import cv2 as cv
import numpy as np
from matplotlib import pyplot as plt
from statistics import stdev
from pathlib import Path
import datetime
class Rectangle:
def __init__(self, x1, y1, w1, h1):
self.x = x1
self.y = y1
self.w = w1
self.h = h1
self.is_surface_breaking=True
self.is_acceptable=True
def __repr__(self):
return "("+str(self.x)+","+str(self.y)+"),"+str(self.w)+"x"+str(self.h)\
+ ", Surface: " + str(self.is_surface_breaking)+", Acceptance: "+str(self.is_acceptable)
def nothing(x):
print("Nothing:"+str(x))
def is_inside_joint(inner: Rectangle, outer: Rectangle) -> bool:
if(inner.x >= outer.x and inner.y >= outer.y and
(inner.x+inner.w) <= (outer.x+outer.w) and (inner.y+inner.h) <= (outer.y+outer.h)):
return True
else:
return False
def error_return():
return 0, 0, 0, 0, 0, 0, 0
def find_smallest_rec(all_rec):
smallest = all_rec[0]
for rec in all_rec:
if rec.w*rec.h < smallest.w*smallest.h:
smallest = rec
return smallest
def discard_irrelevant_results(all_recs):
rec_fields = []
rec_relevant = []
sum_fields = 0
for rec in all_recs:
sum_fields += rec.w*rec.h
average_field = sum_fields/len(all_recs)
# for rec in all_recs:
# if rec.w * rec.h > average_field:
# rec_fields.append(rec.w*rec.h)
# rec_relevant.append(Rectangle(rec.x,rec.y,rec.w,rec.h))
for rec in all_recs:
rec_fields.append(rec.w * rec.h)
all_recs.remove(find_smallest_rec(all_recs))
deviation = stdev(rec_fields)
while deviation/average_field >= 0.85 and len(all_recs) > 1:
# print("dev/mean: "+str(deviation/average_field)+" dev: "+str(deviation)+" mean: "+str(average_field))
all_recs.remove(find_smallest_rec(all_recs))
for rec in all_recs:
sum_fields += rec.w*rec.h
rec_fields.append(rec.w*rec.h)
average_field = sum_fields / len(all_recs)
deviation = stdev(rec_fields)
# print("dev/mean: " + str(deviation / average_field) + " dev: " + str(deviation) + " mean: " + str(average_field))
# plt.hist(rec_fields, rwidth=5)
# plt.title("Average field: "+str(average_field)+"\nIndications: "+str(len(rec_fields)))
# plt.show()
return all_recs
# def check_histograms(img, rec, roi):
# k=45
# #pobieram wymiary obrazu
# ref_up = cv.imread("images\\ref_up.png")
# ref_down = cv.imread("images\\ref_down.png")
# #przycinam obrazy do mierzenia histogramów
# indication_surface_down = img[roi.y+roi.h:roi.y+roi.h+k, rec.x:rec.x+rec.w]
# indication_surface_up = img[roi.y-k:roi.y, rec.x:rec.x+rec.w] #górna część obrazu oobcięta SZTYWNO
# #liczę histogramy
# hist_indication_surface_down = cv.calcHist([indication_surface_down], [0], None, [256], [0,256])
# hist_indication_surface_up = cv.calcHist([indication_surface_up], [0], None, [256], [0,256])
# hist_surface_down = cv.calcHist([ref_down], [0], None, [256], [0,256])
# hist_surface_up = cv.calcHist([ref_up], [0], None, [256], [0,256])
# #porównuję histogramy
# check_up = cv.compareHist(hist_surface_up, hist_indication_surface_up, cv.HISTCMP_CORREL)
# check_down = cv.compareHist(hist_surface_down, hist_indication_surface_down, cv.HISTCMP_CORREL)
# #zwracam poprawione prostokąty zależnie od histogramu
# # if check_down<check_up and check_down<0.92:
