def show_outline(src):
contours = cv2.Canny(image=src, threshold1=125, threshold2=350)
plt_save(255 - contours, title='Canny Contours')
image_gray = cv2.cvtColor(src=src, code=cv2.COLOR_BGR2GRAY)
contours = cv2.Canny(image=image_gray, threshold1=125, threshold2=350)
plt_save(255 - contours, title='Canny Contours Gray')
# Hough tranform for line detection
theta = np.pi / 180
threshold = 50
lines = cv2.HoughLinesP(image=contours, rho=1, theta=theta, threshold=threshold)
if lines is not None:
src_clone = src.copy()
for line in lines:
for x1, y1, x2, y2 in line:
cv2.line(img=src_clone, pt1=(x1, y1), pt2=(x2, y2), color=(0, 255, 0), thickness=2)
pass
pass
plt_save(src_clone, title='Lines with HoughP, threshold: ' + str(threshold))
pass
# Detect circles
# blur = cv2.GaussianBlur(src=image_gray, ksize=(5, 5), sigmaX=1.5)
threshold = 200
min_votes = 100
circles = cv2.HoughCircles(image=image_gray, method=cv2.HOUGH_GRADIENT, dp=2, minDist=20,
param1=threshold, param2=min_votes, minRadius=15, maxRadius=50)
if circles is not None:
src_clone = src.copy()
for circle in circles:
for x1, y1, r in circle:
cv2.circle(img=src_clone, center=(x1, y1), radius=int(r), color=(0, 255, 0), thickness=2)
pass
plt_save(src_clone, title='Circles with HoughP, threshold: ' + str(threshold) + ', min_votes=' + str(min_votes))
pass
# Get the contours
src_clone = src.copy()
contours, _ = cv2.findContours(image=image_gray, mode=cv2.RETR_LIST, method=cv2.CHAIN_APPROX_NONE)
image_contours = cv2.drawContours(image=src_clone, contours=contours, contourIdx=-1, color=(255, 255, 255),
thickness=2)
plt_save(image_contours, title='Contours with RETR_LIST')
from image_filtering import show_filtering
from image_outline import show_outline
from image_transformation import show_transformation
from utils import plt_save
from image_color import show_hsv
from image_enhancement import show_enhancement
if __name__ == '__main__':
file_path = './images/000000507081.jpg'
origin = cv2.imread(file_path)
origin = origin[:, :, [2, 1, 0]]
# x, y = origin.shape[0:2]
# origin = cv2.resize(origin, (int(y / 3), int(x / 3)))
plt_save(image=origin, title='Origin')
# --------------------图像色彩--------------------
# 转换成HSV色彩空间
show_hsv(origin)
# --------------------图像变换--------------------
show_transformation(origin)
# --------------------图像过滤--------------------
show_filtering(origin)
# --------------------提取直线、轮廓、区域--------------------
show_outline(origin)
# -------------------- 图像增强-白平衡等--------------------
def show_filtering(src):
# 低通滤波
# Blur the image with a mean filter
blur = cv2.blur(src=src, ksize=(5, 5))
plt_save(image=blur, title='Mean filtered (5x5)')
# Blur the image with a mean filter 9x9
blur = cv2.blur(src=src, ksize=(9, 9))
plt_save(image=blur, title='Mean filtered (9x9)')
blur = cv2.GaussianBlur(src=src, ksize=(9, 9), sigmaX=1.5)
plt_save(image=blur, title='Gaussian filtered Image (9x9)')
gauss = cv2.getGaussianKernel(ksize=9, sigma=1.5, ktype=cv2.CV_32F)
print('GaussianKernel 1.5 = [', end='')
for item in gauss:
print(item, end='')
pass
print(']')
gauss = cv2.getGaussianKernel(ksize=9, sigma=-1, ktype=cv2.CV_32F)
print('GaussianKernel -1 = [', end='')
for item in gauss:
print(item, end='')
pass
print(']')
# 缩减 采样
blur = cv2.GaussianBlur(src=src, ksize=(11, 11), sigmaX=1.75)
