def entropy(img,dim):
all_entropy = np.zeros(img.shape[dim])
if dim == 0:
for i in range(img.shape[dim]):
all_entropy[i] = shannon_entropy(img[i,:,:])
threshold = np.sort(all_entropy)[-33]
reduced_img = img[all_entropy > threshold,:,:]
elif dim == 1:
for i in range(img.shape[dim]):
all_entropy[i] = shannon_entropy(img[:,i,:])
threshold = np.sort(all_entropy)[-33]
reduced_img = img[:,all_entropy > threshold,:]
elif dim == 2:
for i in range(img.shape[dim]):
all_entropy[i] = shannon_entropy(img[:,:,i])
threshold = np.sort(all_entropy)[-33]
reduced_img = img[:,:,all_entropy > threshold]
return(reduced_img)
def evaluate_entropy_instant(reference_frame, rendition_frame):
# Function that computes the difference in Shannon entropy between
# two images
entropy_difference = shannon_entropy(
reference_frame) - shannon_entropy(rendition_frame)
return entropy_difference
def entropy(image):
if image.ndim == 2:
return shannon_entropy(image)
if image.ndim == 3:
entropies = []
for i in range(image.shape[0]):
entropies.append(shannon_entropy(image[i]))
return entropies
def calcualte_MI(img, img1, img2):
# 分别计算A,C和B,C的MI值，在图像融合中，MI值为二者之和，这里取了平均
img = np.array(img)
img1 = np.array(img1)
img2 = np.array(img2)
mi1 = shannon_entropy(img1) + shannon_entropy(img) - ComEntropy(img1, img)
mi2 = shannon_entropy(img2) + shannon_entropy(img) - ComEntropy(img2, img)
mi = (mi1 + mi2) / 2
return round(mi, 3)
def entropy(reference_frame, rendition_frame):
"""
Function that computes the difference in Shannon entropy between
two images
"""
entropy_difference = shannon_entropy(reference_frame) - shannon_entropy(rendition_frame)
return entropy_difference
def glcm_infmescor_1(matrix: np.ndarray, eps: float = 10e-3) -> np.ndarray:
"""
Measures of Correlation 1
"""
hxy = -np.sum(matrix * np.log(eps + matrix))
hxy_1 = -np.sum(
matrix * np.log(eps + np.sum(matrix, axis=0) * np.sum(matrix, axis=1)))
hx = shannon_entropy(np.sum(matrix, axis=0))
hy = shannon_entropy(np.sum(matrix, axis=1))
measure_1 = (hxy - hxy_1) / max(hx, hy)
return measure_1
def compute_glcm_features(retinal_image):
#Compute Entropy
glcm_image_entropy = shannon_entropy()
#Compute GLCM (Gray Level Co-occurrence Matrix)
glcm_image = greycomatrix()
#Compute GLCM Features
#Contrast
glcm_image_contrast = greycoprops(glcm_image, 'contrast')
#Dissimilarity
glcm_image_dissimilarity = greycoprops(glcm_image, 'dissimilarity')
#Homogeneity
glcm_image_homogeneity = greycoprops(glcm_image, 'homogeneity')
#Energy
glcm_image_energy = greycoprops(glcm_image, 'energy')
#Correlation
glcm_image_correlation = greycoprops(glcm_image, 'correlation')
#ASM
glcm_image_ASM = greycoprops(glcm_image, 'ASM')
def open_and_get_metrics(img_path):
"""Open the image from the static folder and retrieve its metrics for classification
Args:
img_path: The path of the image within the static folder, or uploads folder.
