From 7bbc42127e46ed6846df219866d10852d63ef393 Mon Sep 17 00:00:00 2001
From: Alexander Mordvintsev <no@email>
Date: Fri, 13 Jul 2012 20:27:54 +0000
Subject: [PATCH] refactored find_obj.py sample: - handle bad matching case -
 use BFMatcher and FlannBasedMatcher (and thus fixing a bug: L2^2 metric was
 used for flann)

---
 samples/python2/find_obj.py | 80 +++++++++++++++++--------------------
 1 file changed, 36 insertions(+), 44 deletions(-)

diff --git a/samples/python2/find_obj.py b/samples/python2/find_obj.py
index a3bfce46a..ae054e844 100644
--- a/samples/python2/find_obj.py
+++ b/samples/python2/find_obj.py
@@ -1,35 +1,17 @@
+'''
+Feature-based image matching sample.
+
+USAGE
+     find_obj.py [ <image1> <image2> ]
+'''
+
 import numpy as np
 import cv2
-from common import anorm
-from functools import partial
-
-help_message = '''SURF image match 
-
-USAGE: findobj.py [ <image1> <image2> ]
-'''
 
 FLANN_INDEX_KDTREE = 1  # bug: flann enums are missing
 
-flann_params = dict(algorithm = FLANN_INDEX_KDTREE,
-                    trees = 4)
+flann_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
 
-def match_bruteforce(desc1, desc2, r_threshold = 0.75):
-    res = []
-    for i in xrange(len(desc1)):
-        dist = anorm( desc2 - desc1[i] )
-        n1, n2 = dist.argsort()[:2]
-        r = dist[n1] / dist[n2]
-        if r < r_threshold:
-            res.append((i, n1))
-    return np.array(res)
-
-def match_flann(desc1, desc2, r_threshold = 0.6):
-    flann = cv2.flann_Index(desc2, flann_params)
-    idx2, dist = flann.knnSearch(desc1, 2, params = {}) # bug: need to provide empty dict
-    mask = dist[:,0] / dist[:,1] < r_threshold
-    idx1 = np.arange(len(desc1))
-    pairs = np.int32( zip(idx1, idx2[:,0]) )
-    return pairs[mask]
 
 def draw_match(img1, img2, p1, p2, status = None, H = None):
     h1, w1 = img1.shape[:2]
@@ -65,39 +47,49 @@ def draw_match(img1, img2, p1, p2, status = None, H = None):
 
 
 if __name__ == '__main__':
+    print __doc__
+    
     import sys
     try: fn1, fn2 = sys.argv[1:3]
     except:
         fn1 = '../c/box.png'
         fn2 = '../c/box_in_scene.png'
-    print help_message
 
     img1 = cv2.imread(fn1, 0)
     img2 = cv2.imread(fn2, 0)
 
-    surf = cv2.SURF(1000)
-    kp1, desc1 = surf.detectAndCompute(img1, None)
-    kp2, desc2 = surf.detectAndCompute(img2, None)
-    desc1.shape = (-1, surf.descriptorSize())
-    desc2.shape = (-1, surf.descriptorSize())
+    detector = cv2.SIFT()
+    kp1, desc1 = detector.detectAndCompute(img1, None)
+    kp2, desc2 = detector.detectAndCompute(img2, None)
     print 'img1 - %d features, img2 - %d features' % (len(kp1), len(kp2))
 
-    def match_and_draw(match, r_threshold):
-        m = match(desc1, desc2, r_threshold)
-        matched_p1 = np.array([kp1[i].pt for i, j in m])
-        matched_p2 = np.array([kp2[j].pt for i, j in m])
-        H, status = cv2.findHomography(matched_p1, matched_p2, cv2.RANSAC, 5.0)
-        print '%d / %d  inliers/matched' % (np.sum(status), len(status))
+    bf_matcher = cv2.BFMatcher(cv2.NORM_L2)
+    flann_matcher = cv2.FlannBasedMatcher(flann_params, {})  # bug : need to pass empty dict (#1329)
 
-        vis = draw_match(img1, img2, matched_p1, matched_p2, status, H)
+    def match_and_draw(matcher, r_threshold = 0.75):
+        raw_matches = matcher.knnMatch(desc1, trainDescriptors = desc2, k = 2)
+        p1, p2 = [], []
+        for m in raw_matches:
+            if len(m) == 2 and m[0].distance < m[1].distance * r_threshold:
+                m = m[0]
+                p1.append( kp1[m.queryIdx].pt )
+                p2.append( kp2[m.trainIdx].pt )
+        p1, p2 = np.float32((p1, p2))
+        if len(p1) >= 4:
+            H, status = cv2.findHomography(p1, p2, cv2.RANSAC, 2.0)
+            print '%d / %d  inliers/matched' % (np.sum(status), len(status))
+        else:
+            H, status = None, None
+            print '%d matches found, not enough for homography estimation' % len(p1)
+
+        vis = draw_match(img1, img2, p1, p2, status, H)
         return vis
 
     print 'bruteforce match:',
-    vis_brute = match_and_draw( match_bruteforce, 0.75 )
+    vis_brute = match_and_draw( bf_matcher )
     print 'flann match:',
-    vis_flann = match_and_draw( match_flann, 0.6 ) # flann tends to find more distant second
-                                                   # neighbours, so r_threshold is decreased
-    cv2.imshow('find_obj SURF', vis_brute)
-    cv2.imshow('find_obj SURF flann', vis_flann)
+    vis_flann = match_and_draw( flann_matcher )
+    cv2.imshow('find_obj', vis_brute)
+    cv2.imshow('find_obj flann', vis_flann)
     0xFF & cv2.waitKey()
     cv2.destroyAllWindows()