def cnn_features_extraction(path_database,path_retrieval,path_cnn_trained,folders,image_format,feature_extraction_method,list_of_parameters):
if(feature_extraction_method == "cnn_training" or feature_extraction_method == "cnn_probability_training"):
parameters = Parameters(256 ,path_database,folders,image_format, path_database + "database/",list_of_parameters)
#se precisar fazer treianmento
if(parameters.NUM_EPOCHS > 0):
_,_ = train_cnn_tensorFlow.train(parameters)
file = path_database + "features/result" + "_" + feature_extraction_method + ".csv"
if os.path.isfile(file) and parameters.NUM_EPOCHS == 0:
fname_database = []
reader = csv.reader(open(file),delimiter=',')
x = list(reader)
feature_vectors_database = np.array(x).astype('float')
for f in folders:
a = glob(f+'*.'+image_format)
for i in range(len(a)):
fname_database.append(a[i])
else:
feature_vectors_database, _, fname_database = image_feature_extraction_tensorFlow.features_extraction(parameters)
np.savetxt(file, feature_vectors_database,delimiter = ',')
#calling the extraction of features for the retrieval images
parameters.PATH_TEST = path_retrieval
parameters.CLASSES = []
feature_vectors_retrieval,ims_retrieval,_ = image_feature_extraction_tensorFlow.features_extraction(parameters)
return fname_database, feature_vectors_database, ims_retrieval, feature_vectors_retrieval, file
elif(feature_extraction_method == "cnn" or feature_extraction_method == "cnn_probability"):
file = path_database + "features/result" + "_" + feature_extraction_method + ".csv"
if os.path.isfile(file):
fname_database = []
reader = csv.reader(open(file),delimiter=',')
x = list(reader)
feature_vectors_database = np.array(x).astype('float')
for f in folders:
a = glob(f+'*.'+image_format)
for i in range(len(a)):
fname_database.append(a[i])
else:
feature_vectors_database, fname_database, _ = inception_feature_extraction.features_extraction(path_database + "database/",path_cnn_trained,image_format, feature_extraction_method,True)
np.savetxt(file, feature_vectors_database,delimiter = ',')
feature_vectors_retrieval,_ , ims_retrieval = inception_feature_extraction.features_extraction(path_retrieval,path_cnn_trained,image_format, feature_extraction_method,False)
return fname_database, feature_vectors_database, ims_retrieval, feature_vectors_retrieval, file
def run_command_line(path_database, path_retrieval, path_cnn_trained,
feature_extraction_method, distance, number_of_images,
list_of_parameters):
'''
This is the main function of the pycbir project, the interface will call this function
Parameters:
path_databse:
Complete path of the database folder
path_retrieval:
Complete path of the retrieval images folder, if this value is '', then we will compute the retrieval for one image.
path_image:
Complete path of a single retrievial image
'''
folders = glob(path_database + 'database/*/')
image_format = get_extension(folders)
# this is a particular case of cnn to feature extraction inception
if feature_extraction_method == 'cnn' or feature_extraction_method == 'cnn_probability':
if not ("." + image_format) in path_retrieval:
#ims_database,fname_database,feature_vectors_database,ims_retrieval,_,feature_vectors_retrieval = CNN_feature_extraction.cnn_features_extraction_probability(path_database, path_retrieval, 0, image_format,feature_extraction_method,1,list_of_parameters)
ims_database, fname_database, feature_vectors_database, ims_retrieval, _, feature_vectors_retrieval = CNN_feature_extraction.cnn_features_extraction_using_tensorFlow(
path_database, path_retrieval, path_cnn_trained, image_format,
feature_extraction_method)
else: #O retrieval para uma imagem ainda esta usando o theano...
