[PYTHON] G test measures OCR convolution calculation tool
Overview
This is the first post. I challenged if I could somehow autonomate the rumored convolution calculation when I got out with the G test. I don't think this is a calculation that humans do their best.
Internally --PySimpleGUI (I really wanted to use Kivy) --Area specification by mouse drag using OpenCV --Character recognition by Tesseract OCR I'm using, so I hope you can refer to these individually. Since I am using win32api, it does not work on UNIX as it is. Kivy is cooler and I'm used to it, so I wanted to use it, but I tried using PySimpleGUI for studying. But Kivy is modern and cool, so I recommend it.
The code is also below. https://github.com/poohsanforest/ConvCalcOCR
Required modules
- PySimpleGUI
- numpy
- opencv
- scipy
- pywin32
- pyocr
I don't think it's version dependent. You also need to install Tesseract OCR separately and add it to your environment variables.
How to use
- Enter the name of the window you want to read in the upper field and press the button on the right to display the screen in the lower right. Even if you enter the tab name in Chrome, a black screen will appear, so it is better to enter the tab name in Firefox. (Can't access without Chrome Driver?)
- When you press the filter button, a pop-up will appear. Select the convolution filter part so that the grid fits.
- Use the target button to select the target you want to convolve.
- If the number read in each text box is incorrect, correct it. (Misjudged as a percentage)
- Press the calculate button to display the convolution calculation result.
Test image (open in Firefox)
Appearance part
I am making a GUI part with PySimpleGUI. The redefine_rectangle and draw_shape functions may be helpful when getting coordinates by dragging the mouse.
import sys
import PySimpleGUI as sg
import numpy as np
import cv2
import win32gui
import win32ui
import win32con
import pyocr
import re
from PIL import Image
from scipy import signal
#Since I am using win32api, it does not work on unix as it is
class ConvCalc():
def __init__(self, **kwargs):
self.drawing = False
self.flag_drawing = False
self.flag_filter = True
self.roi_filter = [[0, 0], [50, 50]]
self.roi_target = [[0, 0], [50, 50]]
self.roi_temp = [[0, 0], [50, 50]]
self.image_filter = np.zeros((300, 300, 3), np.uint8)
self.image_target = np.zeros((300, 300, 3), np.uint8)
self.image_capture = np.zeros((300, 300, 3), np.uint8)
self.image_copy = np.zeros((300, 300, 3), np.uint8)
self.tool = None
#Start selecting rectangle
def redefine_rectangle(self, cols, rows, color):
cv2.namedWindow("set ROI")
cv2.setMouseCallback("set ROI", self.draw_shape)
self.flag_drawing = True
while 1:
image_rect = rectangle_grid(self.image_copy.copy(),
tuple(self.roi_filter[0] if self.flag_filter else self.roi_target[0]),
tuple(self.roi_filter[1] if self.flag_filter else self.roi_target[1]),
color,
cols, rows,
1)
cv2.imshow("set ROI", image_rect)
cv2.waitKey(1)
#× Button detection
ret = cv2.getWindowProperty('set ROI', cv2.WND_PROP_ASPECT_RATIO)
#I wonder if I should finish writing or break with the × button and not process the button
if not self.flag_drawing or ret == -1.0:
#Split image creation
divided_image = divide_image(self.image_copy, self.roi_filter if self.flag_filter else self.roi_target, cols, rows, 0.08)
score_list = []
for c, divided in enumerate(divided_image):
txt = self.tool.image_to_string(
Image.fromarray(cv2.cvtColor(divided, cv2.COLOR_BGR2RGB)),
lang="eng",
# builder=pyocr.builders.TextBuilder(tesseract_layout=6),
builder=pyocr.tesseract.DigitBuilder(tesseract_layout=6),
)
score_list.append(float(fix_num(txt)))
#Reshape to dimension of rowcol
score_list_reshaped = fix_textbox(score_list, rows, cols)
#Reflect the recognized numerical value
if self.flag_filter:
self.image_filter = image_rect
self.window['filterarray'].update(score_list_reshaped)
else:
self.image_target = image_rect
self.window['targetarray'].update(score_list_reshaped)
break
#Draw the final range
self.image_capture = rectangle_grid(self.image_target.copy() if self.flag_filter else self.image_filter.copy(),
tuple(self.roi_filter[0] if self.flag_filter else self.roi_target[0]),
tuple(self.roi_filter[1] if self.flag_filter else self.roi_target[1]),
color,
cols, rows,
1)
cv2.destroyWindow("set ROI")
#Draw while being careful not to cause an error when the size is 0
def draw_shape(self, event, x, y, flag, param):
if event == cv2.EVENT_LBUTTONDOWN:
self.drawing = True
self.roi_temp[0][1] = y
self.roi_temp[0][0] = x
elif event == cv2.EVENT_MOUSEMOVE:
if self.flag_filter: #Is there a smarter way?
