def match_2(parentdir, targetimg, templateimg, _type='demo', denoise=True):
"""
:param parentdir: the directory to save the image
:param targetimg: target -> bg image
:param templateimg: template -> block
:param _type: demo or login or register
:param denoise: True or False
:return: distance to the left of the north
"""
target = cv2.imread(targetimg, 0)
template = cv2.imread(templateimg, 0)
# image_arr_rgb.shape
# (H, W, C)
w, h = template.shape[::-1]
temp = parentdir + 'temp_gray_' + _type + '.png'
targ = parentdir + 'targ_gray_' + _type + '.jpg'
cv2.imwrite(temp, template)
cv2.imwrite(targ, target)
template = cv2.imread(temp)
target = cv2.imread(targ)
if denoise:
import utils
noisename = parentdir + 'temp_denoise_' + _type + '.png'
utils.denoise(temp, 15, noisename)
# noisename = cv1.mean_blur(noisename)
template = cv2.imread(noisename)
result = cv2.matchTemplate(target, template, cv2.TM_CCOEFF_NORMED)
# result.shape形状的矩阵，第result.argmax()个元素的展开的坐标，第一个参数是一种整数数组，其元素是维度数组dims的扁平版本的索引。
# 例子：
# Consider a (6,7,8) shape array, what is the index (x,y,z) of the 100th element ?
# >>> print np.unravel_index(100,(6,7,8))
# (1, 5, 4)
# 解释：
# 给定一个矩阵，shape=（6,7,8），即3维的矩阵，求第n个元素的下标是什么？矩阵各维的下标从0开始
# 如果indices参数是一个标量，那么返回的是一个向量，维数=矩阵的维数，向量的值其实就是在矩阵中对应的下标。如6*7*8*9的矩阵，
# 1621/(7*8*9)=3，(1621-3*7*8*9)/(8*9)=1，(1621-3*7*8*9-1*8*9)/9=4，(1621-3*7*8*9-1*8*9-4*9)=1。
# 所以返回的向量为array(3,1,4,1)
# 如果indices参数是一个向量的，那么通过该向量中值求出对应的下标。下标的个数就是矩阵的维数，每一维下标组成一个向量，
# 所以返回的向量的个数=矩阵维数。如7*6的矩阵，第22个元素是 3*6+4，所以对应的下标是(3,4)，那么返回的值是 array([3]),array([4]
# ---------------------
# 原文：https://blog.csdn.net/dn_mug/article/details/70256109
x, y = np.unravel_index(result.argmax(), result.shape) # shape:(H,W)
# 展示圈出来的区域
cv2.rectangle(target, (y, x), (y + w, x + h), (7, 249, 151), 2)
cv2.imshow('Show', target)
cv2.waitKey(0)
cv2.destroyAllWindows()
return y
def parse_document(struct_table, img):
denoise_img = denoise(img)
invoice_units = parse_invoice_units(struct_table, denoise_img)
# total_n = len(y_coords)-1
# total_amount_without_tax = get_cell_digits(5,total_n, table_coords, denoise_img)
# total_amount_tax = get_cell_digits(8,total_n, table_coords, denoise_img)
# total_amount_with_tax = get_cell_digits(9,total_n, table_coords, denoise_img)
# print(total_amount_without_tax,total_amount_tax,total_amount_with_tax)
x_table = min([x['cell'][0] for x in struct_table])
y_table = min([x['cell'][1] for x in struct_table])
invoice_data = None
if y_table > 100:
invoice_header_cell = denoise_img[0:y_table,
x_table:denoise_img.shape[1] // 2]
x, y, w, h = invoice_image_rect(invoice_header_cell)
invoice_header_cell = denoise_img[y:y + h, x:x + w]
if DEBUG:
cv2_imshow(invoice_header_cell)
try:
invoice_data = get_invoice_data(invoice_header_cell)
except Exception as E:
logger.info('Не определен заголовок фактуры')
logger.exception(E)
else:
logger.info('Не определен заголовок фактуры')
return invoice_data, invoice_units
def match_4_for_crawler(parentdir, targetimg, templateimg, _type, i, denoise=True):
template = cv2.imread(templateimg, 0)
