def normalizeHist(self, hist, n_words, n_images):
print 'Applying Okapi transformation...'
k1 = 1.2
b = 0.75
new_hist = [[0 for x in range(n_words)] for x in range(n_images)]
bi = simpleBinarization.SimpleBi()
bi_hist = bi.normalizeHist(hist, n_words, n_images)
bi_hist = np.asarray(bi_hist)
hist_np = np.asarray(hist)
N = n_images
for i in range(0, n_images):
for j in range(0, n_words):
nid = hist[i][j]
nd = hist_np[i].sum()
nt = bi_hist[:, j].sum()
tf = float(k1 * nid) / (nid + k1 * (1 - b + b *
(nd / self.average_words)))
idf = math.log(float(N - nt + 0.5) / (nt + 0.5))
if idf < 0:
new_hist[i][j] = 0
else:
new_hist[i][j] = tf * idf
print 'Okapi method applied'
return new_hist
def normalizeHist(self, hist, n_words, n_images):
print 'Applying tf-idf transformation...'
new_hist = [[0 for x in range(n_words)] for x in range(n_images)]
new_new_hist = [[0 for x in range(n_words)] for x in range(n_images)]
bi = simpleBinarization.SimpleBi()
bi_hist = bi.normalizeHist(hist, n_words, n_images)
N = n_images
for i in range(0, n_images):
for j in range(0, n_words):
nid = hist[i][j]
colunm_bi_hist = [row[j] for row in bi_hist]
nt = sum(colunm_bi_hist)
if nt != 0:
new_hist[i][j] = (float(nid)) * math.log(1 + float(N) / nt)
else:
new_hist[i][j] = 0
for i in range(0, n_images):
for j in range(0, n_words):
tfidf = new_hist[i][j]
sum_line = sum(new_hist[i])
if sum_line != 0:
new_new_hist[i][j] = float(tfidf) / sum_line
else:
new_hist[i][j] = 0
print 'tf-idf applied'
return new_new_hist
def normalizeHist(self, hist, n_words, n_images):
print 'Applying tf-idf transformation...'
new_hist = [[0 for x in range(n_words)] for x in range(n_images)]
bi = simpleBinarization.SimpleBi()
bi_hist = bi.normalizeHist(hist, n_words, n_images)
bi_hist = np.asarray(bi_hist)
N = n_images
for i in range(0,n_images):
for j in range(0,n_words):
nid = hist[i][j]
print nid
nt = bi_hist[:, j].sum()
print nt
if nt!= 0:
new_hist[i][j] = (float(nid))*math.log(1+float(N)/nt)
else:
new_hist[i][j] = 0
print new_hist[i][j]
print 'tf-idf applied'
return new_hist
def applyFilter(self,hist,n_words,n_images):
bi = simpleBinarization.SimpleBi()
bi_hist = bi.normalizeHist(hist, n_words, n_images)
bi_hist = np.array(bi_hist)
words_freq = np.zeros(n_words)
indexes_remove = []
for i in range(0,n_words):
word_freq = float(bi_hist[:,i].sum())/n_images
if word_freq <= self.minThres:
indexes_remove.append(i)
new_hist = np.delete(np.array(hist), indexes_remove, axis=1)
return new_hist
def run(pathImages, method, numpatch, imsample, percentage, codebook, dist,
size, fselec, fselec_perc, histnorm, clust, nclusters, rep):
#################################################################
#
# Initializations and result file configurations
#
#################################################################
im_dataset_name = pathImages.split('/')[-1]
date_time = datetime.datetime.now().strftime('%b-%d-%I%M%p-%G')
name_results_file = 'BOC_' + im_dataset_name + '_' + str(
numpatch
) + '_' + imsample + '_' + codebook + '_' + str(
size
) + '_' + fselec + '_' + histnorm + '_' + clust + '_' + dist + '_' + date_time
#dir_results = 'Results_' + im_dataset_name + '_BOC_' + date_time
dir_results = 'Results_BOC'
if not os.path.exists(dir_results):
os.makedirs(dir_results)
file_count = 2
file_name = os.path.join(dir_results, name_results_file)
while os.path.exists(file_name + ".txt"):
file_name = os.path.join(dir_results,
name_results_file) + "_" + str(file_count)
file_count = file_count + 1
f = open(file_name + ".txt", 'w')
#################################################################
#
# Get images
#
#################################################################
#pathImages = '/Users/Mariana/mieec/Tese/Development/ImageDatabases/Graz-01_sample'
imList = get_imlist(pathImages)
print 'Number of images read = ' + str(len(imList))
f.write("Number of images in dataset read: " + str(len(imList)) + "\n")
#################################################################
#
# Image description
#
#################################################################
kp_vector = [] #vector with the keypoints object
des_vector = [
] #vector wih the descriptors (in order to obtain the codebook)
number_of_kp = [] #vector with the number of keypoints per image
counter = 1
#save current time
start_time = time.time()
labels = []
class_names = []
#ADDED
imPaths = []
#number of divisions of the image
div = int(np.sqrt(numpatch))
n_images = 0
#detect the keypoints and compute the sift descriptors for each image
for im in imList:
if 'DS_Store' not in im:
#ADDED
imPaths.append(im)
print 'image: ' + str(im) + ' number: ' + str(counter)
#read image
img = cv2.imread(im, 1)
img_gray = cv2.imread(im, 0)
img_lab = cv2.cvtColor(img, cv.CV_BGR2Lab)
height, width, comp = img_lab.shape
h_region = height / div
w_region = width / div
des = []
for i in range(0, div):
for j in range(0, div):
#mask
mask = np.zeros(img_gray.shape, dtype=np.uint8)
mask[i * h_region:(i + 1) * h_region,
j * w_region:(j + 1) * w_region] = 1
hist = cv2.calcHist([img_lab], [0, 1, 2], mask,
[256, 256, 256],
[0, 256, 0, 256, 0, 256])
max_color_l, max_color_a, max_color_b = np.where(
hist == np.max(hist))
des.append(
[max_color_l[0], max_color_a[0], max_color_b[0]])
number_of_kp.append(div * div)
if counter == 1:
des_vector = des
else:
des_vector = np.concatenate((des_vector, des), axis=0)
counter += 1
#for evaluation
name1 = im.split("/")[-1]
name = name1.split("_")[0]
if name in class_names:
index = class_names.index(name)
labels.append(index)
else:
class_names.append(name)
index = class_names.index(name)
labels.append(index)
n_images = n_images + 1
#measure the time to compute the description of each image (divide time elapsed by # of images)
elapsed_time = (time.time() - start_time) / len(imList)
print 'Time to compute detector and descriptor for each image = ' + str(
elapsed_time)
f.write(
'Average time to compute detector and descriptor for each image = ' +
str(elapsed_time) + '\n')
average_words = sum(number_of_kp) / float(len(number_of_kp))
print 'Total number of features = ' + str(len(des_vector))
f.write('Total number of features obtained = ' + str(len(des_vector)) +
'\n')
print 'Average number of keypoints per image = ' + str(average_words)
f.write('Average number of keypoints per image = ' + str(average_words) +
'\n')
#################################################################
#
# Image and Keypoint sampling
#
#################################################################
rand_indexes = []
nmi_indexes = []
for iteraction in range(0, rep):
print "\nIteraction #" + str(iteraction + 1) + '\n'
f.write("\nIteraction #" + str(iteraction + 1) + '\n')
print 'Sampling images and keypoints prior to codebook computation...'
