for subset in ['train', 'test']:
print("loading subset %s..." % subset)
d = data.load(subset)
print("computing image moment statistics for subset %s..." % subset)
centroids = np.zeros((d.shape[0], 2))
major_axes = np.zeros((d.shape[0], ))
minor_axes = np.zeros((d.shape[0], ))
angles = np.zeros((d.shape[0], ))
for k, im in enumerate(d):
if k % 1000 == 0:
print("image %d of %d..." % (k + 1, d.shape[0]))
a = data.uint_to_float(im)
ms = skimage.measure.moments(a.astype('float64'), order=1)
x_centroid = ms[1, 0] / ms[0, 0]
y_centroid = ms[0, 1] / ms[0, 0]
mc = skimage.measure.moments_central(a.astype('float64'),
y_centroid,
x_centroid,
order=2)
mu = mc / ms[0, 0]
mudiff = mu[2, 0] - mu[0, 2]
angle = 0.5 * np.arctan2(2.0 * mu[1, 1], mudiff) * (180.0 / np.pi)
if angle < 0.0:
angle += 180
equivalent_diameter,euler_number,extent,filled_area,orientation,perimeter,solidity, centroid[0], centroid[1]
features = {"hu":[],"ORB":[],"tutorial":[],"haralick":[],"lbp":[],"pftas":[],
"zernike_moments":[],"image_size":[]}
image_shapes = np.asarray([img.shape for img in data.load('train')]).astype(np.float32)
# print image_shapes.shape
# pause()
# img_shp = (64,64)
# resized_imgs = np.empty((len(images),np.prod(img_shp)),dtype="uint8")
report = [int((j+1)*images.shape[0]/100.) for j in range(100)]
# count = 0
for i,img in enumerate(images):
img_o = img.copy()
img = data.uint_to_float(img)
# Hu moments
hu = cv2.HuMoments(cv2.moments(img)).flatten()
features["hu"].append(hu)
# ORB
# img_ = cv2.resize(img_o,img_shp)
# orb = cv2.ORB(nfeatures=1,patchSize=5,nlevels=4,edgeThreshold=5)
# kp, des = orb.detectAndCompute(img_,None)
# if (des is None):
# des = np.zeros((32,))
# count+=1
# else:
# des = des.flatten()
# features["ORB"].append(des)
for subset in ['train', 'test']:
print "loading subset %s..." % subset
d = data.load(subset)
print "computing image moment statistics for subset %s..." % subset
centroids = np.zeros((d.shape[0], 2))
major_axes = np.zeros((d.shape[0],))
minor_axes = np.zeros((d.shape[0],))
angles = np.zeros((d.shape[0],))
for k, im in enumerate(d):
if k % 1000 == 0:
print "image %d of %d..." % (k + 1, d.shape[0])
a = data.uint_to_float(im)
ms = skimage.measure.moments(a.astype('float64'), order=1)
x_centroid = ms[1, 0] / ms[0, 0]
y_centroid = ms[0, 1] / ms[0, 0]
mc = skimage.measure.moments_central(a.astype('float64'), y_centroid, x_centroid, order=2)
mu = mc / ms[0, 0]
mudiff = mu[2, 0] - mu[0, 2]
angle = 0.5 * np.arctan2(2.0 * mu[1, 1], mudiff) * (180.0 / np.pi)
if angle < 0.0:
angle += 180
covar = np.array([[mu[2, 0], mu[1, 1]], [mu[1, 1], mu[0, 2]]])
eigvals, eigvecs = np.linalg.eigh(covar)