target = cv2.imread(targetimg, 0)
# 获取图像的长和宽
w, h = template.shape[::-1]
temp = parentdir + 'temp_gray_' + _type + '.png'
targ = parentdir + 'targ_gray_' + _type + '.jpg'
cv2.imwrite(temp, template)
cv2.imwrite(targ, target)
# template = abs(255 - template)
# cv2.imwrite(temp, template)
template = cv2.imread(temp)
target = cv2.imread(targ)
if denoise:
import utils
noisename = parentdir + 'temp_denoise_' + _type + '.png'
# 15不错
utils.denoise(temp, 13, noisename)
# noisename = cv1.mean_blur(noisename)
template = cv2.imread(noisename)
# 模板匹配
result = cv2.matchTemplate(target, template, cv2.TM_CCOEFF_NORMED)
x, y = np.unravel_index(result.argmax(), result.shape)
t = cv2.imread(targetimg, 1)
# 展示圈出来的区域
cv2.rectangle(t, (y, x), (y + w, x + h), (0, 0, 255), 2)
font = cv2.FONT_HERSHEY_SIMPLEX # 定义字体
cv2.putText(t, 'notch 1.000', (y, x - 2), font, 0.4, (255, 0, 0), 1)
# 图像，文字内容， 坐标 ，字体，大小，颜色，字体厚度
# cv2.imshow('Show', t)
cv2.imwrite(parentdir + 'mark/target' + i + _type + '.jpg', t)
cv2.waitKey(0)
cv2.destroyAllWindows()
return y
def wrapper(func, frame_no, partition_number, coords, img_frame, ret_list):
img_frame = utils.denoise(img_frame)
map_ret = func(coords, img_frame)
obj = {
'fno': frame_no,
'pno': partition_number,
'coords': coords,
'return': map_ret,
# 'd_frame': img_frame
}
print(map_ret)
ret_list.append(obj)
if ret_value == False:
break
# frame = imutils.resize(frame, width=700)
clone = frame[:, :]
# get the height and width of the frame
(height, width) = frame.shape[:2]
# convert the roi to grayscale and blur it
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (7, 7), 0)
thresh = cv2.erode(gray, None, iterations=2)
thresh = cv2.dilate(thresh, None, iterations=4)
# calibrating our running average until a threshold is reached
denoised_clone = utils.denoise(clone)
start_time = time()
if num_frames < 6:
running_avg(thresh, avg_wt)
else:
seg_img = segment(thresh)
if seg_img is not None:
# i.e. if segmented
(thresholded, segmented) = seg_img
# print("Center", center)
if num_frames < 50:
selected_partitions = utils.segment_divider(thresholded)
from PIL import Image
from pylab import *
import utils
im = array(Image.open('wisma-nusantara.jpg').convert('L'))
U, T = utils.denoise(im, im)
figure()
gray()
imshow(U)
axis('equal')
axis('off')
show()
sr_gen = generator()
sr_gen.load_weights('weight/srgan/gan_generator.h5')
# Load model 2 -- Illumination improvement
mbllen_gen = Network.build_mbllen((32, 32, 3))
mbllen_gen.load_weights('weight/mbllen/LOL_img_lowlight.h5')
# Load test image
img = readImage(r'C:\Users\ywqqq\Documents\PRS_prj\maskdetection\test.jpg')
# If noise in image, denoise.
gaussian_noise = False
salt_and_pepper_noise = False
if gaussian_noise:
img = denoise(img, 'gaussian')
if salt_and_pepper_noise:
img = denoise(img, 'salt-and-pepper')
# Get the luminance of the image.
# If luminance < 70, then apply illumination improvemtn
imgHSV = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
H, S, V = cv2.split(imgHSV)
if V < 70:
img = resolve_single(mbllen_gen, 'mbllen', img)
