def demo():
import scipy.cluster.vq as vq
## load and show image
img = Image.open('images/117054.jpg')
K = 2
pl.figure(str(K))
pl.subplot(2, 3, 1)
pl.imshow(img)
pl.subplot(2, 3, 4)
pl.imshow(img)
## extract features from image (step size = 7)
X, L = pa2.getfeatures(img, 7)
X = vq.whiten(X.T)
print "Start Learning"
C, Y = vq.kmeans2(X, K, iter=1000, minit='random')
Y = Y + 1
# make segmentation image from labels
segm = pa2.labels2seg(Y, L)
pl.subplot(2, 3, 2)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
pl.subplot(2, 3, 3)
pl.imshow(csegm)
Y = GaussianMixture(n_components=K, covariance_type='full').fit(X)
Y = Y.predict(X) + 1
# Y = EMGMM.clustering(X, K)
# make segmentation image from labels
segm = pa2.labels2seg(Y, L)
pl.subplot(2, 3, 5)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
pl.subplot(2, 3, 6)
pl.imshow(csegm)
# # Y = MeanShift.clustering((vq.whiten(X.T)).T, 5)
# Y = MeanShift(bandwidth=0.9).fit((vq.whiten(X.T)))
# Y = Y.labels_ + 1
# # make segmentation image from labels
# segm = pa2.labels2seg(Y,L)
# pl.subplot(3,3,8)
# pl.imshow(segm)
# # color the segmentation image
# csegm = pa2.colorsegms(segm, img)
# pl.subplot(3,3,9)
# pl.imshow(csegm)
pl.show()
def main():
exp_dict = {}
for data in DATA:
img = Image.open(os.path.join(IMGPATH, data))
pl.subplot(1, 3, 1)
pl.imshow(img)
X_raw, L = pa2.getfeatures(img, 7)
X = vq.whiten(X_raw.T)
for bd in BANDS:
clf = im.GaussianMeanShift(itera=5, bandwidth=bd)
clf.fit_x(X)
clf.cluster()
Y = clf.get_result() + 1
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
csegm = pa2.colorsegms(segm, img)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.savefig(
os.path.join(OUTPATH,
data + '_GMS_' + str(bd) + '_processed.jpg'))
pl.show()
return
def demo():
import scipy.cluster.vq as vq
## load and show image
img = Image.open('images/12003.jpg')
pl.subplot(3,3,1)
pl.imshow(img)
pl.subplot(3,3,4)
pl.imshow(img)
pl.subplot(3,3,7)
pl.imshow(img)
## extract features from image (step size = 7)
X,L = pa2.getfeatures(img, 7)
X = vq.whiten(X.T)
for i in range(3):
Y = MeanShift(bandwidth=0.9*(i+1)).fit(X)
Y = Y.labels_ + 1 # Use matlab 1-index labeling
# make segmentation image from labels
segm = pa2.labels2seg(Y,L)
pl.subplot(3,3,(3 * i + 2))
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
pl.subplot(3,3,(3 * i + 3))
pl.imshow(csegm)
pl.show()
def demo():
import scipy.cluster.vq as vq
## load and show image
img = Image.open('images/12003.jpg')
pl.subplot(1, 3, 1)
pl.imshow(img)
## extract features from image (step size = 7)
X, L = pa2.getfeatures(img, 7)
## Call kmeans function in scipy. You need to write this yourself!
C, Y = vq.kmeans2(vq.whiten(X.T), 2, iter=1000, minit='random')
Y = Y + 1 # Use matlab 1-index labeling
##
# make segmentation image from labels
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.show()
def main():
exp_dict = {}
for data2 in DATA:
img = Image.open(os.path.join(IMGPATH, data2))
pl.subplot(1, 3, 1)
pl.imshow(img)
X_raw, L = pa2.getfeatures(img, 7)
X = vq.whiten(X_raw.T)
for bdp in BANDS:
for bdc in BANDS:
WKM = im.WeightGMeanshift(itera=5,
chrominance_bandwidth=bdp,
location_bandwidth=bdc)
WKM.fit_x(X)
WKM.cluster()
Y = WKM.get_result() + 1
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
csegm = pa2.colorsegms(segm, img)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.savefig(
os.path.join(
OUTPATH, data2 + '_WGMS_' + str(bdp) + '_' + str(bdc) +
'_processed.jpg'))
pl.show()
return
def demo():
import scipy.cluster.vq as vq
## load and show image
img = Image.open('../images/12003.jpg')
pl.subplot(1, 3, 1)
pl.imshow(img)
## extract features from image (step size = 7)
X, L = pa2.getfeatures(img, 7)