if not (self.roi_filter[0][0] == x or self.roi_filter[0][1] == y) and self.drawing == True:
self.roi_filter = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
else:
if not (self.roi_target[0][0] == x or self.roi_target[0][1] == y) and self.drawing == True:
self.roi_target = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
elif event == cv2.EVENT_LBUTTONUP:
self.drawing = False
if self.flag_filter:
if not (self.roi_filter[0][0] == x or self.roi_filter[0][1] == y):
self.roi_filter = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
else:
if not (self.roi_target[0][0] == x or self.roi_target[0][1] == y):
self.roi_target = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
self.flag_drawing = False
# https://qiita.com/dario_okazaki/items/656de21cab5c81cabe59
def main(self):
#Section 1-Option settings and standard layout
sg.theme('Dark Blue 3')
# filter frame
filter_box = sg.Frame('FILTER', font='Any 15', layout=[
[sg.Text('kernelsize_col'), sg.Input("3", size=(3,None), key='kernelsize_0')],
[sg.Text('kernelsize_row'), sg.Input("3", size=(3,None), key='kernelsize_1')],
[sg.Button('filter', key='filter')],
[sg.Image(filename='', key='image_filter')],
[sg.Multiline('[0.0, 0.1, 0.2],\n[0.1, 0.2, 0.4],\n[0.0, 0.2, 0.4]', size=(24, 12), key='filterarray')],
])
target_box = sg.Frame('TARGET', font='Any 15', layout=[
[sg.Text('targetsize_col'), sg.Input("5", size=(3,None), key='targetsize_0')],
[sg.Text('targetsize_row'), sg.Input("5", size=(3,None), key='targetsize_1')],
[sg.Button('target', key='target')],
[sg.Image(filename='', key='image_target')],
[sg.Multiline('[0.0, 2.0, 8.0, 3.0, 5.0],\n[5.0, 1.0, 7.0, 6.0, 3.0],\n[2.0, 9.0, 3.0, 2.0, 1.0],\n[5.0, 4.0, 1.0, 4.0, 8.0],\n[0.0, 5.0, 2.0, 8.0, 5.0]',size=(24, 12), key='targetarray')],
])
calculated_box = sg.Frame('CALCULATED', font='Any 15', layout=[
[sg.Text('stride'), sg.Input("2", size=(3,None), key='stride')],
[sg.Button('calculate', key='calculate')],
[sg.Multiline('calculated',size=(24, 12),key='text_calculated')],
])
layout = [
# [sg.Text('Type the window name and press the button to capture it.\nIf it couldnt find the one, it will capture a whole area of a primary display.')],
[sg.Text('Enter the window name and click the button\n If not found, capture the main display as it is')],
[sg.Input("FireFox", size=(40,None), key='windowname'), sg.Button('windowcapture', key='windowcapture')],
[sg.Image(filename='', key='image_capture', size=(600,300), )],
[filter_box, target_box, calculated_box],
]
#Section 2-Window generation
self.window = sg.Window('ConvCalcOCR', layout)
# init OCR
tools = pyocr.get_available_tools()
if len(tools) == 0:
print("No OCR tool found")
sys.exit(1)
self.tool = tools[0]
print("Will use tool '%s'" % (self.tool.get_name()))
#Section 3-Event loop
while True:
event, values = self.window.read()
if event is None:
print('exit')
break
elif event == 'windowcapture':
self.image_capture = cv2.cvtColor(WindowCapture(values['windowname']), cv2.COLOR_BGRA2BGR)
self.image_copy = self.image_target = self.image_filter = self.image_capture.copy()
img = scale_box(self.image_capture, 600, 300)
imgbytes = cv2.imencode('.png', img)[1].tobytes()
self.window['image_capture'].update(data=imgbytes)
elif event == 'filter':
self.flag_filter = True #Is there a smarter way?