Returns:
image_data: A dictionary of all the collected image metrics
"""
img = io.imread(img_path)
flat_img = flatten(img)
entropy = shannon_entropy(img, base=2)
mean_colors = get_mean_colors(flat_img)
luminance = (0.2126 * mean_colors[0]) + (0.7152 * mean_colors[1]) + (
0.0722 * mean_colors[2])
contrast = get_contrast(flat_img)
contour = get_contour(img)
image_data = {
'file_name': img_path,
'year': img_path.split('/')[-1].split('.')[0],
'shannon_entropy': entropy,
'mean_color_r': mean_colors[0],
'luminance': luminance,
'contrast': contrast,
'contour': contour
}
print(image_data)
return image_data
def get_hand_crafted(one_image):
""" Extracts various features out of the given image
:param array one_image: the image from which features are to be extracted
:return: the features associated with this image
:rtype: Numpy array of size (38, 1)
"""
#Select wavelet decomposition level so as to have the
#same number of approximation coefficients
if (one_image.shape[0] == 1000):
wavedec_level = 9
elif (one_image.shape[0] == 64):
wavedec_level = 5
hist = histogram(one_image, nbins=20, normalize=True)
features = hist[0]
blob_lo = blob_log(one_image,
max_sigma=2.5,
min_sigma=1.5,
num_sigma=5,
threshold=0.05)
shape_ind = shape_index(one_image)
shape_hist = np.histogram(shape_ind, range=(-1, 1), bins=9)
shan_ent = shannon_entropy(one_image)
max_val = one_image.max()
min_val = one_image.min()
variance_val = np.var(one_image)
wavelet_approx = pywt.wavedec2(one_image, 'haar',
level=wavedec_level)[0].flatten()
features = np.concatenate([
features, [blob_lo.shape[0]], shape_hist[0], [shan_ent], [max_val],
[min_val], [variance_val], wavelet_approx
])
return features
def feature_extraction(path):
image_rgb = skimage.io.imread(path)
image_gray = color.rgb2gray(image_rgb)
eight_bit = np.uint8(image_gray)
cm = feature.greycomatrix(eight_bit, [1], [0])
contrast = feature.greycoprops(cm)
dissimilarity = feature.greycoprops(cm, 'dissimilarity')
homogeneity = feature.greycoprops(cm, 'homogeneity')
ASM = feature.greycoprops(cm, 'ASM')
energy = feature.greycoprops(cm, 'energy')
correlation = feature.greycoprops(cm, 'correlation')
ent = measure.shannon_entropy(cm)
thr = flt.threshold_otsu(image_gray)
img = image_gray < thr
filled = scipy.ndimage.morphology.binary_fill_holes(img)
label = measure.label(filled)
for cell in measure.regionprops(label):
eccentricity = cell.eccentricity
return [contrast[0][0], dissimilarity[0][0], homogeneity[0][0], ASM[0][0], energy[0][0], correlation[0][0], ent, eccentricity]
def compute_props(image, masks_labelled, features, img):
properties = regionprops(masks_labelled, intensity_image = image)
#for each nuclei, save the corresponding features
for region in properties:
area = float(region.area)
bboxarea = float(region.bbox_area)
perimeter = region.perimeter
eccentricity = region.eccentricity
total_intensity = float(np.sum(region.intensity_image))
mean_intensity = region.mean_intensity
solidity = region.solidity
centroid = region.centroid
#feats = compute_feats(region.intensity_image,kernels)
#pw = power(region.intensity_image, kernels[0])
lbp = local_binary_pattern(region.intensity_image, 3, 3, method = 'ror')
entr = compute_entropy(region.intensity_image)
shannon_entr = shannon_entropy(region.intensity_image)
# =============================================================================
# res = {"Image": img, "Area": area, "BBox_Area": bboxarea, "Perimeter": perimeter, "Eccentricity": eccentricity,
# "Total Intensity": total_intensity, "Mean Intensity": mean_intensity, "Solidity": solidity, "Gabor Mean":
# feats[0,0], "Gabor Var": feats[0,1], "Gabor Amplitude": np.mean(pw), "Gabor Energy": np.sum(pw**2),
# "Local Binary Pattern": compute_energy(lbp), "Entropy": compute_energy(entr)}
# =============================================================================
res = {"Image": img, "Area": area, "BBox_Area": bboxarea, "Perimeter": perimeter, "Eccentricity": eccentricity,