if imsample != "NONE":
sampleKp = sampleKeypoints.SamplingImandKey(
n_images, number_of_kp, average_words, percentage)
sampleallKp = sampleAllKeypoints.SamplingAllKey(percentage)
names_sampling = np.array(["SAMPLEI", "SAMPLEP"])
sample_method = np.array([sampleKp, sampleallKp])
#Get the sampling method passed in the -g argument
index = np.where(names_sampling == imsample)[0]
if index.size > 0:
sampling_to_use = sample_method[index[0]]
else:
print 'Wrong sampling method passed in the -g argument. Options: NONE, SAMPLEI, SAMPLEP'
sys.exit()
#FOR RESULTS FILE
sampling_to_use.writeFile(f)
des_vector_sampled = sampling_to_use.sampleKeypoints(des_vector)
print 'Total number of features after sampling = ' + str(
len(des_vector_sampled))
f.write('Total number of features after sampling = ' +
str(len(des_vector_sampled)) + '\n')
print 'Images and keypoints sampled...'
else:
print 'No sampling method chosen'
#FOR RESULTS FILE
f.write(
"No method of keypoint sampling chosen. Use all keypoints for codebook construction \n"
)
des_vector_sampled = des_vector
#################################################################
#
# Codebook computation
#
#################################################################
print 'Obtaining codebook...'
#save current time
start_time = time.time()
#Get detector classes
codebook_kmeans = KMeans1.KMeans1(size)
codebook_birch = Birch.Birch(size)
codebook_minibatch = minibatch.MiniBatch(size)
codebook_randomv = randomSamplesBook.RandomVectors(size)
codebook_allrandom = allrandom.AllRandom(size)
names_codebook = np.array(
["KMEANS", "BIRCH", "MINIBATCH", "RANDOMV", "RANDOM"])
codebook_algorithm = np.array([
codebook_kmeans, codebook_birch, codebook_minibatch,
codebook_randomv, codebook_allrandom
])
#Get the codebook algorithm passed in the -c argument
index = np.where(names_codebook == codebook)[0]
if index.size > 0:
codebook_to_use = codebook_algorithm[index[0]]
else:
print 'Wrong codebook construction algorithm name passed in the -c argument. Options: KMEANS, MINIBATCH, RANDOMV and RANDOM'
sys.exit()
#FOR RESULTS FILE
codebook_to_use.writeFileCodebook(f)
#Get centers and projections using codebook algorithm
ceters, projections = codebook_to_use.obtainCodebook(
des_vector_sampled, des_vector)
elapsed_time = (time.time() - start_time)
print 'Time to compute codebook = ' + str(elapsed_time)
f.write('Time to compute codebook = ' + str(elapsed_time) + '\n')
#################################################################
#
# Obtain Histogram
#
#################################################################
print 'Obtaining histograms...'
#print 'projection shape = '+ str(projections.shape)
#print 'size = ' + str(size)
#print 'n of images = ' + str(n_images)
#print 'number of kp' + str(number_of_kp)
hist = histogram.computeHist(projections, size, n_images, number_of_kp)
print hist
print 'Histograms obtained'
################################################################
#
# Feature selection
#
#################################################################
print 'Number of visual words = ' + str(len(hist[0]))
if fselec != "NONE":
print 'Applying feature selection to descriptors...'
filter_max = filterMax.WordFilterMax(fselec_perc[0])
filter_min = filterMin.WordFilterMin(fselec_perc[1])
filter_maxmin = filterMaxMin.WordFilterMaxMin(
fselec_perc[0], fselec_perc[1])
names_filter = np.array(["FMAX", "FMIN", "FMAXMIN"])
filter_method = np.array([filter_max, filter_min, filter_maxmin])
#Get the feature selection method passed in the -f argument
index = np.where(names_filter == fselec)[0]
if index.size > 0:
filter_to_use = filter_method[index[0]]
else:
print 'Wrong codebook construction algorithm name passed in the -f argument. Options: NONE, FMAX, FMIN, FMAXMIN'
sys.exit()
hist = filter_to_use.applyFilter(hist, size, n_images)
#FOR RESULTS FILE
filter_to_use.writeFile(f)
new_size = hist.shape[1]
print 'Visual words Filtered'
print 'Number of visual words filtered = ' + str(size - new_size)
f.write("Number of visual words filtered = " +
str(size - new_size) + '\n')
print 'Final number of visual words = ' + str(new_size)
f.write('Final number of visual words = ' + str(new_size) + '\n')
else:
#FOR RESULTS FILE
filter_min = filterMin.WordFilterMin(0)
hist = filter_min.applyFilter(hist, size, n_images)
new_size = hist.shape[1]
print 'Number of visual words filtered = ' + str(size - new_size)
f.write("No feature selection applied \n")
#################################################################
#
# Histogram Normalization
#
#################################################################
if histnorm != "NONE":
#Get detector classes
norm_sbin = simpleBinarization.SimpleBi()
norm_tfnorm = tfnorm.Tfnorm()
norm_tfidf = tfidf.TfIdf()
norm_tfidf2 = tfidf2.TfIdf2()
norm_tfidfnorm = tfidfnorm.TfIdfnorm()
norm_okapi = okapi.Okapi(average_words)
norm_power = powerNorm.PowerNorm()
names_normalization = np.array([
"SBIN", "TFNORM", "TFIDF", "TFIDF2", "TFIDFNORM", "OKAPI",
"POWER"
])
normalization_method = np.array([
norm_sbin, norm_tfnorm, norm_tfidf, norm_tfidf2,
norm_tfidfnorm, norm_okapi, norm_power
])
#Get the detector passed in the -h argument
index = np.where(names_normalization == histnorm)[0]
if index.size > 0:
normalization_to_use = normalization_method[index[0]]
new_hist = normalization_to_use.normalizeHist(
hist, new_size, n_images)
else:
print 'Wrong normalization name passed in the -h argument. Options: SBIN, TFNORM, TFIDF and TFIDF2'
sys.exit()
#FOR RESULTS FILE
normalization_to_use.writeFile(f)
else:
#FOR RESULTS FILE
f.write("No histogram normalization applied\n")
new_hist = hist
#################################################################
#
# Clustering of the features
#
#################################################################
#save current time
start_time = time.time()
#Get detector classes
clust_dbscan = Dbscan.Dbscan(dist)
clust_kmeans = KMeans1.KMeans1([nclusters])
clust_birch = Birch.Birch(nclusters)
clust_meanSift = meanSift.MeanSift(nclusters)
clust_hierar1 = hierarchicalClustering.Hierarchical(nclusters, dist)
clust_hierar2 = hierarchicalClustScipy.HierarchicalScipy(dist)
clust_community = communityDetection.CommunityDetection(dist)
names_clustering = np.array([
"DBSCAN", "KMEANS", "BIRCH", "MEANSIFT", "HIERAR1", "HIERAR2",
"COMM"
])
clustering_algorithm = np.array([
clust_dbscan, clust_kmeans, clust_birch, clust_meanSift,