self.redefine_rectangle(int(values['kernelsize_0']), int(values['kernelsize_1']), (128, 128, 0))
img = scale_box(self.image_capture, 600, 300)
imgbytes = cv2.imencode('.png', img)[1].tobytes()
self.window['image_capture'].update(data=imgbytes)
elif event == 'target':
self.flag_filter = False
self.redefine_rectangle(int(values['targetsize_0']), int(values['targetsize_1']), (0, 0, 255))
img = scale_box(self.image_capture, 600, 300)
imgbytes = cv2.imencode('.png', img)[1].tobytes()
self.window['image_capture'].update(data=imgbytes)
elif event == 'calculate':
calculated = strideConv(values['targetarray'], values['filterarray'], int(values['stride']))
calculated = np.round(calculated, decimals=2) #Because there may be many decimal places due to calculation error
calculated_shape = np.array(calculated).shape
self.window['text_calculated'].update(fix_textbox(calculated.tolist(), calculated_shape[0], calculated_shape[1]))
#Section 4-Destroy and close windows
self.window.close()
Execution result
Grid type rectangle drawing
Draw with a grid to make it easier to match the range dragged with the mouse
#Rectangle with grid
def rectangle_grid(img, pt1, pt2, color, cols, rows, thickness=1, lineType=cv2.LINE_8, shift=None):
space_x = abs(pt2[0] - pt1[0]) / cols
space_y = abs(pt2[1] - pt1[1]) / rows
for col in range(cols+1):
img = cv2.line(img, (int(pt1[0]+col*space_x), pt1[1]), (int(pt1[0]+col*space_x), pt2[1]), color, thickness, lineType)
for row in range(rows+1):
img = cv2.line(img, (pt1[0], int(pt1[1]+row*space_y)), (pt2[0], int(pt1[1]+row*space_y)), color, thickness, lineType)
return img
Correction of range specification by dragging the mouse
When specifying the range by dragging the mouse, make sure that the image size becomes 0 at the beginning of drawing and the error that occurs and the addition and subtraction at the time of coordinate calculation do not become negative (I did not make a note of the page that I referred to. I'm sorry.)
#Negative coordinate support
def fix_coordinate(iy, ix, y, x):
x_sorted = sorted([ix, x])
y_sorted = sorted([iy, y])
# relu
return [[np.maximum(0, x_sorted[0]), np.maximum(0, y_sorted[0])],
[np.maximum(0, x_sorted[1]), np.maximum(0, y_sorted[1])]]
Image split
Divide the image into any number, and then remove the border of the table in the problem statement from the image by cutting the circumference by the eroderatio ratio.
