Hi,
Le mar. 19 sept. 2023 à 21:15, Ralph Little <[email protected]> a écrit :
> It would be cool to see some samples of that. It might be something that
> is easy to fix.
> Even better would be to create an issue on our GitLab backends page where
> the information you have supplied could be kept and progress tracked.
>
> Done. I hope this is the correct place. I had some issues with formatting
and # characters, so, something got lost in the post on gitlab. Hopefully
the portions of images are relevant.
> If you don't feel happy to do that then you could send some sample images
> to me personally and I will open the issue there. It also means that it
> doesn't get forgotten.
>
> Many thanks.
In the meantime, I played with python and openCV. I could make something
recognizable at least with the attached script. Nonetheless, integrating a
gradient is usually a bad idea because it is too sensitive to noise.
Cheers,
Stephane Louise
#import cv2, numpy and matplotlib libraries
import cv2
import numpy as np
import matplotlib.pyplot as plt
img=cv2.imread("color-ex.png")
#img=cv2.imread("partial.png")
#img= cv2.imread("extract.png")
print("Image size:",len(img),";",len(img[0]))
img2= img.copy()
mb=0
mg=0
mr=0
nb=0
for xi in img[0]:
mb+= xi[0]
mg+= xi[1]
mr+= xi[2]
nb+=1
print("Rm=",mr/nb," Gm=",mg/nb, "Bm=",mb/nb)
# means for each color channel
meanr= int(mr/nb)
meang= int(mg/nb)
meanb= int(mb/nb)
for xi in range(len(img)):
curvalr= 0
curvalg= 0
curvalb= 0
nb=0
for yj in range(len(img[0])):
pixel= img[xi,yj]
# use the means to find the signed gradient value (don't ask about -0.75, it just works better experimentally, that's all)
deltb= pixel[0]-meanb-0.75
deltg= pixel[1]-meang-0.75
deltr= pixel[2]-meanr-0.75
# blue val managemen (integrate)t
curvalb+= deltb
if curvalb<0:
curvalb=0
if curvalb>255:
curvalb=255
# green val management (integrate)
curvalg+= deltg
if curvalg<0:
curvalg=0
if curvalg>255:
curvalg=255
# red val management (integrate)
curvalr+= deltr
if curvalr<0:
curvalr=0
if curvalr>255:
curvalr=255
# some random correction of brightness to improve the output
r= int(curvalr+30) if curvalr<225 else int(curvalr)
g= int(curvalg+30) if curvalg<225 else int(curvalg)
b= int(curvalb+30) if curvalb<225 else int(curvalb)
# I don't know why the color channels are switched between red and blue
img2[xi,yj]= (r, g, b)
# show result
plt.imshow(img2)
#hold the window
plt.waitforbuttonpress()
plt.close('all')