resized1 = cv2.resize(src=blur,
dsize=(0, 0),
fx=0.25,
fy=0.25,
interpolation=cv2.INTER_CUBIC)
plt_save(image=resized1, title='resize CUBIC 0.25')
# resizing with NN
resized2 = cv2.resize(src=resized1,
dsize=(0, 0),
fx=4,
fy=4,
interpolation=cv2.INTER_NEAREST)
plt_save(image=resized2, title='resize NEAREST x4')
# resizing with bilinear
resized3 = cv2.resize(src=resized1,
dsize=(0, 0),
fx=4,
fy=4,
interpolation=cv2.INTER_LINEAR)
plt_save(image=resized3, title='resize LINEAR x4')
# 中值滤波
median_blur = cv2.medianBlur(src=src, ksize=5)
plt_save(image=median_blur, title='Median filtered')
# 定向滤波器
image_gray = cv2.cvtColor(src=src, code=cv2.COLOR_BGR2GRAY)
# Compute Sobel X derivative
sobel_x = cv2.Sobel(src=image_gray,
ddepth=cv2.CV_8U,
dx=1,
dy=0,
ksize=3,
scale=0.4,
delta=128,
borderType=cv2.BORDER_DEFAULT)
plt_save(image=sobel_x, title='Sobel X')
# Compute Sobel Y derivative
sobel_y = cv2.Sobel(src=image_gray,
ddepth=cv2.CV_8U,
dx=0,
dy=1,
ksize=3,
scale=0.4,
delta=128,
borderType=cv2.BORDER_DEFAULT)
plt_save(image=sobel_y, title='Sobel Y')
# Compute norm of Sobel
sobel_x = cv2.Sobel(src=image_gray, ddepth=cv2.CV_16S, dx=1, dy=0)
sobel_y = cv2.Sobel(src=image_gray, ddepth=cv2.CV_16S, dx=0, dy=1)
sobel_1 = abs(sobel_x) + abs(sobel_y)
plt_save(image=sobel_1, title='abs Sobel X+Y')
sobel_2 = cv2.convertScaleAbs(sobel_x) + cv2.convertScaleAbs(sobel_y)
plt_save(image=sobel_2, title='cv2.convertScaleAbs Sobel X+Y')
# Compute Sobel X derivative (7x7)
sobel_x_7x7 = cv2.Sobel(src=image_gray,
ddepth=cv2.CV_8U,
dx=1,
dy=0,
ksize=7,
scale=0.001,
delta=128)
plt_save(image=sobel_x_7x7, title='Sobel X (7x7)')
uint8_sobel_1 = np.uint8(sobel_1)
plt_save(image=uint8_sobel_1, title='uint8 sobel_1')
uint8_sobel_2 = np.uint8(sobel_2)
plt_save(image=uint8_sobel_2, title='uint8 sobel_2')
int8_sobel_1 = np.int8(sobel_1)
plt_save(image=int8_sobel_1, title='int8 sobel_1')
int8_sobel_2 = np.int8(sobel_2)
plt_save(image=int8_sobel_2, title='int8 sobel_2')
# Apply threshold to Sobel norm (low threshold value)
_, thresh_binary = cv2.threshold(src=uint8_sobel_1,
thresh=255,
maxval=255,
type=cv2.THRESH_BINARY)
plt_save(image=thresh_binary,
title='Binary Sobel (low) cv2.threshold uint8_sobel_1')
_, thresh_binary = cv2.threshold(src=uint8_sobel_2,
thresh=255,
maxval=255,
type=cv2.THRESH_BINARY)
plt_save(image=thresh_binary,
title='Binary Sobel (low) cv2.threshold uint8_sobel_2')
# Apply threshold to Sobel norm (high threshold value)
_, thresh_binary = cv2.threshold(src=uint8_sobel_1,
thresh=190,
maxval=255,
type=cv2.THRESH_BINARY)
plt_save(image=thresh_binary,
title='Binary Sobel Image (high) cv2.threshold uint8_sobel_1')
_, thresh_binary = cv2.threshold(src=uint8_sobel_2,
thresh=190,
maxval=255,
type=cv2.THRESH_BINARY)
plt_save(image=thresh_binary,
title='Binary Sobel Image (high) cv2.threshold uint8_sobel_2')
add_weighted_sobel = cv2.addWeighted(sobel_x, 0.5, sobel_y, 0.5, 0)
plt_save(image=add_weighted_sobel, title='cv2.addWeighted abs')
# down-sample and up-sample the image
reduced = cv2.pyrDown(src=src)
rescaled = cv2.pyrUp(src=reduced)
plt_save(image=rescaled, title='Rescaled')
# down-sample and up-sample the image
reduced = cv2.pyrDown(src=image_gray)
rescaled = cv2.pyrUp(src=reduced)
w, h = src.shape[0:2]
rescaled = cv2.resize(rescaled, (h, w))
plt_save(image=rescaled, title='Rescaled')
