问题
Let's take a look at this chessboard (https://prnt.sc/r8vjth) As you can see, there are two greenish squares (g8 and f6), which indicate the enemy's move. Square g8 is empty and is pretty easy to find by locateOnScreen function. Problems begin when I am trying to find f6 because it returns again the g8's position.
For example, in screenshot's position, it finds the g8. First part is correct, and it comes from the h parameter in the code below. But the second g8 is coming from this function, which is not correct.
def pos_loop(h): # h is the first position (Second square's position) which is struct consisting of top, left, width, height # I assume that h is correct because it always gives the right position values
global previous_green
global previous_white
for p1 in pyautogui.locateAllOnScreen("C:\\Users\\Admin\\Desktop\\chesspic\\small_on_white.png"):
if p1.height != h.height and p1.width != h.width and p1 != previous_white and p1 != previous_green:
previous_white = p1
print("\n Returned P1")
return p1
for p2 in pyautogui.locateAllOnScreen("C:\\Users\\Admin\\Desktop\\chesspic\\small_on_green.png"):
if p2.height != h.height and p2.width != h.width and p2 != previous_green and p2 != previous_white:
previous_green = p2
print("\n Returned P2")
return p2
return None
previous_green and previous_white are just global variables, which are set to None at the start of the program. Even though I implemented a few checks, to make sure, that second position is not equal to the first, it is missing it randomly. Most strange thing is that this error is random. Sometimes it finds the right move, sometimes not.
Maybe you know how to solve that finding correct second image (f6) problem.
回答1:
If you are new to image processing, I'll show you something simple that might get you started. I'll load your image as a PIL/Pillow Image, and convert it to a Numpy array - you should be able to get such an Image or array from pyautogui.
from PIL import Image
import numpy as np
# Load your image and make into Numpy array for processing
im = Image.open('chess.png').convert('RGB')
na = np.array(im)
Next, I used a colour dropper to sample the green colours on g8 and f6 to get their RGB values
greenA = [246,246,130]
greenB = [186,202,68]
Now I can make a mask that is True wherever your image is that colour:
maskA = (na[:] == greenA).all(2)
maskB = (na[:] == greenB).all(2)
If you want to visualise the masks, you can make them into images and display them:
Image.fromarray((maskA*255).astype(np.uint8)).show()
Or save them as files:
Image.fromarray((maskA*255).astype(np.uint8)).save('a.png')
If you now sum the pixels across the rows like this:
rowTotals = np.sum(maskA,axis=1)
array([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 65, 65, 4, 4, 4, 59,
59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59,
59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59,
59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 52, 51, 53, 53, 53,
51, 52, 56, 4, 4, 4, 65, 65])
You can see that all the rows that are fully black total up to zero, and only the last few rows are greater than zero - so you can find the coordinates of your coloured square.
Likewise, if you sum the pixels down the columns like this:
colTotals = np.sum(maskA,axis=0)
array([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 65, 65, 4, 4,
4, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59,
59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59,
59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 59, 54, 52, 52,
55, 55, 51, 51, 51, 4, 4, 4, 65, 65, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0])
You can see the first few columns are empty (black or zero) and then there are your coloured pixels, followed by lots more black ones.
Essentially, I am doing this:
来源：https://stackoverflow.com/questions/60447567/finding-correct-second-image-by-smaller-representation-of-first