# # corrected_rect = [(rec.x,rec.y),(rec.x+rec.w,roi.y+roi.h+k)]
# # elif check_up<check_down and check_up<0.92:
# # corrected_rect = [(rec.x,roi.y-k),(rec.x+rec.w,rec.y+rec.h)]
# # else:
# # corrected_rect = [(rec.x,rec.y),(rec.x+rec.w,rec.y+rec.h)]
# if check_down<check_up and check_down<0.15:
# corrected_rect = Rectangle(rec.x, rec.y, rec.w, roi.y+roi.h+k-rec.y)
# corrected_rect.is_surface_breaking=True
# elif check_up<check_down and check_up<=0.15:
# corrected_rect = Rectangle(rec.x, roi.y-k, rec.w, rec.y-roi.y+rec.h+k)
# corrected_rect.is_surface_breaking=True
# # [(rec.x,roi.y-k),(rec.x+rec.w,rec.y+rec.h)]
# else:
# corrected_rect = Rectangle(rec.x,rec.y,rec.w,rec.h)
# print("Input rec: "+str(rec))
# print("Check down: " + str(check_down) + "\nCheck up: " + str(check_up))
# print(corrected_rect)
#
# return corrected_rect
def check_histograms(img, rec, roi):
k = 20 # wysokość fali powierzchniowej
# pobieram wymiary obrazu
height, width, channels = img.shape
# przycinam obrazy do mierzenia histogramów
indication_surface_down = img[roi.y+roi.h:roi.y+roi.h+k, rec.x:rec.x+rec.w]
indication_surface_up = img[roi.y-k:roi.y, rec.x:rec.x+rec.w]
surface_down = img[roi.y+roi.h:roi.y+roi.h+k, roi.x:width]
surface_up = img[roi.y-k:roi.y, roi.x:width]
# liczę histogramy
hist_indication_surface_down = cv.calcHist([indication_surface_down], [0], None, [256], [0,256])
hist_indication_surface_up = cv.calcHist([indication_surface_up], [0], None, [256], [0,256])
hist_surface_down = cv.calcHist([surface_down], [0], None, [256], [0,256])
hist_surface_up = cv.calcHist([surface_up], [0], None, [256], [0,256])
# porównuję histogramy
check_up = cv.compareHist(hist_surface_up, hist_indication_surface_up, cv.HISTCMP_CORREL)
check_down = cv.compareHist(hist_surface_down, hist_indication_surface_down, cv.HISTCMP_CORREL)
# zwracam poprawione prostokąty zależnie od histogramu
# if check_down<check_up and check_down<0.92:
# corrected_rect = [(rec.x,rec.y),(rec.x+rec.w,roi.y+roi.h+k)]
# elif check_up<check_down and check_up<0.92:
# corrected_rect = [(rec.x,roi.y-k),(rec.x+rec.w,rec.y+rec.h)]
# else:
# corrected_rect = [(rec.x,rec.y),(rec.x+rec.w,rec.y+rec.h)]
if check_down<check_up and check_down < 0.9:
corrected_rect = Rectangle(rec.x, rec.y, rec.w, roi.y+roi.h+k-rec.y)
corrected_rect.is_surface_breaking=True
elif check_up<check_down and check_up <= 0.9:
corrected_rect = Rectangle(rec.x, roi.y-k, rec.w, rec.y-roi.y+rec.h+k)
corrected_rect.is_surface_breaking = True
# [(rec.x,roi.y-k),(rec.x+rec.w,rec.y+rec.h)]
else:
corrected_rect = Rectangle(rec.x,rec.y,rec.w,rec.h)
print("Input rec: "+str(rec))
print("Check down: " + str(check_down) + "\nCheck up: " + str(check_up))
print(corrected_rect)
return corrected_rect
def display_process(images):
# titles = ['ORIGINAL', 'FILTER', 'CANNY', 'DILATE', 'CONTOURS']
# for i in range(5):
# plt.subplot(3, 2, i+1)
# plt.imshow(images[i], 'gray')
# plt.title(titles[i])
# plt.xticks([])
# plt.yticks([])
# plt.show()
plt.subplot(1, 2, 1)
plt.imshow(images[1], 'gray')
plt.title('FILTER')
plt.xticks([])
plt.yticks([])