"Total Intensity": total_intensity, "Mean Intensity": mean_intensity, "Solidity": solidity,
"Local Binary Pattern": compute_energy(lbp), "Entropy": compute_amplitude(entr),
"Shannon Entropy": shannon_entr, "Centroid_y": centroid[0], "Centroid_x": centroid[1]}
row = len(features)
features.loc[row] = res
return features
def svm_clf(path):
data_set = []
data1 =fits.open(path)
data = data1[0].data
data = data[132:991, 132:991]
# 检测图像背景，获得去除背景后的图像
# Detect the background of the picture and obtain the picture after removing the background
data = data.astype(np.float64)
bkg = sep.Background(data, mask=None, bw=64, bh=64, fw=3, fh=3)
data_sub = data - bkg # 得到去噪后的数据
objects = sep.extract(data_sub, 2.5, err=bkg.globalrms, deblend_nthresh=1)
# 获得亮星数目
# Get the number of bright stars
number = 0
for i in range(len(objects)):
a = objects[i][15]
b = objects[i][16]
a = max(a, b)
b = min(a, b)
# 控制星象大小
# Control star size
if a < 32 and b > 2.5:
number = number + 1
else:
number = number
m1, s1 = np.mean(data_sub), np.std(data_sub)
data_sub = data_sub.astype(np.uint16)
# 获得灰度共生矩阵参数
# Obtain gray level co-occurrence matrix parameters
gray = color.rgb2gray(data_sub)
image = img_as_ubyte(gray)
bins = np.array([0, 16, 32, 48, 64, 80, 96, 112, 128, 144, 160, 176, 192, 208, 224, 240, 255]) # 16-bit
inds = np.digitize(image, bins)
max_value = inds.max() + 1
matrix_coocurrence = greycomatrix(inds, [1], [0, np.pi / 4, np.pi / 2, 3 * np.pi / 4], levels=max_value,
normed=False, symmetric=False)
cons = np.sum(contrast_feature(matrix_coocurrence)) / 4
diss = np.sum(dissimilarity_feature(matrix_coocurrence)) / 4
h**o = np.sum(homogeneity_feature(matrix_coocurrence)) / 4
asmm = np.sum(asm_feature(matrix_coocurrence)) / 4
ener = np.sum(energy_feature(matrix_coocurrence)) / 4
corr = np.sum(correlation_feature(matrix_coocurrence)) / 4
# 熵的计算
# Entropy calculation
shan = shannon_entropy(image)
data_set = [[m1,number,corr,s1,h**o,shan,asmm,ener,cons]]
# 加载保存好的模型进行预测
# Load the saved model for prediction
clf = joblib.load('./rf_clf.m')
a=clf.predict(data_set)
a=int(a[0])
print(a)
cnn_set=[path,m1,number,corr,s1,h**o,shan,asmm,ener,cons,diss,a]
return cnn_set
def shannon_split_weak(image):
entropies = [(shannon_entropy(image[:, i, :]), i)
for i in range(image.shape[1])]
entropies = sorted(entropies, key=lambda x: x[0], reverse=True)
return list(
list(
zip(*(entropies[args.threshold // 2:args.threshold] +
entropies[-args.threshold // 2:-args.threshold:-1])))[1])
def entropy_image(regionmask, intensity):
"""Compute Shannon Entropy of a given image
Args:
regionmask=binary image
intensity= intensity image
"""
feat = Entropy_Image([shannon_entropy((intensity * regionmask))])
return feat
def display_metrics(Method, Prediction, aTargetImage, computingTime):
number_of_angles = 22
number_of_distances = 2
prediction = Prediction
# target = Target;
target_image = aTargetImage
method = Method
prediction[0] = round(prediction[0] * prediction[1])
pred_angles = np.zeros(number_of_angles)
# target_angles = np.zeros(number_of_angles);
for i in range(len(pred_angles)):
pred_angles[i] = prediction[i + number_of_distances]
# target_angles[i] = target[i+number_of_distances];
setXRayParameters(prediction[0], prediction[1])
pred_image = bone_rotation(pred_angles)