# Get the resolution of the image.
# If the resolution < 10000, then apply super resolution
r = img.shape[0] * img.shape[1]
data = data.reshape(-1, 784)
return data
def load_label(file_name):
file_path = dataset_dir + '/' + file_name
with gzip.open(file_path, 'rb') as f:
labels = np.frombuffer(f.read(), np.uint8, offset=8)
return labels
dataset = {}
dataset['train_img'] = load_img(key_file['train_img'])
dataset['train_label'] = load_label(key_file['train_label'])
dataset['test_img'] = load_img(key_file['test_img'])
dataset['test_label'] = load_label(key_file['test_label'])
import scipy.misc
import scipy as sp
for i in range(len(dataset['test_img'])):
sp.misc.imsave('./datasets/mnist_test_image/test' + str(i) + '.jpg',
dataset['test_img'][i].reshape(28, 28))
testdatasets = glob.glob("./datasets/mnist_test_image/*")
for i in testdatasets:
denoise(i)
def match_3(parentdir, targetimg, templateimg, _type="demo", denoise=True):
"""
:param parentdir:
:param _type: demo or login or regist
:param denoise: True or False
:param targetimg: path of the bg image
:param templateimg: path of the slide block
:return: distance
"""
# 参考https://blog.csdn.net/weixin_42081389/article/details/87935735
# a=np.array([[2,3,4,5],[5,6,78,9],[1,3,4,5]])
# print(a)
# # 函数功能：假设有一个矩阵a,现在需要求这个矩阵的最小值，最大值，并得到最大值，最小值的索引，注意是先x后y，先列之间距离，再横向行距。
# min_val,max_val,min_indx,max_indx=cv2.minMaxLoc(a)
# print(min_val,max_val,min_indx,max_indx)
img = cv2.imread(targetimg, 0)
# img2 = img.copy()
template = cv2.imread(templateimg, 0)
temp = parentdir + 'temp_gray_' + _type + '.png'
targ = parentdir + 'targ_gray_' + _type + '.jpg'
cv2.imwrite(targ, img)
# cv2.imshow('messi', img)
# cv2.imshow('face', template)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# 也可以写作： h, w = template.shape[:2]，因为shape是(H,W,C)
w, h = template.shape[::-1]
# All the 6 methods for comparison in a list
# methods = ['cv2.TM_CCOEFF', 'cv2.TM_CCOEFF_NORMED', 'cv2.TM_CCORR',
# 'cv2.TM_CCORR_NORMED', 'cv2.TM_SQDIFF', 'cv2.TM_SQDIFF_NORMED']
# 这个要不要翻转颜色很关键，试试就知道
# template = abs(255 - template)
meth = 'cv2.TM_CCOEFF_NORMED'
# img = img2.copy()
'''
exec可以用来执行储存在字符串文件中的python语句
例如可以在运行时生成一个包含python代码的字符串
然后使用exec语句执行这些语句
eval语句用来计算存储在字符串中的有效python表达式
'''
cv2.imwrite(temp, template)
target = cv2.imread(targ)
template = cv2.imread(temp)
method = eval(meth)
if denoise:
import utils
noisename = parentdir + 'temp_denoise_' + _type + '.png'
utils.denoise(temp, 15, noisename)
# noisename = cv1.mean_blur(noisename)
template = cv2.imread(noisename)
# 匹配应用
res = cv2.matchTemplate(target, template, method)
# print(res)
# print(np.array(res))
# print("res.shape: ", res.shape)
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)
# 使用不同的方法，对结果的解释不同
# 方法判断
# if method in [cv2.TM_SQDIFF, cv2.TM_SQDIFF_NORMED]:
# top_left = min_loc
# else:
# top_left = max_loc
# 最大值的坐标就是距离左边界和上边界的距离
top_left = max_loc
# print('偏移像素', top_left[0])
bottom_right = (top_left[0] + w, top_left[1] + h)
cv2.rectangle(img, top_left, bottom_right, 255, 2)
cv2.imshow('Show', img)
plt.subplot(121), plt.imshow(res, cmap='gray')
plt.title('Matching Result'), plt.xticks([]), plt.yticks([])
plt.subplot(122), plt.imshow(img, cmap='gray')
plt.title('Detected Point'), plt.xticks([]), plt.yticks([])
plt.suptitle('method: ' + meth)
plt.savefig(parentdir + 'mark3/target_res.jpg')
plt.show()
# cv2.imwrite(parentdir + 'mark3/target' + i + _type + '.jpg', img)
cv2.imwrite(parentdir + 'mark3/target' + _type + '.jpg', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
return top_left[0]
