def computeCornerResponse(img, radius, max_number_points,
harris_threshold_rel):
max_dimension = max([img.shape[0], img.shape[1]])
gray_img = toGrayScale(img)
edge_discard = 20
harris_img, interest_points = harris.get_harris_corners(
gray_img, edge_discard, harris_threshold_rel)
num_points_found = len(interest_points[0])
interest_points_transposed = np.transpose(interest_points)
dists = harris.dist2(interest_points_transposed,
interest_points_transposed)
max_global = float("-inf")
max_global_index = None
for index in xrange(num_points_found):
y = interest_points[0][index]
x = interest_points[1][index]
if harris_img[y, x] > max_global:
max_global_index = index
max_global = harris_img[y, x]
indexes_selected = [max_global_index]
for r in reversed(xrange(radius, max_dimension)):
for candidate_index in xrange(num_points_found):
isGood = True
for good_index in indexes_selected:
if dists[candidate_index, good_index] < r * r:
isGood = False
break
if isGood:
indexes_selected.append(candidate_index)
print "Found " + str(len(indexes_selected)) + " out of " + str(
max_number_points) + " points expected."
if len(indexes_selected) >= max_number_points:
break
if len(indexes_selected) >= max_number_points:
break
points = []
for index in indexes_selected:
points.append((interest_points[1][index], interest_points[0][index]))
print "Total points used: " + str(len(points))
figure, axis = plt.subplots(ncols=3)
axis[0].imshow(img, vmin=0, vmax=1)
axis[1].imshow(harris_img, vmin=0, vmax=1)
axis[2].imshow(img, vmin=0, vmax=1)
for x, y in points:
marker = plt.Circle((x, y), 2, color='r')
axis[2].add_artist(marker)
plt.show()
return points
def feature_match(desc_imA, desc_imB):
results = {}
for point_A, vector_A in desc_imA.items():
dists = {}
for point_B, vector_B in desc_imB.items():
dists[point_B] = harris.dist2(vector_A, vector_B)[0][0]
dists = sorted((value, key) for (key, value) in dists.items())
if dists[0][0] / dists[1][0] < .3:
results[point_A] = dists[0][1]
return results
def feature_match(desc_imA, desc_imB):
threshold = .3
results = {}
for point_A, vector_A in desc_imA.items():
dists = {}
for point_B, vector_B in desc_imB.items():
dists[point_B] = harris.dist2(vector_A, vector_B)[0][0]
dists = sorted((value, key) for (key, value) in dists.items())
best = dists[0]
secondbest = dists[1]
if best[0] / secondbest[0] < threshold:
results[point_A] = best[1]
return results
def computeCornerResponse(img, radius, max_number_points,harris_threshold_rel):
max_dimension = max([img.shape[0],img.shape[1]])
gray_img = img
edge_discard = EDGE_DISCARD
harris_img, interest_points = harris.get_harris_corners(gray_img,edge_discard,harris_threshold_rel)
num_points_found = len(interest_points[0])
interest_points_transposed = np.transpose(interest_points)
dists = harris.dist2(interest_points_transposed,interest_points_transposed)
max_global = float("-inf")
max_global_index = None
for index in xrange(num_points_found):
y = interest_points[0][index]
x = interest_points[1][index]
if harris_img[y,x] > max_global:
max_global_index = index
max_global = harris_img[y,x]
indexes_selected = [max_global_index]
for r in reversed(xrange(radius,max_dimension)):
for candidate_index in xrange(num_points_found):
isGood = True
for good_index in indexes_selected:
if dists[candidate_index,good_index] < r*r:
isGood = False
break
if isGood:
indexes_selected.append(candidate_index)
# print "Found " + str(len(indexes_selected)) + " out of " + str(max_number_points) + " points expected."
if len(indexes_selected) >= max_number_points:
break
if len(indexes_selected) >= max_number_points:
break
points = []
for index in indexes_selected:
points.append((interest_points[1][index],interest_points[0][index]))
return points
def ANMS(points, eps, H):
radius_values = {}
for center in points:
H_i = H[center[0], center[1]]
interest_points = []
for point in points:
H_j = H[point[0], point[1]]
if H_i < (eps * H_j):
interest_points.append(point)
if len(interest_points) > 0:
radius_values[center] = np.amin(
harris.dist2(np.array([center]), np.array(interest_points)))
radius_values = sorted(
(value, key) for (key, value) in radius_values.items())[::-1]
top_500 = [[], []]
for i in range(500):
top_500[0].append(radius_values[i][1][0])
top_500[1].append(radius_values[i][1][1])
return top_500
def ANMS(points, eps, H):
remain_num = 500
radius = {}
for center in points:
rC, cC = center[0], center[1]
interest_points = []
H_C = H[rC, cC]
for point in points:
H_p = H[point[0], point[1]]
if H_C < (eps * H_p):
interest_points.append(point)
if len(interest_points) > 0:
radius[center] = np.amin(
harris.dist2(np.array([center]), np.array(interest_points)))
radius_values = sorted(((value, key) for (key, value) in radius.items()),
reverse=True)
top_500 = [[], []]
for i in range(remain_num):
top_500[0].append(radius_values[i][1][0])
top_500[1].append(radius_values[i][1][1])
return top_500