# setXRayParameters(target[0], target[1]);
# target_image = bone_rotation(target_angles);
# diff = [];
#
# for i in range(number_of_distances):
# diff.append(abs(prediction[i]-target[i]));
entropy = shannon_entropy(pred_image)
SSIM = structural_similarity(pred_image, target_image)
MAE = mean_absolute_error(target_image, pred_image)
RMSE = root_mean_squared_error(target_image, pred_image)
RE = relative_error(target_image, pred_image)
ZNCC = zero_mean_normalised_cross_correlation(target_image, pred_image)
computing_time = computingTime
print('Prediction:', prediction)
# print('Target: ', target);
# print('SOD and SDD errors: ', diff);
print(
'Metrics: \n SSIM: %.8f \t MAE: %.8f \t RMSE: %.8f \t RE: %.8f \t ZNCC: %.8f \
\t Entropy: %8f' % (SSIM, MAE, RMSE, RE, ZNCC, entropy))
row = [[
method, prediction[0], prediction[1], pred_angles, entropy, SSIM, MAE,
RMSE, RE, ZNCC, computing_time
]]
df2 = pd.DataFrame(row,
columns=[
'Methods', 'SOD', 'SDD', 'Rotating Angles',
'Entropy', 'SSIM', 'MAE', 'RMSE', 'Relative Error',
'ZNCC', 'Time'
])
return pred_image, df2
def basic_statistical_features(image):
"""calculates the set of basic statistical features
Calculates the standard statistical features per channel every 10th percentile,
sum of the pixel values and different moments
Parameters
----------
image : 3D array, shape (M, N, C)
The input image with multiple channels.
Returns
-------
features : dict
dictionary including percentiles, moments and sum per channel
"""
# storing the feature values
features = dict()
for ch in range(image.shape[2]):
# percentiles
features["min_intensity_Ch" + str(ch + 1)] = image[:, :, ch].min()
features["percentile10_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.1)
features["percentile20_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.2)
features["percentile30_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.3)
features["percentile40_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.4)
features["percentile50_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.5)
features["percentile60_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.6)
features["percentile70_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.7)
features["percentile80_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.8)
features["percentile90_intensity_Ch" + str(ch + 1)] = np.percentile(
image[:, :, ch], 0.9)
features["max_intensity_Ch" + str(ch + 1)] = image[:, :, ch].max()
# pixel sum
features["total_intensity_Ch" + str(ch + 1)] = image[:, :, ch].sum()
# moments
features["mean_intensity_Ch" + str(ch + 1)] = image[:, :, ch].mean()
features["std_intensity_Ch" + str(ch + 1)] = image[:, :, ch].std()
features["kurtosis_intensity_Ch" + str(ch + 1)] = kurtosis(
image[:, :, ch].ravel())
features["skew_intensity_Ch" + str(ch + 1)] = skew(image[:, :,
ch].ravel())
features["shannon_entropy_Ch" + str(ch + 1)] = shannon_entropy(
image[:, :, ch])
return features
def _calc_features(self):
"""
calculate feature values
:return: feature values
"""
features = {}
unif = []
ent = []
for i in np.arange(self.glcm.shape[3]):
mat = self.glcm[:, :, 0, i]
feature_unif = (mat**2).sum()
unif.append(feature_unif)
feature_ent = shannon_entropy(mat)
ent.append(feature_ent)
matrix = self.matrix[:, :, 0, 0]
features['Uniformity'] = list(unif)
features['Invariant Uniformity'] = (matrix**2).sum()
features['GLCM Entropy'] = list(ent)
features['GLCM Invariant Entropy'] = shannon_entropy(matrix)
features['Correlation'] = greycoprops(self.glcm, 'correlation')[0]
aux_corr = greycoprops(self.matrix, 'correlation')
features['Invariant Correlation'] = float(aux_corr[0][0])