clust_hierar1, clust_hierar2, clust_community
])
#Get the detector passed in the -a argument
index = np.where(names_clustering == clust)[0]
if index.size > 0:
clustering_to_use = clustering_algorithm[index[0]]
else:
print 'Wrong clustering algorithm name passed in the -a argument. Options: DBSCAN, KMEANS, BIRCH, MEANSIFT, HIERAR1, HIERAR2, COMM'
sys.exit()
clusters = clustering_to_use.obtainClusters(new_hist)
#FOR RESULTS FILE
clustering_to_use.writeFileCluster(f)
elapsed_time = (time.time() - start_time)
print 'Time to run clustering algorithm = ' + str(elapsed_time)
f.write('Time to run clustering algorithm = ' + str(elapsed_time) +
'\n')
print 'Number of clusters obtained = ' + str(max(clusters) + 1)
f.write('Number of clusters obtained = ' + str(max(clusters) + 1) +
'\n')
nclusters = max(clusters) + 1
print 'Clusters obtained = ' + str(np.asarray(clusters))
#date_time = datetime.datetime.now().strftime('%b-%d-%I%M%p-%G')
#np.savetxt('saveClusters_'+date_time+'_.txt', clusters, '%i', ',')
#ADDED
#################################################################
#
# Create folder with central images for each cluster
#
#################################################################
#obtain representative images for each cluster
central_ims = clust_community.obtainCenteralImages(new_hist, clusters)
central_folder = os.path.join(dir_results, 'CenterImages')
if not os.path.exists(central_folder):
os.makedirs(central_folder)
count = 0
for central_im in central_ims:
filename = os.path.join(central_folder,
'Cluster_' + str(count) + '.jpg')
img = cv2.imread(imPaths[central_im], 1)
cv2.imwrite(filename, img)
count = count + 1
#ADDED
#################################################################
#
# Separate Clusters into folders
#
#################################################################
clusters_folder = os.path.join(dir_results, 'Clusters')
if not os.path.exists(clusters_folder):
os.makedirs(clusters_folder)
clust_dir = []
for iclust in range(0, nclusters):
direc = os.path.join(clusters_folder, 'Cluster_' + str(iclust))
if not os.path.exists(direc):
os.makedirs(direc)
clust_dir.append(direc)
for im in range(0, len(imPaths)):
im_name = imPaths[im].split('/')[-1]
#print clust_dir[int(clusters[im])]
filename = os.path.join(clust_dir[int(clusters[im])], im_name)
#print filename
img = cv2.imread(imPaths[im], 1)
cv2.imwrite(filename, img)
#################################################################
#
# Evaluation
#
#################################################################
users = 0
if users == 1:
rand_index = evaluationUsers.randIndex(clusters)
rand_indexes.append(rand_index)
print 'rand_index = ' + str(rand_index)
f.write("Rand Index = " + str(rand_index) + "\n")
else:
if len(clusters) == len(labels):
f.write("\nResults\n")
f.write('Clusters Obtained = ' + str(np.asarray(clusters)))
f.write('Labels = ' + str(np.asarray(labels)))
rand_index = metrics.adjusted_rand_score(labels, clusters)
rand_indexes.append(rand_index)
print 'rand_index = ' + str(rand_index)
f.write("Rand Index = " + str(rand_index) + "\n")
NMI_index = metrics.normalized_mutual_info_score(
labels, clusters)
nmi_indexes.append(NMI_index)
print 'NMI_index = ' + str(NMI_index)
f.write("NMI Index = " + str(NMI_index) + "\n")
if rep > 1:
f.write("\nFINAL RESULTS\n")
f.write("Avg Rand Index = " + str(float(sum(rand_indexes)) / rep) +
"\n")
f.write("Std Rand Index = " + str(statistics.stdev(rand_indexes)) +
"\n")
if users != 1:
f.write("Avg NMI Index = " + str(float(sum(nmi_indexes)) / rep) +
"\n")
f.write("Std NMI Index = " + str(statistics.stdev(nmi_indexes)) +
"\n")
f.close()
def run(pathImages, method, keypnt, numpatch, equalnum, imdes, imsample,
percentage, codebook, dist, size, fselec, fselec_perc, histnorm, clust,
K, pca, nclusters, rep):
#################################################################
#
# Initializations and result file configurations
#
#################################################################
im_dataset_name = pathImages.split('/')[-1]
date_time = datetime.datetime.now().strftime('%b-%d-%I%M%p-%G')
name_results_file = 'BOF_' + im_dataset_name + '_' + keypnt + '_' + str(
numpatch
) + '_' + str(
equalnum
) + '_' + imdes + '_' + imsample + '_' + codebook + '_' + str(
size
) + '_' + fselec + '_' + histnorm + '_' + clust + '_' + dist + '_' + date_time
#dir_results = 'Results_' + im_dataset_name + '_BOF_' + date_time
dir_results = 'Results_BOF'
if not os.path.exists(dir_results):
os.makedirs(dir_results)
file_count = 2
file_name = os.path.join(dir_results, name_results_file)
while os.path.exists(file_name + ".txt"):
file_name = os.path.join(dir_results,
name_results_file) + "_" + str(file_count)
file_count = file_count + 1
f = open(file_name + ".txt", 'w')
#################################################################
#
# Get images
#
#################################################################
#pathImages = '/Users/Mariana/mieec/Tese/Development/ImageDatabases/Graz-01_sample'
imList = get_imlist(pathImages)
print 'Number of images read = ' + str(len(imList))
f.write("Number of images in dataset read: " + str(len(imList)) + "\n")
#################################################################
#
# Image description
#
#################################################################
#Get detector classes
det_sift = siftLib.Sift(numpatch, equalnum)
det_surf = surfLib.Surf(numpatch, equalnum)
det_fast = fastDetector.Fast(numpatch, equalnum)
det_star = starDetector.Star(numpatch, equalnum)
det_orb = orbLib.Orb(numpatch, equalnum)
det_random = randomDetector.Random(numpatch)
names_detectors = np.array(
["SIFT", "SURF", "FAST", "STAR", "ORB", "RANDOM"])
detectors = np.array(
[det_sift, det_surf, det_fast, det_star, det_orb, det_random])
#Get the detector passed in the -k argument
index = np.where(names_detectors == keypnt)[0]
if index.size > 0:
detector_to_use = detectors[index[0]]
else:
print 'Wrong detector name passed in the -k argument. Options: SIFT, SURF, FAST, STAR, ORB and RANDOM'
sys.exit()
#FOR RESULTS FILE
detector_to_use.writeParametersDet(f)
#Get descriptor classes
des_sift = siftLib.Sift(numpatch, equalnum)
des_surf = surfLib.Surf(numpatch, equalnum)
des_orb = orbLib.Orb(numpatch)
des_brief = briefDescriptor.Brief()
des_freak = freakDescriptor.Freak()
names_descriptors = np.array(["SIFT", "SURF", "ORB", "BRIEF", "FREAK"])
descriptors = np.array([des_sift, des_surf, des_orb, des_brief, des_freak])