#Image split
def divide_image(image, roi, cols, rows, eroderatio):
#Crop
cropped = image[roi[0][1]:roi[1][1], roi[0][0]:roi[1][0]]
# https://pystyle.info/opencv-split-and-concat-images/
chunks = []
for row_cropped in np.array_split(cropped, rows, axis=0):
for chunk in np.array_split(row_cropped, cols, axis=1):
#Remove borders that tend to be judged as characters
erode = int(eroderatio*min(chunk.shape[:2]))
chunk = chunk[erode:chunk.shape[0]-erode, erode:chunk.shape[1]-erode]
chunks.append(chunk)
return chunks
Correction of read numbers
In many cases, "." And "-" cannot be read normally, so ignore them once and format them assuming that the integer part has only one digit in the G test.
#Formatting assuming that the integer part is one digit
def fix_num(num):
fixed_num = re.sub("\\D", "", num) #Delete other than numbers
if fixed_num == '' or fixed_num == '-': # "-"Was treated as a number
return 0.0
else:
fixed_num = fixed_num[0] + '.' + fixed_num[1:]
return fixed_num
Screen display of read numerical values
Since there is no line break in the list of numerical values read by OCR and the display is broken as it is, disassemble and insert the line feed code.
#Shaping for the display
def fix_textbox(score_list, rows, cols):
score_array = np.reshape(np.array(score_list), (rows, cols))
score_list_reshaped= ["{}".format(l) for l in score_array.tolist()]
return ',\n'.join(score_list_reshaped)
Convolution calculation
The numerical value read by OCR is calculated by scipy after returning it because the parentheses of List are deleted for screen display.
#Convolution calculation
# https://stackoverflow.com/questions/48097941/strided-convolution-of-2d-in-numpy/48098534
def strideConv(v1, v2, s):
arr = np.array(eval('[' + v1 + ']'))
arr2 = np.array(eval('[' + v2 + ']'))
return signal.convolve2d(arr, arr2[::-1, ::-1], mode='valid')[::s, ::s]
Whole code
import sys
import PySimpleGUI as sg
import numpy as np
import cv2
import win32gui
import win32ui
import win32con
import pyocr
import re
from PIL import Image
from scipy import signal
#Since I am using win32api, it does not work on unix as it is
class ConvCalc():
def __init__(self, **kwargs):
self.drawing = False
self.flag_drawing = False
self.flag_filter = True
self.roi_filter = [[0, 0], [50, 50]]
self.roi_target = [[0, 0], [50, 50]]
self.roi_temp = [[0, 0], [50, 50]]
self.image_filter = np.zeros((300, 300, 3), np.uint8)
self.image_target = np.zeros((300, 300, 3), np.uint8)
self.image_capture = np.zeros((300, 300, 3), np.uint8)
self.image_copy = np.zeros((300, 300, 3), np.uint8)
self.tool = None
#Start selecting rectangle
def redefine_rectangle(self, cols, rows, color):
cv2.namedWindow("set ROI")
cv2.setMouseCallback("set ROI", self.draw_shape)
self.flag_drawing = True
while 1:
image_rect = rectangle_grid(self.image_copy.copy(),
tuple(self.roi_filter[0] if self.flag_filter else self.roi_target[0]),
tuple(self.roi_filter[1] if self.flag_filter else self.roi_target[1]),
color,
cols, rows,
1)
cv2.imshow("set ROI", image_rect)
cv2.waitKey(1)
#× Button detection
ret = cv2.getWindowProperty('set ROI', cv2.WND_PROP_ASPECT_RATIO)
#I wonder if I should finish writing or break with the × button and not process the button
if not self.flag_drawing or ret == -1.0:
#Split image creation
divided_image = divide_image(self.image_copy, self.roi_filter if self.flag_filter else self.roi_target, cols, rows, 0.08)
score_list = []
for c, divided in enumerate(divided_image):
txt = self.tool.image_to_string(
Image.fromarray(cv2.cvtColor(divided, cv2.COLOR_BGR2RGB)),
lang="eng",
# builder=pyocr.builders.TextBuilder(tesseract_layout=6),
builder=pyocr.tesseract.DigitBuilder(tesseract_layout=6),
)
score_list.append(float(fix_num(txt)))
#Reshape to dimension of rowcol
score_list_reshaped = fix_textbox(score_list, rows, cols)
#Reflect the recognized numerical value
if self.flag_filter:
self.image_filter = image_rect
self.window['filterarray'].update(score_list_reshaped)
else:
self.image_target = image_rect
self.window['targetarray'].update(score_list_reshaped)
break
#Draw the final range
self.image_capture = rectangle_grid(self.image_target.copy() if self.flag_filter else self.image_filter.copy(),
tuple(self.roi_filter[0] if self.flag_filter else self.roi_target[0]),
tuple(self.roi_filter[1] if self.flag_filter else self.roi_target[1]),
color,
cols, rows,
1)
cv2.destroyWindow("set ROI")
#Draw while being careful not to cause an error when the size is 0
def draw_shape(self, event, x, y, flag, param):
if event == cv2.EVENT_LBUTTONDOWN:
self.drawing = True
self.roi_temp[0][1] = y
self.roi_temp[0][0] = x
elif event == cv2.EVENT_MOUSEMOVE:
if self.flag_filter: #Is there a smarter way?
