python - OpenCV multicolor thresholding -

I am trying to multicolor thresholding on an opencv cv2 image. The problem I am trying to solve is the following:

• Every "valid" list is in R, G, B
• If one pixel r, j For all valid values,

for example

• [221, make pixel (0,0,0) again, otherwise, make it (255) , 255, 255) 180, 50] is considered valid in R channel
• [23, 18, 2] is considered to be valid in live channel
• [84, 22, 48 ] B is considered to be valid in the channel

Then if the following value (RGB commands in a pixel

(221, 23, 48)

(221, 23, 48)

(221, 23, 84)

(221, 18, 84)

(221, 18, 22)

(221, 18, 48)

/ Li>

(50, 2, 48)

It will change in (0,0,0), otherwise (255255255)

Currently, I'm doing it with looped nested for loop:

  in range for x (width): for y in range (height): imcv [y, x] = threshold [Y, x])  

Where thresholds function displays the arguments described above. Note that although I did this place, in-place transformation is not required.

The method I currently use, but it is still very slow I believe there should be a better way in OpenCV / Numpy. I am very new to both the structure and can not understand how

I researched OpenCV thresholding functions, it seems that they can only work on a channel gray scale image, it also series continuous range What do I need to do thresholding on all 3 channels at discrete values? I think there is a custom function to pass there, but I am unable to find the right API in my docs.

I also saw the numpy API that I could use, like ufunc . It seems that what I want to achieve here to achieve it, or I did not see it.

Any help is appreciated.

Edit:

Thanks to both Abidaram's and care, both solutions achieved more than x1500 improvement in performance.

  ncalls tottime percall cumtime percall filename: lineno (function) 1 1.576 1.576 1.576 1.576 test.py:48 (preprocess2_image) 1 0.000 0.000 0.001 0.00.py test.py:79 (preprocess2_image3) 1 0.000 0.000 0.001 0.001 test.py:66 (preprocess2_image2)  

please Try:

z1 = np.dstack ([np.in1d ​​(IMG [..., 0], b), NPNN1D (IMG [..., 1], G), NPN.N.DD (IMG [..., 2], R.]]). Reshape (img.shape) q = np.all (z1, axis = 2) out = np.uint8 (q) * 255

np.in1d ​​(a, B) gives you a similar level of boolean array with true , if that element is in B, else incorrect This is a vector equivalent of the in method in Python or in short:

np.in1d ​​(a, b) & lt; ==> [i if i in b other wrong] i

You do it for all channels, i.e. check the first channel for valid values ​​in B , Second with G and second with R.

Then you place them in the z-direction stack np.dstack . Why z-direction? Because we want in the format BGR-BGR-BGR ...

But remember, this is a 1D array, so we resize it into our original image size using the X.reshape (img.shape) method.

So now you have a boolean mask where true is valid if it is valid and false.

This is in the first line of code.

Now you want to see a valid BGR combination. A combination is valid if all B, G, R components are true. Then you apply np.all () in the z-direction then you get a boolean mask quote

q will be a boolean mask with valid colors and others as correct.

Then you can change the integer data type, True -> 1 and False -> 0

Then you multiply with 255. If you want an inverted image, you can use np.bitwise_not