#Get the detector passed in the -d argument
index = np.where(names_descriptors == imdes)[0]
if index.size > 0:
descriptor_to_use = descriptors[index[0]]
else:
print 'Wrong descriptor name passed in the -d argument. Options: SIFT, SURF, ORB, BRIEF and FREAK'
sys.exit()
#FOR RESULTS FILE
descriptor_to_use.writeParametersDes(f)
kp_vector = [] #vector with the keypoints object
des_vector = [
] #vector wih the descriptors (in order to obtain the codebook)
number_of_kp = [] #vector with the number of keypoints per image
counter = 1
#save current time
start_time = time.time()
labels = []
class_names = []
#ADDED
imPaths = []
#detect the keypoints and compute the sift descriptors for each image
for im in imList:
if 'DS_Store' not in im:
#ADDED
imPaths.append(im)
print 'image: ' + str(im) + ' number: ' + str(counter)
#read image
img = cv2.imread(im, 0)
#mask in order to avoid keypoints in border of image. size = 40 pixels
border = 40
height, width = img.shape
mask = np.zeros(img.shape, dtype=np.uint8)
mask[border:height - border, border:width - border] = 1
#get keypoints from detector
kp = detector_to_use.detectKp(img, mask)
#get features from descriptor
des = descriptor_to_use.computeDes(img, kp)
number_of_kp.append(len(kp))
kp_vector.append(kp)
if counter == 1:
des_vector = des
else:
des_vector = np.concatenate((des_vector, des), axis=0)
counter += 1
#for evaluation
name1 = im.split("/")[-1]
name = name1.split("_")[0]
if name in class_names:
index = class_names.index(name)
labels.append(index)
else:
class_names.append(name)
index = class_names.index(name)
labels.append(index)
#measure the time to compute the description of each image (divide time elapsed by # of images)
elapsed_time = (time.time() - start_time) / len(imList)
print 'Time to compute detector and descriptor for each image = ' + str(
elapsed_time)
f.write(
'Average time to compute detector and descriptor for each image = ' +
str(elapsed_time) + '\n')
n_images = len(kp_vector)
average_words = sum(number_of_kp) / float(len(number_of_kp))
print 'Total number of features = ' + str(len(des_vector))
f.write('Total number of features obtained = ' + str(len(des_vector)) +
'\n')
print 'Average number of keypoints per image = ' + str(average_words)
f.write('Average number of keypoints per image = ' + str(average_words) +
'\n')
#################################################################
#
# Dimentionality reduction
#
#################################################################
if pca != None:
start_time = time.time()
print 'Applying PCA...'
pca = PCA(n_components=pca)
descriptors_reduced = pca.fit(des_vector).transform(des_vector)
print 'PCA Applied.'
print 'time to apply PCA = ' + str(time.time() - start_time)
des_vector = descriptors_reduced
#################################################################
#
# Image and Keypoint sampling
#
#################################################################
rand_indexes = []
nmi_indexes = []
for iteraction in range(0, rep):
print "\nIteraction #" + str(iteraction + 1) + '\n'
f.write("\nIteraction #" + str(iteraction + 1) + '\n')
print 'Sampling images and keypoints prior to codebook computation...'
if imsample != "NONE":
sampleKp = sampleKeypoints.SamplingImandKey(
n_images, number_of_kp, average_words, percentage)
sampleallKp = sampleAllKeypoints.SamplingAllKey(percentage)
names_sampling = np.array(["SAMPLEI", "SAMPLEP"])
sample_method = np.array([sampleKp, sampleallKp])
#Get the detector passed in the -g argument
index = np.where(names_sampling == imsample)[0]
if index.size > 0:
sampling_to_use = sample_method[index[0]]
else:
print 'Wrong sampling method passed in the -g argument. Options: NONE, SAMPLEI, SAMPLEP'
sys.exit()
#FOR RESULTS FILE
sampling_to_use.writeFile(f)
des_vector_sampled = sampling_to_use.sampleKeypoints(des_vector)
print 'Total number of features after sampling = ' + str(
len(des_vector_sampled))
f.write('Total number of features after sampling = ' +
str(len(des_vector_sampled)) + '\n')
print 'Images and keypoints sampled...'
else:
print 'No sampling method chosen'
#FOR RESULTS FILE
f.write(
"No method of keypoint sampling chosen. Use all keypoints for codebook construction \n"
)
des_vector_sampled = des_vector
#################################################################
#
# Codebook computation
#
#################################################################
print 'Obtaining codebook...'
#save current time
start_time = time.time()
#Get detector classes
codebook_kmeans = KMeans1.KMeans1(size)
codebook_birch = Birch.Birch(size)
codebook_minibatch = minibatch.MiniBatch(size)
codebook_randomv = randomSamplesBook.RandomVectors(size)
codebook_allrandom = allrandom.AllRandom(size)
names_codebook = np.array(
["KMEANS", "BIRCH", "MINIBATCH", "RANDOMV", "RANDOM"])
codebook_algorithm = np.array([
codebook_kmeans, codebook_birch, codebook_minibatch,
codebook_randomv, codebook_allrandom
])
#Get the detector passed in the -c argument
index = np.where(names_codebook == codebook)[0]
if index.size > 0:
codebook_to_use = codebook_algorithm[index[0]]
else:
print 'Wrong codebook construction algorithm name passed in the -c argument. Options: KMEANS, MINIBATCH, RANDOMV and RANDOM'
sys.exit()
#FOR RESULTS FILE
codebook_to_use.writeFileCodebook(f)
#Get centers and projections using codebook algorithm
centers, projections = codebook_to_use.obtainCodebook(
des_vector_sampled, des_vector)
#compute the number of unique descriptor vectors
codebook_randomv.unique_vectors(centers)
elapsed_time = (time.time() - start_time)
print 'Time to compute codebook = ' + str(elapsed_time)
f.write('Time to compute codebook = ' + str(elapsed_time) + '\n')
#################################################################
#
# Obtain Histogram
#
#################################################################
print 'Obtaining histograms...'
#print 'projection shape = '+ str(projections.shape)
#print 'size = ' + str(size)
#print 'n of images = ' + str(n_images)
#print 'number of kp' + str(number_of_kp)
hist = histogram.computeHist(projections, size, n_images, number_of_kp)
#print hist
print 'Histograms obtained'
################################################################
#
# Feature selection
#
#################################################################
print 'Number of visual words = ' + str(len(hist[0]))
if fselec != "NONE":
print 'Applying feature selection to descriptors...'