if not (self.roi_filter[0][0] == x or self.roi_filter[0][1] == y) and self.drawing == True:
self.roi_filter = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
else:
if not (self.roi_target[0][0] == x or self.roi_target[0][1] == y) and self.drawing == True:
self.roi_target = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
elif event == cv2.EVENT_LBUTTONUP:
self.drawing = False
if self.flag_filter:
if not (self.roi_filter[0][0] == x or self.roi_filter[0][1] == y):
self.roi_filter = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
else:
if not (self.roi_target[0][0] == x or self.roi_target[0][1] == y):
self.roi_target = fix_coordinate(self.roi_temp[0][1], self.roi_temp[0][0], y, x)
self.flag_drawing = False
# https://qiita.com/dario_okazaki/items/656de21cab5c81cabe59
def main(self):
#Section 1-Option settings and standard layout
sg.theme('Dark Blue 3')
# filter frame
filter_box = sg.Frame('FILTER', font='Any 15', layout=[
[sg.Text('kernelsize_col'), sg.Input("3", size=(3,None), key='kernelsize_0')],
[sg.Text('kernelsize_row'), sg.Input("3", size=(3,None), key='kernelsize_1')],
[sg.Button('filter', key='filter')],
[sg.Image(filename='', key='image_filter')],
[sg.Multiline('[0.0, 0.1, 0.2],\n[0.1, 0.2, 0.4],\n[0.0, 0.2, 0.4]', size=(24, 12), key='filterarray')],
])
target_box = sg.Frame('TARGET', font='Any 15', layout=[
[sg.Text('targetsize_col'), sg.Input("5", size=(3,None), key='targetsize_0')],
[sg.Text('targetsize_row'), sg.Input("5", size=(3,None), key='targetsize_1')],
[sg.Button('target', key='target')],
[sg.Image(filename='', key='image_target')],
[sg.Multiline('[0.0, 2.0, 8.0, 3.0, 5.0],\n[5.0, 1.0, 7.0, 6.0, 3.0],\n[2.0, 9.0, 3.0, 2.0, 1.0],\n[5.0, 4.0, 1.0, 4.0, 8.0],\n[0.0, 5.0, 2.0, 8.0, 5.0]',size=(24, 12), key='targetarray')],
])
calculated_box = sg.Frame('CALCULATED', font='Any 15', layout=[
[sg.Text('stride'), sg.Input("2", size=(3,None), key='stride')],
[sg.Button('calculate', key='calculate')],
[sg.Multiline('calculated',size=(24, 12),key='text_calculated')],
])
layout = [
# [sg.Text('Type the window name and press the button to capture it.\nIf it couldnt find the one, it will capture a whole area of a primary display.')],
[sg.Text('Enter the window name and click the button\n If not found, capture the main display as it is')],
[sg.Input("FireFox", size=(40,None), key='windowname'), sg.Button('windowcapture', key='windowcapture')],
[sg.Image(filename='', key='image_capture', size=(600,300), )],
[filter_box, target_box, calculated_box],
]
#Section 2-Window generation
self.window = sg.Window('ConvCalcOCR', layout)
# init OCR
tools = pyocr.get_available_tools()
if len(tools) == 0:
print("No OCR tool found")
sys.exit(1)
self.tool = tools[0]
print("Will use tool '%s'" % (self.tool.get_name()))
#Section 3-Event loop
while True:
event, values = self.window.read()
if event is None:
print('exit')
break
elif event == 'windowcapture':
self.image_capture = cv2.cvtColor(WindowCapture(values['windowname']), cv2.COLOR_BGRA2BGR)
self.image_copy = self.image_target = self.image_filter = self.image_capture.copy()
img = scale_box(self.image_capture, 600, 300)
imgbytes = cv2.imencode('.png', img)[1].tobytes()
self.window['image_capture'].update(data=imgbytes)
elif event == 'filter':
self.flag_filter = True #Is there a smarter way?
