()
 detector_FeatureDetector_2 = cv2.FastFeatureDetector_create()   
 detector_DescriptorExtractor_1 = cv2.BRISK_create()
 detector_DescriptorExtractor_2 = cv2.BRISK_create()
 detector_DescriptorMatcher_1 = cv2.BFMatcher(cv2.NORM_HAMMING2, crossCheck = True)
 detector_DescriptorMatcher_2 = cv2.BFMatcher(cv2.NORM_HAMMING2, crossCheck = True)
 image_1 = cv2.imread('/Users/rui/image1.png')
 image_2 = cv2.imread('/Users/rui/image2.png')
 obj_descriptor_keypoints_1 = detector_FeatureDetector.detect(image_1)
 obj_descriptor_keypoints_2 = detector_FeatureDetector.detect(image_2)
 keypoints1, obj_descriptor_descriptors_1 = detector_DescriptorExtractor.compute(image_1, obj_descriptor_keypoints_1)
 keypoints2, obj_descriptor_descriptors_2 = detector_DescriptorExtractor.compute(image_2, obj_descriptor_keypoints_2)
 matches = detector_DescriptorMatcher.knnMatch(obj_descriptor_descriptors_1, obj_descriptor_descriptors_2, k=6)

But detector_DescriptorMatcher.knnMatch() only works when k=1. If k has a different value than 1, the following error is returned:

OpenCV Error: Assertion failed (K == 1 && update == 0 && mask.empty()) in batchDistance, file /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_graphics_opencv/opencv/work/opencv-3.0.0/modules/core/src/stat.cpp, line 3682

Traceback (most recent call last):
  File "/Users/rui/main.py", line 191, in <module>

matches = detector_DescriptorMatcher.knnMatch(obj_descriptor_descriptors, obj_descriptor_descriptors_movie_frame, k=6)

cv2.error: /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_graphics_opencv/opencv/work/opencv-3.0.0/modules/core/src/stat.cpp:3682: error: (-215) K == 1 && update == 0 && mask.empty() in function batchDistance

Answer 1

The error is caused by configuring BFMatcher with crossCheck = True. For k > 1, set crossCheck = False (constructor default).

From the docs:

If crossCheck==true, then the knnMatch() method with k=1 will only return pairs (i,j) such that for i-th query descriptor the j-th descriptor in the matcher’s collection is the nearest and vice versa, i.e. the BFMatcher will only return consistent pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches. This is alternative to the ratio test, used by D. Lowe in SIFT paper.