def binarization(image_name, plot=False):
# remove the noise
img = cv.GaussianBlur(cv.imread(image_name, 0), (5, 5), 0)
# compute histogram
values = img.flatten()
hist, bins = np.histogram(values, range(257))
# normalize to get probabilities of each intensity level
norm_hist = hist / np.sum(hist)
# p1, p2 - probabilities of the two classes separated by a threshold t
# m1, m2 - means of these two classes
# v - inter-class variance
thresh = -1
max_v = -1
for t in range(256):
p1 = np.sum(norm_hist[:t])
p2 = np.sum(norm_hist[t:])
m1 = np.sum([i
for i in range(t)] * hist[:t]) / p1 if p1 != 0 else np.nan
m2 = np.sum([i for i in range(t, 256)] *
hist[t:]) / p2 if p2 != 0 else np.nan
v = p1 * p2 * (m1 - m2)**2
if v > max_v:
thresh = t
max_v = v
if plot:
plot_comparison(img, thresh, "Own Otsu's thresholding", otsu=True)
res = np.where(img > thresh, 255, 0)
return thresh, res
def otsu_thresholding_filtered(image_name, plot=False):
# Additionally image is filtered with a 5x5 gaussian kernel to remove the noise
img = cv.imread(image_name, 0)
blur = cv.GaussianBlur(img, (5, 5), 0)
image, thresh = cv.threshold(blur, 0, 255, cv.THRESH_BINARY + cv.THRESH_OTSU)
if plot:
plot_comparison(img, thresh, "Otsu's thresholding filtered")
return thresh
def otsu_thresholding(image_name, plot=False):
# Automatically calculates a threshold value from image histogram
# (for bimodal images, otherwise binarization is not accurate)
img = cv.imread(image_name, 0)
image, thresh = cv.threshold(img, 0, 255, cv.THRESH_BINARY + cv.THRESH_OTSU)
if plot:
plot_comparison(img, thresh, "Otsu's thresholding")
return thresh
def adaptive_gaussian_thresholding(image_name, block_size, plot=False):
# The threshold value is a gaussian-weighted sum of the neighbourhood values minus the constant C
img = cv.imread(image_name, 0)
# blockSize determines the size of the neighbourhood area
thresh = cv.adaptiveThreshold(img, 255, cv.ADAPTIVE_THRESH_GAUSSIAN_C, cv.THRESH_BINARY, blockSize=block_size, C=5)
if plot:
plot_comparison(img, thresh, "Adaptive gaussian thresholding")
return thresh
def global_thresholding(image_name, plot=False):
img = cv.imread(image_name, 0)
image, thresh = cv.threshold(img, 127, 255, cv.THRESH_BINARY)
if plot:
plot_comparison(img, thresh, "Global thresholding")
return thresh