self.redefine_rectangle(int(values['kernelsize_0']), int(values['kernelsize_1']), (128, 128, 0))
img = scale_box(self.image_capture, 600, 300)
imgbytes = cv2.imencode('.png', img)[1].tobytes()
self.window['image_capture'].update(data=imgbytes)
elif event == 'target':
self.flag_filter = False
self.redefine_rectangle(int(values['targetsize_0']), int(values['targetsize_1']), (0, 0, 255))
img = scale_box(self.image_capture, 600, 300)
imgbytes = cv2.imencode('.png', img)[1].tobytes()
self.window['image_capture'].update(data=imgbytes)
elif event == 'calculate':
calculated = strideConv(values['targetarray'], values['filterarray'], int(values['stride']))
calculated = np.round(calculated, decimals=2) #Because there may be many decimal places due to calculation error
calculated_shape = np.array(calculated).shape
self.window['text_calculated'].update(fix_textbox(calculated.tolist(), calculated_shape[0], calculated_shape[1]))
#Section 4-Destroy and close windows
self.window.close()
#Rectangle with grid
def rectangle_grid(img, pt1, pt2, color, cols, rows, thickness=1, lineType=cv2.LINE_8, shift=None):
space_x = abs(pt2[0] - pt1[0]) / cols
space_y = abs(pt2[1] - pt1[1]) / rows
for col in range(cols+1):
img = cv2.line(img, (int(pt1[0]+col*space_x), pt1[1]), (int(pt1[0]+col*space_x), pt2[1]), color, thickness, lineType)
for row in range(rows+1):
img = cv2.line(img, (pt1[0], int(pt1[1]+row*space_y)), (pt2[0], int(pt1[1]+row*space_y)), color, thickness, lineType)
return img
#Image split
def divide_image(image, roi, cols, rows, eroderatio):
#Crop
cropped = image[roi[0][1]:roi[1][1], roi[0][0]:roi[1][0]]
# https://pystyle.info/opencv-split-and-concat-images/
chunks = []
for row_cropped in np.array_split(cropped, rows, axis=0):
for chunk in np.array_split(row_cropped, cols, axis=1):
#Remove borders that tend to be judged as characters
erode = int(eroderatio*min(chunk.shape[:2]))
chunk = chunk[erode:chunk.shape[0]-erode, erode:chunk.shape[1]-erode]
chunks.append(chunk)
return chunks
#Negative coordinate support
def fix_coordinate(iy, ix, y, x):
x_sorted = sorted([ix, x])
y_sorted = sorted([iy, y])
# relu
return [[np.maximum(0, x_sorted[0]), np.maximum(0, y_sorted[0])],
[np.maximum(0, x_sorted[1]), np.maximum(0, y_sorted[1])]]
#Formatting assuming that the integer part is one digit
def fix_num(num):
fixed_num = re.sub("\\D", "", num) #Delete other than numbers
if fixed_num == '' or fixed_num == '-': # "-"Was treated as a number
return 0.0
else:
fixed_num = fixed_num[0] + '.' + fixed_num[1:]
return fixed_num
#Shaping for the display
def fix_textbox(score_list, rows, cols):
score_array = np.reshape(np.array(score_list), (rows, cols))
score_list_reshaped= ["{}".format(l) for l in score_array.tolist()]
return ',\n'.join(score_list_reshaped)
#Image resizing
# https://pystyle.info/opencv-resize/#outline__3_5
def scale_box(img, width, height):
"""Fix the aspect ratio and resize it so that it fits in the specified size.