## Call kmeans function in scipy. You need to write this yourself!
# C,Y = vq.kmeans2(vq.whiten(X.T), 2, iter=1000, minit='random')
d = 4
K = 4
N = X.shape[1]
Mju = RandomMju(150, 155, K, d)
##########Kmeans##########
#Y = Clustering_Kmeans(Mju, X, K, N, d, 1000, 0.000000000001)
##########EM#########
#d = 4
#K = 2
#SIGMA = RandomSIGMA(100, 110, K, d)
#Pi = [0.1, 0.1, 0.05, 0.1, 0.1, 0.1, 0.1, 0.1, 0.05, 0.2]
#Y = Clustering_EM(Mju, SIGMA, Pi, X, K, d, X.shape[1], 800, 0.00000000000001)
###################
#####Meanshift####
h = 45
maxRound = 30
limitation = 0.00001
cluster_gap = 8
X_peak = Peak_MeanShift(X, d, N, h, maxRound, limitation)
cluster = Clustering_MeanShift(X, X_peak, N, d, cluster_gap)
Y = cluster[1]
##########################
# Use matlab 1-index labeling
# make segmentation image from labels
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
picname = "seastar"
title = "Menshift, h=" + str(h)
pl.title(title)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.savefig('/Users/gaobrook/Desktop/Clustering/problem2/' + title + '_' +
picname + '.png',
dpi=200)
pl.show()
def pic_prod(Y, pic, pic_label):
Y = Y + 1 # Use matlab 1-index labeling
##
pl.figure(pic)
pl.subplot(1, 3, 1)
pl.title('original')
pl.imshow(img)
# make segmentation image from labels
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
pl.title(pic_label)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
def main():
import scipy.cluster.vq as vq
## load and show image
img = Image.open('images/resized.jpg')
pl.subplot(1,3,1)
pl.imshow(img)
## extract features from image (step size = 7)
X,L = pa2.getfeatures(img, 9)
X_T = np.transpose(X)
print(X_T)
"""
Change the methods to one of them: KMeans(X_T, k), EM_GMM(X_T, k), MeanShift(X_T, h)
"""
ms = EM_GMM(X_T, k=5)
ms.fit()
Y = ms.evaluate()
"""
# Sample code, using the methods in scipy.
# Call kmeans function in scipy. You need to write this yourself!
C,Y = vq.kmeans2 (vq.whiten(X.T), 2, iter=1000, minit='random')
Y = Y + 1 # Use matlab 1-index labeling
##
"""
# make segmentation image from labels
segm = pa2.labels2seg(Y,L)
pl.subplot(1,3,2)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
pl.subplot(1,3,3)
pl.imshow(csegm)
pl.show()
def main():
exp_dict = {}
for data in DATA:
img = Image.open(os.path.join(IMGPATH, data))
pl.subplot(1, 3, 1)
pl.imshow(img)
X_raw, L = pa2.getfeatures(img, 7)
X = vq.whiten(X_raw.T)
for method in METHODS[:2]:
for k in K:
if (method == 'KM'):
clf = im.Kmeans(k=k)
if (method == 'EMGMM'):
clf = im.EMGMM(k=k)
clf.fit_x(X)
clf.cluster()
Y = clf.get_result() + 1
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
csegm = pa2.colorsegms(segm, img)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.savefig(
os.path.join(
OUTPATH,
data + '_' + method + '_' + str(k) + '_processed.jpg'))
pl.show()
for data2 in DATA:
img = Image.open(os.path.join(IMGPATH, data2))
pl.subplot(1, 3, 1)
pl.imshow(img)
X_raw, L = pa2.getfeatures(img, 7)
X = vq.whiten(X_raw.T)
for l in Lambda:
for k in K:
WKM = im.WeightedKmeans4D(k=k, Lambda=l)
WKM.fit_x(X)
WKM.cluster()
Y = WKM.get_result() + 1
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
csegm = pa2.colorsegms(segm, img)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.savefig(
os.path.join(
OUTPATH, data2 + '_WKM_' + str(k) + '_' + str(l) +
'_processed.jpg'))
pl.show()
return
def demo():
import scipy.cluster.vq as vq
## load and show image
img = Image.open('../images/12003.jpg')
pl.subplot(1, 3, 1)
pl.imshow(img)
## extract features from image (step size = 7)
X, L = pa2.getfeatures(img, 7)
## Call kmeans function in scipy. You need to write this yourself!
# C,Y = vq.kmeans2(vq.whiten(X.T), 2, iter=1000, minit='random')
d = 4
K = 4
N = X.shape[1]
Mju = RandomMju(150, 155, K, d)
##########Kmeans##########
#Y = Clustering_Kmeans(Mju, X, K, N, d, 1000, 0.000000000001)
##########EM#########
#d = 4
#K = 2
#SIGMA = RandomSIGMA(100, 110, K, d)
#Pi = [0.1, 0.1, 0.05, 0.1, 0.1, 0.1, 0.1, 0.1, 0.05, 0.2]
#Y = Clustering_EM(Mju, SIGMA, Pi, X, K, d, X.shape[1], 800, 0.00000000000001)
###################
#####Meanshift####
h = 45
maxRound = 30
limitation = 0.00001
cluster_gap = 8
X_peak = Peak_MeanShift(X, d, N, h, maxRound, limitation)
cluster = Clustering_MeanShift(X, X_peak, N, d, cluster_gap)
Y = cluster[1]
##########################
# Use matlab 1-index labeling
# make segmentation image from labels
segm = pa2.labels2seg(Y, L)
pl.subplot(1, 3, 2)
pl.imshow(segm)
# color the segmentation image
csegm = pa2.colorsegms(segm, img)
picname ="seastar"
title = "Menshift, h="+str(h)
pl.title(title)
pl.subplot(1, 3, 3)
pl.imshow(csegm)
pl.savefig('/Users/gaobrook/Desktop/Clustering/problem2/'+title+'_'+picname+'.png', dpi=200)
pl.show()