def demo():
src_image_feature_path = 'wine_images/src_image_feature_path.path'
feature_path_dict = rw.read_dict(src_image_feature_path)
key_list = feature_path_dict.keys()
img_name, img_kp, img_des = rw.read_feature(feature_path_dict[key_list[0]].strip())
print( img_name )
img_name2, img_kp2, img_des2 = rw.read_feature(feature_path_dict[key_list[100]].strip())
print( img_name2)
r_good_match, good_match = match(img_kp, img_des.astype(np.float32),img_kp2, img_des2.astype(np.float32), 0.75, 'bf')
print(len(good_match))
# MIN_MATCH_COUNT = 10
# ransanc_good_match = []
# if len(good_match) > MIN_MATCH_COUNT:
# # 获取关键点的坐标
# src_pts = np.float32([ img_kp[m[0].queryIdx].pt for m in good_match ]).reshape(-1,1,2)
# dst_pts = np.float32([ img_kp2[m[0].trainIdx].pt for m in good_match ]).reshape(-1,1,2)
# H, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC,5.0)
# for i in range(len(mask)):
# if mask[i]:
# ransanc_good_match.append(good_match[i])
# print( 'ransanc_good_match:', len( ransanc_good_match))
print( len( r_good_match ) )
print( img_des.shape )
def read_des_feature(feature_path_dict):
#from feature path dict to get des dict
des_dict = dict()
kp_dict = dict()
for key in feature_path_dict.keys():
img_name, img_kp, img_des = rw.read_feature(feature_path_dict[key])
kp_dict[img_name] = img_kp
des_dict[img_name] = img_des
return kp_dict, des_dict
def demo1():
#save and load feature, save_feature function and read_feature function
kp_aa = cv2.KeyPoint()
kps_d = [kp_aa, kp_aa]
des_a = np.array([[1, 2, 3], [2, 3, 4], [5, 6, 7]])
rwOperate.save_feature('image1.jpg', kps_d, des_a, './test/data.surf')
img_name, img_kp, img_des = rwOperate.read_feature('./test/data.surf')
print(img_name)
print(img_kp, type(img_kp))
print(img_des, type(img_des))
def getDescriptors(src_image_feature_path):
#read image feature for
#path = '../datafolder/test'
image_feature_path = os.path.join(src_image_feature_path,
'src_image_feature_path.path')
if os.path.exists(image_feature_path):
path_dict = rwOperate.read_dict(image_feature_path)
else:
print('have no src_image_feature_path.path to read!',
image_feature_path)
return -1
descriptors = list()
for key in path_dict.keys():
one_image_feature_path = path_dict[key]
_, _, img_des = rwOperate.read_feature(one_image_feature_path)
descriptors.extend(img_des.tolist())
descriptors = np.asarray(descriptors)
return descriptors
def save_VLAD_to_proto(src_image_feature_path, visualDictionary):
image_feature_path = os.path.join(src_image_feature_path,
'src_image_feature_path.path')
if os.path.exists(image_feature_path):
path_dict = rwOperate.read_dict(image_feature_path)
else:
print('have no src_image_feature_path.path to read!')
descriptors_dict = dict()
for key in path_dict.keys():
try:
one_image_feature_path = path_dict[key]
_, _, img_des = rwOperate.read_feature(one_image_feature_path)
v = VLAD.VLAD(img_des, visualDictionary)
descriptors_dict[key] = v
except:
print('extract vlad feature failure')
continue
rwOperate.save_dict_des(
descriptors_dict,
os.path.join(src_image_feature_path, 'descriptors_dict.vlad'))
def read_des_feature_by_path(feature_path):
'''
通过key读取特征数据，该方式占用内存少，但是耗时长
'''
img_name, img_kp, img_des = rw.read_feature( feature_path )
return img_name, img_kp, img_des
read_image_st = time.time()
img = cv2.imread(image_path)
img = cv2.resize(img, (360, 360))
print('read image time:', time.time() - read_image_st)
kp, des = feature_extract.detect(surf, img)
f_train_path = []
for nameIdx in dest_name_list:
f_train_path.append(feature_path_dict[nameIdx])
f_train_des = []
temp_image_name = []
for pathIdx in f_train_path:
img_name, img_kp, img_des = rwOperate.read_feature(pathIdx)
r_good_match, good_match = feature_match.match(
kp, des.astype(np.float32), img_kp, img_des.astype(np.float32),
0.75, 'bf')
if svm_clf.predict([[float(len(r_good_match)),
float(len(good_match))]]):
print(img_name)
temp_image_name.append(img_name)
search_time.append(time.time() - st_time)
search_result.append(temp_image_name)
print('all time:', time.time() - st_time)
with open('result.txt', 'w+') as f:
f.write(str(search_time))