features['Dissimilarity'] = greycoprops(self.glcm, 'dissimilarity')[0]
aux_diss = greycoprops(self.matrix, 'dissimilarity')
features['Invariant Dissimilarity'] = float(aux_diss[0][0])
features['Contrast'] = greycoprops(self.glcm, 'contrast')[0]
aux_cont = greycoprops(self.matrix, 'contrast')
features['Invariant Contrast'] = float(aux_cont[0][0])
features['Homogeneity'] = greycoprops(self.glcm, 'homogeneity')[0]
aux_hom = greycoprops(self.matrix, 'homogeneity')
features['Invariant Homogeneity'] = float(aux_hom[0][0])
features['Energy'] = greycoprops(self.glcm, 'energy')[0]
aux_eng = greycoprops(self.matrix, 'energy')
features['Invariant Energy'] = float(aux_eng[0][0])
return features
def read_hyper_data(hs_img, directory_path, sheet_number, choose_best = False):
print("-------Begining to read hyperspectral data----------- ")
#if a sheet_number is provided, only read that sheet number
if sheet_number != None:
start_time = time.time()
data, sheet_cnt = sheet_to_array(hs_img, sheet_number)
end_time = time.time()
print("Total time taken during reading count_worksheets: " + str(end_time - start_time))
#preprocess the hs_img using data read from sheet
hyp_im = preprocess_hyperdata(data)
print("________Read sheet " + str(sheet_number)+ " of hyperspectral image.__________")
cv2.imwrite(directory_path+"/full_hyperspec_img.png", hyp_im)
return hyp_im
#if NO sheet_number is provided, read the entire workbook
else:
start_time = time.time()
datas, sheet_cnt = sheet_to_array(hs_img, sheet_number)
end_time = time.time()
print("Total time taken during reading count_worksheets: " + str(end_time - start_time))
imgs = []
entropies =[]
#iterates through the number of channels
for i in range(len(datas)):
#reads data value from the sheet of hs_img
data = datas[i]
#preprocess the hs_img using data read from sheet
hyp_im = preprocess_hyperdata(data)
#output preprocessed hyp_im to path
cv2.imwrite(directory_path+"/full_hyperspec_img_"+str(i)+".png", hyp_im)
#the following steps are for calculating the best hyper_img
imgs.append(hyp_im)
#calculate the entropy of image
entropy = measure.shannon_entropy(hyp_im)
# print("entropy"+str(i), entropy)
entropies.append(entropy)
if choose_best == True:
#output best preprocessed hs_img to path
best_hyper_index = np.argmax(entropies)
best_hyper_img = imgs[best_hyper_index]
cv2.imwrite(directory_path+"/full_hyperspec_img.png", best_hyper_img)
return best_hyper_img, best_hyper_img.shape
#check if all the hyperspectral image matrix have been read.
try:
hyperspectral_data = np.concatenate((hyperspectral_data, im[..., np.newaxis]), axis = 2)
except:
hyperspectral_data = im[..., np.newaxis]
assert(sheet_cnt == hyperspectral_data.shape[2])
print("________Read all the channels of hyperspectral image.__________")
return imgs
def getShapeValue(region, mode='area'):
# Making different orders based on the shape signatures
order = [ 0, 1, 2, 3, 4 ]
value = 0
# Choose which method to measure a shape signature
if mode == 'shannon': value = shannon_entropy(region.image)
elif mode == 'moi': value = momentOfInertia(region)
else: value = region.area
return value, order
def getShapeValue(region, mode='area'):
# Making different orders based on the shape signatures
order = [0, 1, 2, 3, 4]
value = 0
# Choose which method to measure a shape signature
if mode == 'shannon': value = shannon_entropy(region.image)
elif mode == 'moi': value = momentOfInertia(region)
else: value = region.area
return value, order
def add_noise2(array, exposure=None, bit_depth=None):
'''
returnes the snr as peak snr for reduced exposure images, returns shannon entropy for exposure = None.