plt.subplot(1,2,2)
plt.imshow(images[2], 'gray')
plt.title('EDGES')
plt.xticks([])
plt.yticks([])
plt.show()
def find_roi(filename):
thresh1 = 100
thresh2 = 255
dil_kernel_size = 7
dil_it = 3
filter_kernel = 5
roi_img = cv.imread(filename)
roi_imgray = cv.cvtColor(roi_img, cv.COLOR_BGR2GRAY)
roi_imgray=cv.GaussianBlur(roi_imgray,(filter_kernel,filter_kernel),0)
_, roi_imthresh = cv.threshold(roi_imgray, thresh1, thresh2, cv.THRESH_BINARY_INV)
roi_kernel = np.ones((dil_kernel_size, dil_kernel_size), np.uint8)
roi_imdil = cv.dilate(roi_imthresh, roi_kernel, iterations=dil_it)
h, w, c = roi_img.shape
roi_mat_img = np.transpose(roi_imdil)
first_column = roi_mat_img[0]
for i in range(len(first_column)):
if i == len(first_column)-1:
y_min = 0
y_max = len(first_column)-1
break
if first_column[i] == 255 and first_column[i + 1] == 0 and i > 10:
y_min = i
break
for j in range(i, len(first_column) - 1):
if first_column[j] == 0 and first_column[j + 1] == 255 and j - y_min > 50:
y_max = j
break
roi = Rectangle(0, y_min, w, y_max-y_min)
return roi
def convert_to_mm(roi, thickness, all_recs):
px_to_mm = int(thickness)/(roi.h+60)
all_recs_in_mm = []
for rec in all_recs:
rec_in_mm = Rectangle(round(rec.x*px_to_mm, 1), round(rec.y*px_to_mm, 1),
round(rec.w*px_to_mm, 1), round(rec.h*px_to_mm, 1))
all_recs_in_mm.append(rec_in_mm)
x_max = 0; y_max = 0; w_max = 0; h_max = 0
for rec in all_recs_in_mm:
if rec.w * rec.h > w_max * h_max:
x_max = rec.x
y_max = rec.y
w_max = rec.w
h_max = rec.h
return x_max, y_max, w_max, h_max, all_recs_in_mm
def confirm_indications(all_indications, acceptance_level, thickness):
joint_acceptance = True
if acceptance_level == 1:
l_max = 0.75*float(thickness)
h3 = 1.5
h2 = 2
h1 = 1
elif acceptance_level == 2:
l_max = thickness
h3 = 2
h2 = 2
h1 = 1
elif acceptance_level == 3:
l_max = min(1.5*thickness, 20)
h3 = 2
h2 = 2
h1 = 1
for ind in all_indications:
if ind.w > l_max and ind.h > h1:
ind.is_acceptable = False
joint_acceptance = False
elif ind.w <= l_max:
if ind.is_surface_breaking and ind.h > h3:
ind.is_acceptable = False
joint_acceptance = False
elif not ind.is_surface_breaking and ind.h > h2:
ind.is_acceptable = False
joint_acceptance = False
elif ind.h < h2 or ind.h < h3:
if ind.w > l_max:
ind.is_acceptable = False
joint_acceptance = False
for ind in all_indications:
print(ind)
date = datetime.date.today()
time = datetime.datetime.now().time()
with open(str(date) + "_" + str(time).replace(':', '_')[:-8] + "_results.txt", "w") as f:
i = 0
for rec in all_indications:
f.write("Pattern " + str(i) + ": x0 = " + str(rec.x) + "; y0 = " + str(rec.y) + "; w = "
+ str(rec.w) + "; h = " + str(rec.h) + "; " + str(rec.is_acceptable) + "\n")
i += 1
return joint_acceptance
def inspect(filepath, list_of_parameters, acceptance_level, thickness):
try:
filter_size=list_of_parameters[0]
canny_min=list_of_parameters[1]
canny_max=list_of_parameters[2]
dilation_kernel_size=list_of_parameters[3]
dilation_it=list_of_parameters[4]
roi_x_min=list_of_parameters[5]
roi_x_max=list_of_parameters[6]
roi_y_min=list_of_parameters[7]
roi_y_max=list_of_parameters[8]