"""
h, w = img.shape[:2]
aspect = w / h
if width / height >= aspect:
nh = height
nw = round(nh * aspect)
else:
nw = width
nh = round(nw / aspect)
return cv2.resize(img, dsize=(nw, nh))
#Convolution calculation
# https://stackoverflow.com/questions/48097941/strided-convolution-of-2d-in-numpy/48098534
def strideConv(v1, v2, s):
arr = np.array(eval('[' + v1 + ']'))
arr2 = np.array(eval('[' + v2 + ']'))
return signal.convolve2d(arr, arr2[::-1, ::-1], mode='valid')[::s, ::s]
#Window capture
# https://qiita.com/danupo/items/e196e0e07e704796cd42
def WindowCapture(window_name: str, bgr2rgb: bool = False):
#Find the name of the currently active window
process_list = []
def callback(handle, _):
process_list.append(win32gui.GetWindowText(handle))
win32gui.EnumWindows(callback, None)
#Find the target window name
for process_name in process_list:
if window_name in process_name:
hnd = win32gui.FindWindow(None, process_name)
break
else:
#Get the entire screen if not found
hnd = win32gui.GetDesktopWindow()
#Get window size
x0, y0, x1, y1 = win32gui.GetWindowRect(hnd)
width = x1 - x0
height = y1 - y0
#Get device context for a window
windc = win32gui.GetWindowDC(hnd)
srcdc = win32ui.CreateDCFromHandle(windc)
memdc = srcdc.CreateCompatibleDC()
#Copy pixel information from device context,bmp
bmp = win32ui.CreateBitmap()
bmp.CreateCompatibleBitmap(srcdc, width, height)
memdc.SelectObject(bmp)
memdc.BitBlt((0, 0), (width, height), srcdc, (0, 0), win32con.SRCCOPY)
#Export bmp
if bgr2rgb is True:
img = np.frombuffer(bmp.GetBitmapBits(True), np.uint8).reshape(height, width, 4)
img = cv2.cvtColor(img, cv2.COLOR_bgr2rgb)
else:
img = np.fromstring(bmp.GetBitmapBits(True), np.uint8).reshape(height, width, 4)
#Clean up
# srcdc.DeleteDC()
memdc.DeleteDC()
# win32gui.ReleaseDC(hnd, windc)
win32gui.DeleteObject(bmp.GetHandle())
return img
if __name__ == '__main__':
ConvCalc().main()
reference
- https://qiita.com/dario_okazaki/items/656de21cab5c81cabe59
- https://pystyle.info/opencv-split-and-concat-images/
- https://pystyle.info/opencv-resize/#outline__3_5
- https://stackoverflow.com/questions/48097941/strided-convolution-of-2d-in-numpy/48098534
- https://qiita.com/danupo/items/e196e0e07e704796cd42
Finally
――I think there are many places that cannot be reached, so please feel free to point out. ――How many convolution calculations will you ask in the G test? ――I think I should have studied the time I was making this. ――Next, I would like to make an article about Kivy.
Postscript
In the 2020.11.7 exam, I didn't get a single convolution calculation question: angry:
Recommended Posts