filter_max = filterMax.WordFilterMax(fselec_perc[0])
filter_min = filterMin.WordFilterMin(fselec_perc[1])
filter_maxmin = filterMaxMin.WordFilterMaxMin(
fselec_perc[0], fselec_perc[1])
names_filter = np.array(["FMAX", "FMIN", "FMAXMIN"])
filter_method = np.array([filter_max, filter_min, filter_maxmin])
#Get the detector passed in the -f argument
index = np.where(names_filter == fselec)[0]
if index.size > 0:
filter_to_use = filter_method[index[0]]
else:
print 'Wrong codebook construction algorithm name passed in the -f argument. Options: NONE, FMAX, FMIN, FMAXMIN'
sys.exit()
hist = filter_to_use.applyFilter(hist, size, n_images)
#FOR RESULTS FILE
filter_to_use.writeFile(f)
new_size = hist.shape[1]
print 'Visual words Filtered'
print 'Number of visual words filtered = ' + str(size - new_size)
f.write("Number of visual words filtered = " +
str(size - new_size) + '\n')
print 'Final number of visual words = ' + str(new_size)
f.write('Final number of visual words = ' + str(new_size) + '\n')
else:
#FOR RESULTS FILE
filter_min = filterMin.WordFilterMin(0)
hist = filter_min.applyFilter(hist, size, n_images)
new_size = hist.shape[1]
print 'Number of visual words filtered = ' + str(size - new_size)
f.write("No feature selection applied \n")
#################################################################
#
# Histogram Normalization
#
#################################################################
if histnorm != "NONE":
#Get detector classes
norm_sbin = simpleBinarization.SimpleBi()
norm_tfnorm = tfnorm.Tfnorm()
norm_tfidf = tfidf.TfIdf()
norm_tfidf2 = tfidf2.TfIdf2()
norm_tfidf3 = tfidf3.Tfidf3()
norm_power = powerNorm.PowerNorm()
norm_tfidfnorm = tfidfnorm.TfIdfnorm()
norm_okapi = okapi.Okapi(average_words)
names_normalization = np.array(
["SBIN", "TFNORM", "TFIDF", "TFIDF2", "TFIDFNORM", "OKAPI"])
normalization_method = np.array([
norm_sbin, norm_tfnorm, norm_tfidf, norm_tfidf2,
norm_tfidfnorm, norm_okapi
])
#Get the detector passed in the -h argument
index = np.where(names_normalization == histnorm)[0]
if index.size > 0:
normalization_to_use = normalization_method[index[0]]
new_hist = normalization_to_use.normalizeHist(
hist, new_size, n_images)
else:
print 'Wrong normalization name passed in the -h argument. Options: SBIN, TFNORM, TFIDF and TFIDF2'
sys.exit()
#FOR RESULTS FILE
normalization_to_use.writeFile(f)
else:
#FOR RESULTS FILE
f.write("No histogram normalization applied\n")
new_hist = hist
#################################################################
#
# Clustering of the features
#
#################################################################
#save current time
start_time = time.time()
#Get detector classes
clust_dbscan = Dbscan.Dbscan(dist)
clust_kmeans = KMeans1.KMeans1([nclusters])
clust_kmeans2 = kmeans2.KMeans2([nclusters])
clust_birch = Birch.Birch(nclusters)
clust_meanSift = meanSift.MeanSift(nclusters)
clust_hierar1 = hierarchicalClustering.Hierarchical(nclusters, dist)
clust_hierar2 = hierarchicalClustScipy.HierarchicalScipy(dist)
clust_community = communityDetection.CommunityDetection(dist)
names_clustering = np.array([
"DBSCAN", "KMEANS", "BIRCH", "MEANSIFT", "HIERAR1", "HIERAR2",
"COMM"
])
clustering_algorithm = np.array([
clust_dbscan, clust_kmeans, clust_birch, clust_meanSift,
clust_hierar1, clust_hierar2, clust_community
])
#Get the detector passed in the -a argument
index = np.where(names_clustering == clust)[0]
if index.size > 0:
clustering_to_use = clustering_algorithm[index[0]]
else:
print 'Wrong clustering algorithm name passed in the -a argument. Options: DBSCAN, KMEANS, BIRCH, MEANSIFT, HIERAR1, HIERAR2, COMM'
sys.exit()
clusters = clustering_to_use.obtainClusters(new_hist)
#FOR RESULTS FILE
clustering_to_use.writeFileCluster(f)
elapsed_time = (time.time() - start_time)
print 'Time to run clustering algorithm = ' + str(elapsed_time)
f.write('Time to run clustering algorithm = ' + str(elapsed_time) +
'\n')
#ADDED
nclusters = int(max(clusters) + 1)
print 'Number of clusters obtained = ' + str(max(clusters) + 1)
f.write('Number of clusters obtained = ' + str(max(clusters) + 1) +
'\n')
print 'Clusters obtained = ' + str(np.asarray(clusters))
#date_time = datetime.datetime.now().strftime('%b-%d-%I%M%p-%G')
#np.savetxt('saveClusters_'+date_time+'_.txt', clusters, '%i', ',')
#ADDED
#################################################################
#
# Create folder with central images for each cluster
#
#################################################################
###obtain representative images for each cluster
#central_ims = clust_community.obtainCenteralImages(new_hist, clusters)
#central_folder = os.path.join(dir_results,'CenterImages')
#if not os.path.exists(central_folder):
#os.makedirs(central_folder)
#count=0
#for central_im in central_ims:
#filename = os.path.join(central_folder,'Cluster_'+str(count)+'.jpg')
#img = cv2.imread(imPaths[central_im],1)
#cv2.imwrite(filename, img)
#count = count + 1
##ADDED
##################################################################
##
## Separate Clusters into folders
##
##################################################################
#clusters_folder = os.path.join(dir_results,'Clusters')
#if not os.path.exists(clusters_folder):
#os.makedirs(clusters_folder)
#clust_dir = []
#for iclust in range(0,nclusters):
#direc = os.path.join(clusters_folder,'Cluster_'+str(iclust))
#if not os.path.exists(direc):
#os.makedirs(direc)
#clust_dir.append(direc)
#for im in range(0,len(imPaths)):
#im_name = imPaths[im].split('/')[-1]
##print clust_dir[int(clusters[im])]
#filename = os.path.join(clust_dir[int(clusters[im])],im_name)
##print filename
#img = cv2.imread(imPaths[im],1)
#cv2.imwrite(filename, img)
##calculate distances between images and closest images
#closest_im = distances.calculateClosest(new_hist,dist)
##print closest_im
#if not os.path.exists('ClosestImages'):
#os.makedirs('ClosestImages')
#file_name = os.path.join('ClosestImages',name_results_file)
#f2 = open(file_name + ".txt", 'w')
#counter = 0
#counter2 = 1
#for ims in closest_im:
#for im in ims:
#f2.write(str(counter2) + '-' + str(counter) + '-' + str(im) + '\n')
#counter2 = counter2 + 1
#counter = counter + 1
#f2.close()
#################################################################
#
# Evaluation
#
#################################################################
users = 0
#labels = np.load('IndividualClustersMatrix.npy')
if users == 1:
rand_index = evaluationUsers.randIndex(clusters)
rand_indexes.append(rand_index)
print 'rand_index = ' + str(rand_index)
f.write("Rand Index = " + str(rand_index) + "\n")
else:
if len(clusters) == len(labels):
f.write("\nResults\n")
f.write('Clusters Obtained = ' + str(np.asarray(clusters)))
f.write('Labels = ' + str(np.asarray(labels)))
rand_index = metrics.adjusted_rand_score(labels, clusters)
rand_indexes.append(rand_index)
print 'rand_index = ' + str(rand_index)
f.write("Rand Index = " + str(rand_index) + "\n")
NMI_index = metrics.normalized_mutual_info_score(
labels, clusters)
nmi_indexes.append(NMI_index)
print 'NMI_index = ' + str(NMI_index)
f.write("NMI Index = " + str(NMI_index) + "\n")
if rep > 1:
f.write("\nFINAL RESULTS\n")
f.write("Avg Rand Index = " + str(float(sum(rand_indexes)) / rep) +
"\n")
f.write("Std Rand Index = " + str(statistics.stdev(rand_indexes)) +
"\n")
if users != 1:
f.write("Avg NMI Index = " + str(float(sum(nmi_indexes)) / rep) +
"\n")
f.write("Std NMI Index = " + str(statistics.stdev(nmi_indexes)) +
"\n")
f.close()
filter_min = filterMin.WordFilterMin(0)
hist = filter_min.applyFilter(hist, size, n_images)
new_size = hist.shape[1]
print 'Number of visual words filtered = ' + str(size - new_size)
f.write("No feature selection applied \n")
#################################################################
#
# Histogram Normalization
#
#################################################################