subtract = cv2.subtract(src1=rescaled, src2=image_gray)
subtract = np.uint8(subtract)
plt_save(image=subtract, title='cv2.subtract')
gauss05 = cv2.GaussianBlur(src=image_gray, ksize=(0, 0), sigmaX=0.5)
gauss15 = cv2.GaussianBlur(src=image_gray, ksize=(0, 0), sigmaX=1.5)
subtract = cv2.subtract(src1=gauss15, src2=gauss05, dtype=cv2.CV_16S)
subtract = np.uint8(subtract)
plt_save(image=subtract, title='cv2.subtract gauss15 - gauss05')
gauss20 = cv2.GaussianBlur(src=image_gray, ksize=(0, 0), sigmaX=2.0)
gauss22 = cv2.GaussianBlur(src=image_gray, ksize=(0, 0), sigmaX=2.2)
subtract = cv2.subtract(src1=gauss22, src2=gauss20, dtype=cv2.CV_32F)
subtract = np.uint8(subtract)
plt_save(image=subtract, title='cv2.subtract gauss22 - gauss20')
title='Test Set Visualization',
xlabel='Latent dim 1',
ylabel='Latent dim 2')
print('\nCorresponding mapping coordinates in the latent space - one image per digit:\n')
encoded_x_test = encoder.predict(x_test)
digits = {}
i = 0
while len(digits) < 10:
digits[y_test[i]] = encoded_x_test[i]
i += 1
ut.output('g.c', pd.DataFrame(digits))
# (d)
z_sample = np.array([[-2.5, 0.55]])
decoded_x = generator.predict(z_sample)
ut.image(decoded_x)
ut.plt_save('g.d')
# (e)
source, target = digits[3], digits[5]
(x1, y1), (x2, y2) = source, target
a = (y2 - y1) / (x2 - x1)
b = y1 - (a * x1)
f = lambda _x_: a * _x_ + b
x_samples = np.linspace(x1, x2, num=10)
samples = [np.array([[_x, f(_x)]]) for _x in x_samples]
ut.images(map(generator.predict, samples), map(str, x_samples), n_cols=3)
ut.plt_save('g.e')
title='Test Set Visualization',
xlabel='Latent dim 1',
ylabel='Latent dim 2')
print('\nf. Corresponding mapping coordinates in the latent space - one image per digit:\n')
encoded_x_test = encoder.predict(x_test)
digits = {}
i = 0
while len(digits) < 10:
digits[y_test[i]] = encoded_x_test[i]
i += 1
ut.output('f.c', pd.DataFrame(digits))
# (d)
z_sample = np.array([[-2.5, 0.55]])
decoded_x = generator.predict(z_sample)
ut.image(decoded_x)
ut.plt_save('f.d')
# (e)
source, target = digits[3], digits[5]
(x1, y1), (x2, y2) = source, target
a = (y2 - y1) / (x2 - x1)
b = y1 - (a * x1)
f = lambda _x_: a * _x_ + b
x_samples = np.linspace(x1, x2, num=10)
samples = [np.array([[_x, f(_x)]]) for _x in x_samples]
ut.images(map(generator.predict, samples), map(str, x_samples), n_cols=3)
ut.plt_save('f.e')
def show_transformation(src):
# 用形态学滤波器腐蚀和膨胀图像
element_3x3 = np.ones((3, 3), np.uint8)
# 腐蚀 3x3
eroded = cv2.erode(src=src, kernel=element_3x3)
plt_save(image=eroded, title='eroded')
# 膨胀 3x3 3次
dilated = cv2.dilate(src=src, kernel=element_3x3, iterations=3)
plt_save(image=dilated, title='dilated 3 times')
# 腐蚀 7x7
element_7x7 = np.ones((7, 7), np.uint8)
eroded_7x7 = cv2.erode(src=src, kernel=element_7x7, iterations=1)
plt_save(image=eroded_7x7, title='eroded 7x7')
# 腐蚀 3x3 3次
eroded_3 = cv2.erode(src=src, kernel=element_3x3, iterations=3)
plt_save(image=eroded_3, title='eroded 3 times')
# 用形态学滤波器开启和闭合图像
image_gray = cv2.cvtColor(src=src, code=cv2.COLOR_RGB2GRAY)
# plt_save(image=image_gray, title='image_gray')
# Close the image
element_5x5 = np.ones((5, 5), np.uint8)
closed = cv2.morphologyEx(src=image_gray,
op=cv2.MORPH_CLOSE,
kernel=element_5x5)
plt_save(image=closed, title='closed')
# Open the image
opened = cv2.morphologyEx(src=image_gray,