'''
## Change bit_depth if necessary
if bit_depth == 16:
array = img_as_uint(array / np.max(array))
clip = 2**16 - 1
elif bit_depth == 8:
noise = img_as_ubyte(array / np.max(array))
clip = 2**8 - 1
else:
clip = None
if exposure == None:
noise = array
snr = shannon_entropy(noise)
else:
noise = np.clip(np.random.poisson(array / np.max(array) * exposure), 0,
clip)
snr = shannon_entropy(
noise) #compare_psnr(array/np.max(array)*exposure, noise)
return noise, snr
def segment(data_path, SAVE_PATH):
labels = [label for label in os.listdir(data_path)] #Duyet cac labels
kernel = np.ones((3,3))
if not os.path.exists(SAVE_PATH):
os.mkdir(SAVE_PATH)
for label in labels: #Duyet cac labels
dir = os.path.join(data_path, label)
for img_name in os.listdir(dir): #Duyet cac anh
im_path = os.path.join(dir, img_name)
img = cv2.imread(im_path, cv2.IMREAD_GRAYSCALE) #doc anh xam
imgsave_name = '{}'.format(img_name) #Ten anh de luu
save_directory = os.path.join(SAVE_PATH, label)
save_path = os.path.join(save_directory, imgsave_name)
h, w = img.shape
diff = 10
for i in range(w): #duyet canh ben tren
if img[0, i] > 220:
cv2.floodFill(img, None, (i, 0), 0, loDiff=diff, upDiff=diff)
for i in range(w-1, -1, -1): #duyet canh ben duoi
if img[h - 1, i] > 220:
cv2.floodFill(img, None, (i, h-1), 0, loDiff=diff, upDiff=diff)
img_org = copy.copy(img) #tao ban sao cua img
_, img = cv2.threshold(img,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU) #nhi phan hoa theo nguong OTSU
img = 255 - img #invert anh
img = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel) #toan tu open
img = cv2.dilate(img,kernel,iterations = 7) #toan tu gian no
x1, y1, x2, y2 = findROICoord(img)
x1 = np.maximum(x1 - 60, 0) #tranh tran khoi kich thuoc anh goc
y1 = np.maximum(y1 - 60, 0)
x2 = np.minimum(x2 + 60, w - 1)
y2 = np.minimum(y2 + 60, h - 1)
try:
img_org = img_org[y1:y2, x1:x2]
if not os.path.exists(save_directory):
os.mkdir(save_directory)
img_org = resize(img_org, (128, 128))
entropy = shannon_entropy(img_org)
if entropy <= 5.5:
print(entropy)
print(save_path)
continue
cv2.imwrite(save_path, img_org)
# clear_output()
except Exception as e:
print(im_path, save_path, str(e))
def prep_data(filename_path):
nii_output_dir = os.getenv('TMP_DIR_PATH',
"/Users/pmartyn/PycharmProjects/Thesis/tmp/")
if not os.path.exists(nii_output_dir):
os.makedirs(nii_output_dir)
p = 0.5
img = nib.load(filename_path)
affine = img.affine
img = img.get_fdata()
prob = Extractor().run(img)
mask = prob > p
brain = img[:]
brain[~mask] = 0
image_array = nib.Nifti1Image(brain, affine).get_fdata()
total_slices = image_array.shape[2]
image_entropies = []
for current_slice in range(0, total_slices):
image_data = np.rot90(
np.rot90(np.rot90(image_array[:, :, current_slice])))
gray_image_data = color.rgb2gray(image_data)
ent = measure.shannon_entropy(gray_image_data)
dictionary = {
"filename": current_slice,
"entropy-value": ent,
"image-data": gray_image_data
}
image_entropies.append(dictionary)
image_entropies.sort(key=itemgetter('entropy-value'), reverse=True)
for dictionary in image_entropies[:6]:
filename = dictionary["filename"]
image_data = dictionary["image-data"]
img_to_crop = Image.fromarray(image_data)