if histnorm != "NONE":
#Get detector classes
norm_sbin = simpleBinarization.SimpleBi()
norm_tfnorm = tfnorm.Tfnorm()
norm_tfidf = tfidf.TfIdf()
norm_tfidf2 = tfidf2.TfIdf2()
norm_tfidfnorm = tfidfnorm.TfIdfnorm()
norm_okapi = okapi.Okapi(average_words)
names_normalization = np.array(
["SBIN", "TFNORM", "TFIDF", "TFIDF2", "TFIDFNORM", "OKAPI"])
normalization_method = np.array([
norm_sbin, norm_tfnorm, norm_tfidf, norm_tfidf2,
norm_tfidfnorm, norm_okapi
])
#Get the detector passed in the -h argument
index = np.where(names_normalization == histnorm)[0]
def run(pathImages, method, keypnt, numpatch, equalnum, imdes, imsample,
percentage, codebook, dist, size, fselec, fselec_perc, histnorm, clust,
K, pca, nclusters, rep, levels):
#################################################################
#
# Initializations and result file configurations
#
#################################################################
#warnings.simplefilter("error")
if os.path.exists('save_HIST.txt') == True:
os.remove('save_HIST.txt')
if os.path.exists('save_dist.txt') == True:
os.remove('save_dist.txt')
if os.path.exists('saveClustersKmeans.txt') == True:
os.remove('saveClustersKmeans.txt')
im_dataset_name = pathImages.split('/')[-1]
date_time = datetime.datetime.now().strftime('%b-%d-%I%M%p-%G')
name_results_file = im_dataset_name + '_' + keypnt + '_' + str(
numpatch
) + '_' + str(equalnum) + '_' + imdes + '_' + 'levels:' + str(
levels
) + '_' + imsample + '_' + codebook + '_' + str(
size
) + '_' + fselec + '_' + histnorm + '_' + clust + '_' + dist + '_' + date_time
#dir_results = 'Results_' + im_dataset_name + '_SPM_' + date_time
dir_results = 'Results_SPM'
if not os.path.exists(dir_results):
os.makedirs(dir_results)
file_count = 2
file_name = os.path.join(dir_results, name_results_file)
while os.path.exists(file_name + ".txt"):
file_name = os.path.join(dir_results,
name_results_file) + "_" + str(file_count)
file_count = file_count + 1
f = open(file_name + ".txt", 'w')
#################################################################
#
# Get images
#
#################################################################
#pathImages = '/Users/Mariana/mieec/Tese/Development/ImageDatabases/Graz-01_sample'
imList = get_imlist(pathImages)
print 'Number of images read = ' + str(len(imList))
f.write("Number of images in dataset read: " + str(len(imList)) + "\n")
#################################################################
#
# Image description
#
#################################################################
#Number of regions
n_regions = np.power(4, levels - 1)
#Get detector classes
det_sift = siftLib.Sift(numpatch / n_regions, equalnum)
det_surf = surfLib.Surf(numpatch / n_regions, equalnum)
det_fast = fastDetector.Fast(numpatch / n_regions, equalnum)
det_star = starDetector.Star(numpatch / n_regions, equalnum)
det_orb = orbLib.Orb(numpatch / n_regions, equalnum)
det_random = randomDetector.Random(numpatch / n_regions)
names_detectors = np.array(
["SIFT", "SURF", "FAST", "STAR", "ORB", "RANDOM"])
detectors = np.array(
[det_sift, det_surf, det_fast, det_star, det_orb, det_random])
#Get the detector passed in the -k argument
index = np.where(names_detectors == keypnt)[0]
if index.size > 0:
detector_to_use = detectors[index[0]]
else:
print 'Wrong detector name passed in the -k argument. Options: SIFT, SURF, FAST, STAR, ORB and RANDOM'
sys.exit()
#FOR RESULTS FILE
detector_to_use.writeParametersDet(f)
#Get descriptor classes
des_sift = siftLib.Sift(numpatch / n_regions, equalnum)
des_surf = surfLib.Surf(numpatch / n_regions, equalnum)
des_orb = orbLib.Orb(numpatch / n_regions)
des_brief = briefDescriptor.Brief()
des_freak = freakDescriptor.Freak()
names_descriptors = np.array(["SIFT", "SURF", "ORB", "BRIEF", "FREAK"])
descriptors = np.array([des_sift, des_surf, des_orb, des_brief, des_freak])
#Get the detector passed in the -d argument
index = np.where(names_descriptors == imdes)[0]
if index.size > 0:
descriptor_to_use = descriptors[index[0]]
else:
print 'Wrong descriptor name passed in the -d argument. Options: SIFT, SURF, ORB, BRIEF and FREAK'
sys.exit()
#FOR RESULTS FILE
descriptor_to_use.writeParametersDes(f)
kp_vector = [] #vector with the keypoints object
des_vector = [
] #vector wih the descriptors (in order to obtain the codebook)
number_of_kp = [] #vector with the number of keypoints per image
counter = 1
#save current time
start_time = time.time()
labels = []
class_names = []
#Border
border = 40
side = int(np.sqrt(n_regions))
des_vector_byregion = [0] * n_regions
number_of_kp_region = [0] * n_regions
filled = [0] * n_regions
#matrixes of the indexes
mat_indexes = np.array([[0, 1, 4, 5, 16, 17, 20, 21],
[2, 3, 6, 7, 18, 19, 22, 23],
[8, 9, 12, 13, 24, 25, 28, 29],
[10, 11, 14, 15, 26, 27, 30, 31],
[32, 33, 36, 37, 48, 49, 52, 53],
[34, 35, 38, 39, 50, 51, 54, 55],
[40, 41, 44, 45, 56, 57, 60, 61],
[42, 43, 46, 47, 58, 59, 62, 63]])
#detect the keypoints and compute the sift descriptors for each image
for im in imList:
if 'DS_Store' not in im:
print 'image: ' + str(im) + ' number: ' + str(counter)
#read image
img = cv2.imread(im, 0)
# region
for i in range(0, side):
for j in range(0, side):
#mask in order to avoid keypoints in border of image. size = 40 pixels
height, width = img.shape
h_region = (height - 2 * border) / np.sqrt(n_regions)
w_region = (width - 2 * border) / np.sqrt(n_regions)
mask = np.zeros(img.shape, dtype=np.uint8)
mask[border + i * h_region:border + (i + 1) * h_region,
border + j * w_region:border + (j + 1) * w_region] = 1
#get keypoints from detector
kp = detector_to_use.detectKp(img, mask)
#get features from descriptor
des = descriptor_to_use.computeDes(img, kp)
number_of_kp.append(len(kp))
#print i*np.sqrt(n_regions)+j
#print number_of_kp_region[int(i*np.sqrt(n_regions)+j)]
if filled[mat_indexes[i, j]] == 1:
#descriptors of all the regions (in a list)
des_vector_byregion[mat_indexes[
i, j]] = np.concatenate(
(des_vector_byregion[mat_indexes[i, j]], des),
axis=0)
#number of descriptors in each region
number_of_kp_region[mat_indexes[
i, j]] = np.concatenate(
(number_of_kp_region[mat_indexes[i, j]],
np.array([len(kp)])),
axis=0)
else:
des_vector_byregion[mat_indexes[i, j]] = des
number_of_kp_region[mat_indexes[i, j]] = np.array(
[len(kp)])
filled[mat_indexes[i, j]] = 1
#print des_vector_byregion
#print number_of_kp_region
#for evaluation
name1 = im.split("/")[-1]
name = name1.split("_")[0]
if name in class_names:
index = class_names.index(name)
labels.append(index)
else:
class_names.append(name)
index = class_names.index(name)
labels.append(index)
counter += 1
#measure the time to compute the description of each image (divide time elapsed by # of images)
elapsed_time = (time.time() - start_time) / len(imList)
print 'Time to compute detector and descriptor for each image = ' + str(
elapsed_time)
f.write(
'Average time to compute detector and descriptor for each image = ' +
str(elapsed_time) + '\n')
n_images = counter - 1
average_words = sum(number_of_kp) / float(len(number_of_kp))
#all the descriptors together
des_vector = np.concatenate(np.array(des_vector_byregion))
print 'Total number of features = ' + str(len(des_vector))
f.write('Total number of features obtained = ' + str(len(des_vector)) +
'\n')
print 'Average number of keypoints per image = ' + str(average_words)
f.write('Average number of keypoints per image = ' + str(average_words) +
'\n')
#################################################################
#
# Image and Keypoint sampling
#
#################################################################
rand_indexes = []
nmi_indexes = []
for iteraction in range(0, rep):
print "\nIteraction #" + str(iteraction + 1) + '\n'
f.write("\nIteraction #" + str(iteraction + 1) + '\n')
print 'Sampling images and keypoints prior to codebook computation...'