op=cv2.MORPH_OPEN,
kernel=element_5x5)
plt_save(image=opened, title='opened')
closed = cv2.morphologyEx(src=image_gray,
op=cv2.MORPH_CLOSE,
kernel=element_5x5)
closed_opened = cv2.morphologyEx(src=closed,
op=cv2.MORPH_OPEN,
kernel=element_5x5)
plt_save(image=closed_opened, title='Closed -> Opened')
opened = cv2.morphologyEx(src=image_gray,
op=cv2.MORPH_OPEN,
kernel=element_5x5)
opened_closed = cv2.morphologyEx(src=opened,
op=cv2.MORPH_CLOSE,
kernel=element_5x5)
plt_save(image=opened_closed, title='Opened -> Closed')
# 在灰度图像中应用形态学运算
edge = cv2.morphologyEx(src=image_gray,
op=cv2.MORPH_GRADIENT,
kernel=element_3x3)
plt_save(image=255 - edge, title='Gradient | Edge')
# Apply threshold to obtain a binary image
threshold = 80
_, thresh_binary = cv2.threshold(src=edge,
thresh=threshold,
maxval=255,
type=cv2.THRESH_BINARY)
plt_save(image=thresh_binary,
title='Gradient | Edge -> Thresh Binary | Edge')
# 7x7 Black Top-hat Image
black_hat = cv2.morphologyEx(src=image_gray,
op=cv2.MORPH_BLACKHAT,
kernel=element_7x7)
plt_save(image=255 - black_hat, title='7x7 Black Top-hat')
# Apply threshold to obtain a binary image
threshold = 25
_, thresh_binary = cv2.threshold(src=black_hat,
thresh=threshold,
maxval=255,
type=cv2.THRESH_BINARY)
plt_save(image=255 - thresh_binary,
title='7x7 Black Top-hat -> Thresh Binary | Edge')
# Apply the black top-hat transform using a 7x7 structuring element
closed = cv2.morphologyEx(src=thresh_binary,
op=cv2.MORPH_CLOSE,
kernel=element_7x7)
plt_save(image=255 - closed, title='7x7 Black Top-hat -> Closed')
pass
ylabel='Latent dim 2')
print(
'\nCorresponding mapping coordinates in the latent space - one image per digit:\n'
)
encoded_x_test = encoder.predict(x_test)
digits = {}
i = 0
while len(digits) < 10:
digits[y_test[i]] = encoded_x_test[i]
i += 1
ut.output('c', pd.DataFrame(digits))
# (d)
z_sample = np.array([[-2.5, 0.55]])
decoded_x = generator.predict(z_sample)
ut.image(decoded_x)
ut.plt_save('d')
# (e)
source, target = digits[3], digits[5]
(x1, y1), (x2, y2) = source, target
a = (y2 - y1) / (x2 - x1)
b = y1 - (a * x1)
f = lambda _x_: a * _x_ + b
x_samples = np.linspace(x1, x2, num=10)
samples = [np.array([[_x, f(_x)]]) for _x in x_samples]
ut.images(map(generator.predict, samples), map(str, x_samples), n_cols=3)
ut.plt_save('e')
def show_enhancement(src):
balance_img1 = white_balance(src)
plt_save(balance_img1, 'White Balance 1')
# 灰度世界算法
balance_img2 = grey_world(src)
plt_save(balance_img2, 'White Balance 2')
# 直方图均衡化
balance_img3 = his_equl_color(src)
plt_save(balance_img3, 'White Balance 3')
# 视网膜-大脑皮层(Retinex)增强算法
# 单尺度Retinex
ssr1 = single_scale_retinex(src, 300)
ssr1[ssr1 < 0] = 0
plt_save(ssr1, 'Single Scale Retinex 1')
ssr2 = s_s_r(src)
plt_save(ssr2, 'Single Scale Retinex 2')
# 多尺度Retinex
msr1 = multi_scale_retinex(src, [15, 80, 250])
msr1[msr1 < 0] = 0
plt_save(msr1, 'Multi Scale Retinex 1')
msr2 = m_s_r(src, sigma_list=[15, 80, 250])
plt_save(msr2, 'Multi Scale Retinex 2')
msrcr1 = m_s_r_c_r(src, sigma_list=[15, 80, 250])
plt_save(msrcr1, 'Multi Scale Retinex With Color Restoration 1')
# 自动白平衡 AWB
awb = automatic_white_balance(src)
plt_save(awb, 'Automatic White Balance')
# 自动色彩均衡 ACE
balance_img4 = automatic_color_equalization(src)
plt_save(balance_img4, 'Automatic Color Equalization')