# The crop rectangle, as a (left, upper, right, lower)-tuple.
cropped = img_to_crop.crop((25, 20, 250, 275))
cropped = np.array(cropped)
io.imsave(nii_output_dir + str(filename) + ".jpeg", cropped)
def main():
os.chdir(sys.argv[1])
data = pd.read_csv("positions.csv")
folderpath = "icm/"
entropy = []
new_data = pd.DataFrame()
new_data['filename'] = pd.unique(data['filename'])
for imagename in pd.unique(data['filename']):
print(os.path.abspath(imagename))
e = shannon_entropy(imread(imagename + '.png'))
if e < 0:
e = 0
entropy.append(e)
new_data['entropy'] = entropy
new_data.to_csv("entropy.csv", index=False)
def feature_extraction(image):
img = cv2.imread(image)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
eight_bit = np.uint8(gray)
cm = feature.greycomatrix(eight_bit, [1], [0])
contrast = feature.greycoprops(cm)
dissimilarity = feature.greycoprops(cm, 'dissimilarity')
homogeneity = feature.greycoprops(cm, 'homogeneity')
ASM = feature.greycoprops(cm, 'ASM')
energy = feature.greycoprops(cm, 'energy')
correlation = feature.greycoprops(cm, 'correlation')
ent = measure.shannon_entropy(cm)
return [contrast[0][0], dissimilarity[0][0], homogeneity[0][0], ASM[0][0], energy[0][0], correlation[0][0], ent]
def region_texture_metrics(region, image=None, tag='', glcm=False):
"""Texture analysis for a of scikit-image region"""
database = pd.Series(dtype=object)
# Check to see whether intensity_image is present or image argument
# has been supplied
if image is not None:
region_image = bbox_sample(region, image)
else:
region_image = region.intensity_image
# Obtain indices of pixels in region mask
indices = np.where(region.image)
intensity_sample = region_image[indices]
# _, _, database[f"{tag} Fourier SDI"] = (0, 0, 0)
# fourier_transform_analysis(segment_image)
database[f"{tag} Mean"] = np.mean(intensity_sample)
database[f"{tag} STD"] = np.std(intensity_sample)
database[f"{tag} Entropy"] = shannon_entropy(intensity_sample)
if glcm:
glcm = greycomatrix(
(region_image * region.image * IMAGE_MAX).astype('uint8'), [1, 2],
[0, np.pi / 4, np.pi / 2, np.pi * 3 / 4],
256,
symmetric=True,
normed=True)
glcm[0, :, :, :] = 0
glcm[:, 0, :, :] = 0
greycoprops = greycoprops_edit(glcm)
metrics = [
"Contrast", "Homogeneity", "Energy", "Entropy", "Autocorrelation",
"Clustering", "Mean", "Covariance", "Correlation"
]
for metric in metrics:
value = greycoprops[metric.lower()].mean()
database[f"{tag} GLCM {metric}"] = value
return database
def get_image_entropies(dir_path):
image_entropies = []
counter = 0
for filename in os.listdir(dir_path):
if filename.endswith('.jpg'):
path = dir_path + filename
rgbImg = io.imread(path)
grayImg = img_as_ubyte(color.rgb2gray(rgbImg))
ent = measure.shannon_entropy(grayImg)
dictionary = {
"filename": filename,
"entropy-value": ent,
"counter": counter
}
counter = counter + 1
image_entropies.append(dictionary)
image_entropies.sort(key=itemgetter('entropy-value'), reverse=True)
print(image_entropies)
return image_entropies
def compute_gabor_feats(self, image, kernels):
features = {}
f_var = []