if imsample != "NONE":
sampleKp = sampleKeypoints.SamplingImandKey(
n_images, number_of_kp, average_words, percentage)
sampleallKp = sampleAllKeypoints.SamplingAllKey(percentage)
names_sampling = np.array(["SAMPLEI", "SAMPLEP"])
sample_method = np.array([sampleKp, sampleallKp])
#Get the detector passed in the -g argument
index = np.where(names_sampling == imsample)[0]
if index.size > 0:
sampling_to_use = sample_method[index[0]]
else:
print 'Wrong sampling method passed in the -g argument. Options: NONE, SAMPLEI, SAMPLEP'
sys.exit()
#FOR RESULTS FILE
sampling_to_use.writeFile(f)
des_vector_sampled = sampling_to_use.sampleKeypoints(des_vector)
print 'Total number of features after sampling = ' + str(
len(des_vector_sampled))
f.write('Total number of features after sampling = ' +
str(len(des_vector_sampled)) + '\n')
print 'Images and keypoints sampled...'
else:
print 'No sampling method chosen'
#FOR RESULTS FILE
f.write(
"No method of keypoint sampling chosen. Use all keypoints for codebook construction \n"
)
des_vector_sampled = des_vector
#################################################################
#
# Codebook computation
#
#################################################################
print 'Obtaining codebook...'
#save current time
start_time = time.time()
#Get detector classes
codebook_kmeans = KMeans1.KMeans1(size)
codebook_birch = Birch.Birch(size)
codebook_minibatch = minibatch.MiniBatch(size)
codebook_randomv = randomSamplesBook.RandomVectors(size)
codebook_allrandom = allrandom.AllRandom(size)
names_codebook = np.array(
["KMEANS", "BIRCH", "MINIBATCH", "RANDOMV", "RANDOM"])
codebook_algorithm = np.array([
codebook_kmeans, codebook_birch, codebook_minibatch,
codebook_randomv, codebook_allrandom
])
#Get the detector passed in the -c argument
index = np.where(names_codebook == codebook)[0]
if index.size > 0:
codebook_to_use = codebook_algorithm[index[0]]
else:
print 'Wrong codebook construction algorithm name passed in the -c argument. Options: KMEANS, MINIBATCH, RANDOMV and RANDOM'
sys.exit()
#FOR RESULTS FILE
codebook_to_use.writeFileCodebook(f)
#Get centers and projections using codebook algorithm
centers, projections = codebook_to_use.obtainCodebook(
des_vector_sampled, des_vector)
#compute the number of unique descriptor vectors
codebook_randomv.unique_vectors(centers)
elapsed_time = (time.time() - start_time)
print 'Time to compute codebook = ' + str(elapsed_time)
f.write('Time to compute codebook = ' + str(elapsed_time) + '\n')
#################################################################
#
# Obtain Histogram
#
#################################################################
des_byregion = des_vector_byregion
numkp_region = number_of_kp_region
hist_total = []
for level in range(levels - 1, -1, -1):
print 'Level = ' + str(level)
n_regions = np.power(4, level)
for i in range(0, n_regions):
print 'Obtaining histograms...'
#print 'projection shape = '+ str(projections.shape)
#print 'size = ' + str(size)
#print 'n of images = ' + str(n_images)
#print 'number of kp' + str(number_of_kp)
#print len(des_vector_byregion)
#print len(des_vector_byregion[0])
#print len(des_vector_byregion[0][0])
result = scipy.cluster.vq.vq(np.array(des_byregion[i]),
centers)
projections_region = result[0]
#print 'projections = ' + str(projections_region)
#print n_images
#print number_of_kp_region[i]
#print len(number_of_kp_region)
#print len(number_of_kp_region[0])
hist = histogram.computeHist(projections_region, size,
n_images, numkp_region[i])
#print hist
print 'Histograms obtained'
#print hist
################################################################
#
# Feature selection
#
#################################################################
print 'Number of visual words = ' + str(len(hist[0]))
if fselec != "NONE":
print 'Applying feature selection to descriptors...'