except IndexError:
filter_size = 9
canny_min = 20
canny_max = 77
dilation_kernel_size = 3
dilation_it = 3
roi_x_min = -1
roi_x_max = -1
roi_y_min = -1
roi_y_max = -1
try_file = cv.imread(filepath)
try:
try_file.size == 0
except AttributeError:
return error_return()
if roi_x_min == -1 and roi_x_max == -1 and roi_y_min == -1 and roi_y_max == -1:
roi = find_roi(filepath)
else:
roi = Rectangle(roi_x_min, roi_y_min, roi_x_max-roi_x_min, roi_y_max-roi_y_min)
img = cv.imread(filepath)
height, width, channels = img.shape
imgray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
filter_img = cv.medianBlur(imgray,filter_size)
# filter_img = cv.GaussianBlur(imgray,(filter_size,filter_size),0)
# filter_img = cv.bilateralFilter(imgray, 10, 10, 10)
edges = cv.Canny(filter_img, canny_min, canny_max)
cv.imwrite("edges.png", edges)
try:
img_to_dilate = edges[roi.y:roi.y+roi.h, roi.x:roi.x+roi.w]
except IndexError:
return error_return()
kernel = np.ones((dilation_kernel_size, dilation_kernel_size), np.uint8)
img_to_dilate = cv.dilate(img_to_dilate, kernel, iterations=dilation_it)
edges[roi.y:roi.y+roi.h,roi.x:roi.x+roi.w] = cv.morphologyEx(img_to_dilate, cv.MORPH_CLOSE, kernel)
dilation = cv.morphologyEx(edges, cv.MORPH_ERODE, kernel)
contours, hierarchy = cv.findContours(dilation, cv.RETR_TREE, cv.CHAIN_APPROX_NONE)
number_of_recs=0
all_recs = []
for i in range(len(contours)):
x, y, w, h = cv.boundingRect(contours[i])
rec = Rectangle(x, y, w, h)
if is_inside_joint(rec, roi):
cv.drawContours(img, contours, i, (0, 0, 255), 1)
all_recs.append(rec)
number_of_recs += 1
if len(all_recs) == 0:
return error_return()
cv.imwrite("contours.png",img)
filtered_recs = discard_irrelevant_results(all_recs)
corrected_recs = filtered_recs
if len(corrected_recs) == 0:
return error_return()
# corrected_recs = []
# for rec in filtered_recs:
# corrected_recs.append(check_histograms(img, rec, roi))
x_max = 0; y_max = 0; w_max = 0; h_max = 0
for rec in corrected_recs:
cv.rectangle(img, (rec.x, rec.y), (rec.x + rec.w, rec.y + rec.h), (0, 255, 0), 1)
if rec.w * rec.h > w_max * h_max:
x_max = rec.x
y_max = rec.y
w_max = rec.w
h_max = rec.h
images = [imgray, filter_img, edges, dilation, img]
filepath_inspected = filepath.replace(".png", "_inspected.png")
cv.imwrite(filepath_inspected, img)
x_max, y_max, w_max, h_max, all_recs_in_mm = convert_to_mm(roi, thickness, corrected_recs)
acceptance = confirm_indications(all_recs_in_mm, acceptance_level, thickness)
with open(Path(filepath).stem+"_results_in_pixels.txt", "w") as f:
i = 0
for rec in all_recs:
f.write("Pattern " + str(i) + ": x0 = " + str(rec.x) + "; y0 = " + str(rec.y) + "; w = " +
str(rec.w) + "; h = " + str(rec.h) + "\n")
i += 1
return acceptance, x_max, y_max, w_max, h_max, filepath_inspected, images
#throwMe = [9, 20, 77, 3, 3, -1, -1, -1, -1]
#acc, x_max, y_max, w_max, h_max, filepath_inspected, images = inspect("images\\test1.png", throwMe, 3, 12)
#cv.imshow("result", images[0])
#cv.waitKey()
#cv.destroyAllWindows()
# pierwsza strona do podpisu, reszta mailem, prezka w grudniu mailem i przyjść przegadać
#