ims_database, fname_database, feature_vectors_database, ims_retrieval, _, feature_vectors_retrieval = CNN_feature_extraction.cnn_features_extraction_probability(
path_database, path_retrieval, -1, image_format,
feature_extraction_method, 1, list_of_parameters)
#training the CNN with the database
elif feature_extraction_method == 'cnn_training' or feature_extraction_method == 'cnn_probability_training':
if not ("." + image_format) in path_retrieval:
parameters = Parameters(256, path_database, folders, image_format,
path_database + "database/",
list_of_parameters)
#calling the training process
if (parameters.NUM_EPOCHS > 0):
ims_database, fname_database = train_cnn_tensorFlow.train(
parameters)
#calling the extraction of features for the database images
feature_vectors_database, ims_database, fname_database = image_feature_extraction_tensorFlow.features_extraction(
parameters)
#calling the extraction of features for the retrieval images
parameters.PATH_TEST = path_retrieval
parameters.CLASSES = []
feature_vectors_retrieval, ims_retrieval, _ = image_feature_extraction_tensorFlow.features_extraction(
parameters)
feature_vectors_retrieval = feature_vectors_retrieval[:, :-1]
#funcionando chamando a rede antigo do theano
#ims_database,fname_database,feature_vectors_database,ims_retrieval,_,feature_vectors_retrieval = CNN_feature_extraction.cnn_features_extraction_probability(path_database, path_retrieval, 0, image_format,feature_extraction_method,1,list_of_parameters)
#funcionando chamando a rede antiga do tensorFlow
#ims_database,fname_database,feature_vectors_database,ims_retrieval,_,feature_vectors_retrieval = CNN_feature_extraction.cnn_training_using_tensorFlow(path_database, path_retrieval, path_cnn_trained, image_format,feature_extraction_method,list_of_parameters)
else: #O retrieval para uma imagem ainda esta usando o theano...
ims_database, fname_database, feature_vectors_database, ims_retrieval, _, feature_vectors_retrieval = CNN_feature_extraction.cnn_features_extraction_probability(
path_database, path_retrieval, -1, image_format,
feature_extraction_method, 1, list_of_parameters)
else:
#check if there is a file computed for this descriptor-distance
parameters_name = ""
for parameter in list_of_parameters:
parameters_name = parameters_name + "_" + parameter
file = path_database + "features/result" + "_" + feature_extraction_method + parameters_name + ".csv"
if os.path.isfile(file):
fname_database = []
reader = csv.reader(open(file), delimiter=',')
x = list(reader)
feature_vectors_database = np.array(x).astype('float')
ims_database = []
cont = 0
for f in folders:
a = glob(f + '*.' + image_format)
for i in range(len(a)):
fname_database.append(a[i])
file = imread(a[i])
ims_database.append(file)
else:
#get the number of features
number_of_features = get_number_of_features(
folders, image_format, feature_extraction_method,
list_of_parameters)
#computing features for the database
ims_database, fname_database, feature_vectors_database = descriptor_all_database(
path_database + 'database/', folders, image_format,
feature_extraction_method, number_of_features,
list_of_parameters)
np.savetxt(file, feature_vectors_database, delimiter=',')
#get the number of features
number_of_features = get_number_of_features(folders, image_format,
feature_extraction_method,
list_of_parameters)
#computing features for the retrieval image(s)
if not ("." + image_format) in path_retrieval:
ims_retrieval, _, feature_vectors_retrieval = descriptor_all_database(
path_retrieval, [], image_format, feature_extraction_method,
number_of_features, list_of_parameters)
else:
print(path_retrieval)
ims_retrieval, _, feature_vectors_retrieval = descriptor_all_database(
path_retrieval, -1, image_format, feature_extraction_method,
number_of_features, list_of_parameters)
#compute the ranked outputs
result = searching(feature_vectors_database[:, :-1],
feature_vectors_database[:, -1],
feature_vectors_retrieval,
similarity_metric=distance)
#show the ranked output
show_retrieval(ims_database, ims_retrieval, result, number_of_images,
path_database, feature_extraction_method, distance,
fname_database, feature_vectors_database[:, -1], folders)