filter_max = filterMax.WordFilterMax(fselec_perc[0])
filter_min = filterMin.WordFilterMin(fselec_perc[1])
filter_maxmin = filterMaxMin.WordFilterMaxMin(
fselec_perc[0], fselec_perc[1])
names_filter = np.array(["FMAX", "FMIN", "FMAXMIN"])
filter_method = np.array(
[filter_max, filter_min, filter_maxmin])
#Get the detector passed in the -f argument
index = np.where(names_filter == fselec)[0]
if index.size > 0:
filter_to_use = filter_method[index[0]]
else:
print 'Wrong codebook construction algorithm name passed in the -f argument. Options: NONE, FMAX, FMIN, FMAXMIN'
sys.exit()
hist = filter_to_use.applyFilter(hist, size, n_images)
#FOR RESULTS FILE
filter_to_use.writeFile(f)
new_size = hist.shape[1]
print 'Visual words Filtered'
print 'Number of visual words filtered = ' + str(size -
new_size)
f.write("Number of visual words filtered = " +
str(size - new_size) + '\n')
print 'Final number of visual words = ' + str(new_size)
f.write('Final number of visual words = ' + str(new_size) +
'\n')
else:
#FOR RESULTS FILE
filter_min = filterMin.WordFilterMin(0)
hist = filter_min.applyFilter(hist, size, n_images)
new_size = hist.shape[1]
print 'Number of visual words filtered = ' + str(size -
new_size)
f.write("No feature selection applied \n")
#################################################################
#
# Histogram Normalization
#
#################################################################
if histnorm != "NONE":
#Get detector classes
norm_sbin = simpleBinarization.SimpleBi()
norm_tfnorm = tfnorm.Tfnorm()
norm_tfidf = tfidf.TfIdf()
norm_tfidf2 = tfidf2.TfIdf2()
norm_tfidfnorm = tfidfnorm.TfIdfnorm()
norm_okapi = okapi.Okapi(average_words)
names_normalization = np.array([
"SBIN", "TFNORM", "TFIDF", "TFIDF2", "TFIDFNORM",
"OKAPI"
])
normalization_method = np.array([
norm_sbin, norm_tfnorm, norm_tfidf, norm_tfidf2,
norm_tfidfnorm, norm_okapi
])
#Get the detector passed in the -h argument
index = np.where(names_normalization == histnorm)[0]
if index.size > 0:
normalization_to_use = normalization_method[index[0]]
new_hist = normalization_to_use.normalizeHist(
hist, new_size, n_images)
else:
print 'Wrong normalization name passed in the -h argument. Options: SBIN, TFNORM, TFIDF and TFIDF2'
sys.exit()
#FOR RESULTS FILE
normalization_to_use.writeFile(f)
else:
#FOR RESULTS FILE
f.write("No histogram normalization applied\n")
new_hist = hist
hist_total.append(np.array(new_hist))
#concatenate des_vector_byregion TODOOOOOOOOOO
des_vector_aux = []
number_of_kp_aux = []
if level != 0:
side = 4
ntimes = int(np.power(4, level - 1))
for h in range(0, ntimes):
#print len(des_byregion)
#print h*side
#print (h+1)*side
des_vector_aux.append(
np.concatenate(des_byregion[h * side:(h + 1) * side],
axis=0))
count = 0
for n in numkp_region[h * side:(h + 1) * side]:
if count != 0:
sum_np = [sum(x) for x in zip(sum_np, n)]
else:
sum_np = n
count = count + 1
number_of_kp_aux.append(sum_np)
des_byregion = des_vector_aux
numkp_region = number_of_kp_aux
#print hist_total
hist_total = np.concatenate(hist_total, axis=1)
print len(hist_total[0])
#################################################################
#
# Clustering of the features
#
#################################################################
#save current time
start_time = time.time()
#Get detector classes
clust_dbscan = Dbscan.Dbscan(dist)
clust_kmeans = KMeans1.KMeans1([nclusters])
clust_birch = Birch.Birch(nclusters)
clust_meanSift = meanSift.MeanSift(nclusters)
clust_hierar1 = hierarchicalClustering.Hierarchical(nclusters, dist)
clust_hierar2 = hierarchicalClustScipy.HierarchicalScipy(dist)
clust_community = communityDetection.CommunityDetection(dist)
names_clustering = np.array([
"DBSCAN", "KMEANS", "BIRCH", "MEANSIFT", "HIERAR1", "HIERAR2",
"COMM"
])
clustering_algorithm = np.array([
clust_dbscan, clust_kmeans, clust_birch, clust_meanSift,
clust_hierar1, clust_hierar2, clust_community
])
#Get the detector passed in the -a argument
index = np.where(names_clustering == clust)[0]
if index.size > 0:
clustering_to_use = clustering_algorithm[index[0]]
else:
print 'Wrong clustering algorithm name passed in the -a argument. Options: DBSCAN, KMEANS, BIRCH, MEANSIFT, HIERAR1, HIERAR2, COMM'
sys.exit()
clusters = clustering_to_use.obtainClusters(hist_total)
#FOR RESULTS FILE
clustering_to_use.writeFileCluster(f)
elapsed_time = (time.time() - start_time)
print 'Time to run clustering algorithm = ' + str(elapsed_time)
f.write('Time to run clustering algorithm = ' + str(elapsed_time) +
'\n')
print 'Number of clusters obtained = ' + str(max(clusters) + 1)
f.write('Number of clusters obtained = ' + str(max(clusters) + 1) +
'\n')
print 'Clusters obtained = ' + str(np.asarray(clusters))
#date_time = datetime.datetime.now().strftime('%b-%d-%I%M%p-%G')
#np.savetxt('saveClusters_'+date_time+'_.txt', clusters, '%i', ',')
##ADDED
##################################################################
##
## Create folder with central images for each cluster
##
##################################################################
#dir_results = 'Results_' + im_dataset_name + '_SPM_' + date_time
##obtain representative images for each cluster
#central_ims = clust_community.obtainCenteralImages(new_hist, clusters)
#central_folder = os.path.join(dir_results,'CenterImages')
#if not os.path.exists(central_folder):
#os.makedirs(central_folder)
#count=0
#for central_im in central_ims:
#filename = os.path.join(central_folder,'Cluster_'+str(count)+'.jpg')
#img = cv2.imread(imPaths[central_im],1)
#cv2.imwrite(filename, img)
#count = count + 1
##ADDED
##################################################################
##
## Separate Clusters into folders
##
##################################################################
#clusters_folder = os.path.join(dir_results,'Clusters')
#if not os.path.exists(clusters_folder):
#os.makedirs(clusters_folder)
#clust_dir = []
#for iclust in range(0,nclusters):
#direc = os.path.join(clusters_folder,'Cluster_'+str(iclust))
#if not os.path.exists(direc):
#os.makedirs(direc)
#clust_dir.append(direc)
#for im in range(0,len(imPaths)):
#im_name = imPaths[im].split('/')[-1]
##print clust_dir[int(clusters[im])]
#filename = os.path.join(clust_dir[int(clusters[im])],im_name)
##print filename
#img = cv2.imread(imPaths[im],1)
#cv2.imwrite(filename, img)
#################################################################
#
# Evaluation
#
#################################################################
users = 0
if users == 1:
rand_index = evaluationUsers.randIndex(clusters)
rand_indexes.append(rand_index)
print 'rand_index = ' + str(rand_index)
f.write("Rand Index = " + str(rand_index) + "\n")
else:
if len(clusters) == len(labels):
f.write("\nResults\n")
f.write('Clusters Obtained = ' + str(np.asarray(clusters)))
f.write('Labels = ' + str(np.asarray(labels)))
rand_index = metrics.adjusted_rand_score(labels, clusters)
rand_indexes.append(rand_index)
print 'rand_index = ' + str(rand_index)
f.write("Rand Index = " + str(rand_index) + "\n")
NMI_index = metrics.normalized_mutual_info_score(
labels, clusters)
nmi_indexes.append(NMI_index)
print 'NMI_index = ' + str(NMI_index)
f.write("NMI Index = " + str(NMI_index) + "\n")
if rep > 1:
f.write("\nFINAL RESULTS\n")
f.write("Avg Rand Index = " + str(float(sum(rand_indexes)) / rep) +
"\n")
f.write("Std Rand Index = " + str(statistics.stdev(rand_indexes)) +
"\n")
f.write("Avg NMI Index = " + str(float(sum(nmi_indexes)) / rep) + "\n")
f.write("Std NMI Index = " + str(statistics.stdev(nmi_indexes)) + "\n")
f.close()