f_mean = []
f_energy = []
f_ent = []
for k, kernel in enumerate(kernels):
filtered = ndi.convolve(image, kernel, mode='wrap')
f_mean.append(np.mean(filtered))
f_var.append(np.var(filtered))
f_energy.append(
np.sum(np.power(filtered.ravel(), 2)) / len(filtered.ravel()))
f_ent.append(shannon_entropy(filtered))
features['Gabor Variance'] = mean(f_var)
features['Gabor Mean'] = mean(f_mean)
features['Gabor Energy'] = mean(f_energy)
features['Gabor Entropy'] = mean(f_ent)
return features
def glcm_feature(image):
glcm = greycomatrix(image, [2, 8, 16],
[0, np.pi / 4, np.pi / 2, np.pi * 3 / 4],
256,
symmetric=True,
normed=True)
#print(len(glcm))
arr = np.empty((0, 0))
for prop in {
'contrast', 'dissimilarity', 'homogeneity', 'correlation', 'ASM'
}: #, 'energy'
temp = greycoprops(glcm, prop)
temp = np.array(temp).reshape(-1)
arr = np.append(arr, temp)
entropy_tem = []
for k in range(glcm.shape[2]):
for j in range(glcm.shape[3]):
entropy_tem.append(measure.shannon_entropy(glcm[:, :, k, j]))
entropy_feat = np.array(entropy_tem)
arr = np.append(arr, entropy_feat)
return arr #.reshape([-1]) #4*3*6=72
def extract_features(array):
props_array = []
angles = [0, np.pi / 4, np.pi / 2, 3 * np.pi / 4]
for img in array:
ubyte_img = img_as_ubyte(img)
img_features = np.array([])
for i in range(3):
glcm = greycomatrix(ubyte_img[:, :, i], [1], angles)
filt_glcm = glcm[:, :, :, :]
constrast = greycoprops(filt_glcm, 'contrast')
energy = greycoprops(filt_glcm, 'energy')
homogeneity = greycoprops(filt_glcm, 'homogeneity')
correlation = greycoprops(filt_glcm, 'correlation')
entropy = shannon_entropy(ubyte_img[:, :, i])
img_features = np.insert(
img_features, 0,
np.concatenate(
(constrast, energy, homogeneity, correlation)).flatten())
img_features = np.insert(img_features, 0, entropy)
props_array.append(img_features)
return props_array
def _calc_features(self):
"""
calculate feature values
:return: feature values
"""
features = {}
props = regionprops(self.bin_img, self.img)
features['Mean Intensity'] = np.mean(self.imgzero)
#mean intensity of image
features['Std'] = np.std(self.imgzero)
#standard deviation
features['Variance'] = np.var(self.imgzero)
#variance
features['Skewness'] = skew(self.imgzero)
#skewness of distribution
features['Kurtosis'] = kurtosis(self.imgzero)
#kurtosis of distribution
features['Contrast'] = np.std(self.hist)
#contrast can be defined as std of histogram of intensity
features['Max Intensity'] = props[0].max_intensity
#Value with the greatest intensity in the region.
features['Min Intensity'] = props[0].min_intensity
#Value with the greatest intensity in the region.
features['Entropy'] = shannon_entropy(self.img, base=2)
#The Shannon entropy is defined as S = -sum(pk * log(pk)), where pk are frequency/probability of pixels of value k.
return features
def test_shannon_ones():
img = np.ones((10, 10))
res = shannon_entropy(img, base=np.e)
assert_almost_equal(res, 0.0)
def test_shannon_all_unique():
img = np.arange(64)
res = shannon_entropy(img, base=2)
assert_almost_equal(res, np.log(64) / np.log(2))
