How to effectively loop through each pixel for saving time with numpy?

Question

As you know looping through each pixels and accessing their values with opencv takes too long. As a beginner I'm trying to learn opencv myself when I tried this approach it took me around 7-10 seconds of time to loop through image and perform operations.

code is as below

original_image = cv2.imread(img_f)

image = np.array(original_image)
for y in range(image.shape[0]):
    for x in range(image.shape[1]):
        # remove grey background
        if 150 <= image[y, x, 0] <= 180 and \
                150 <= image[y, x, 1] <= 180 and \
                150 <= image[y, x, 2] <= 180:
            image[y, x, 0] = 0
            image[y, x, 1] = 0
            image[y, x, 2] = 0

        # remove green dashes
        if image[y, x, 0] == 0 and \
                image[y, x, 1] == 169 and \
                image[y, x, 2] == 0:
            image[y, x, 0] = 0
            image[y, x, 1] = 0
            image[y, x, 2] = 0

in above code i'm just trying to remove grey and green pixel colors.

I found similar question asked here but im not able to understand how to use numpy in my usecase as i'm beginner in python and numpy.

Any help or suggestion for solving this will be appreciated thanks

np.where searches for pixels that match the condition and you can then set those to some value or color. See sparkbyexamples.com/numpy/numpy-where-multiple-conditions. — fmw42
– fmw42, Commented Dec 3, 2022 at 5:45

kmkurn · Accepted Answer · 2022-12-03 08:52:16Z

3

You can take advantage of NumPy's vectorized operations to eliminate all loops which should be much faster.

# Remove grey background
is_grey = ((150 <= image) & (image <= 180)).all(axis=2, keepdims=True)
image = np.where(is_grey, 0, image)

# Remove green dashes
is_green_dash = (image[..., 0] == 0) & (image[..., 1] == 169) & (image[..., 2] == 0)
is_green_dash = is_green_dash[..., np.newaxis]  # append a new dim at the end
image = np.where(is_green_dash, 0, image)

Both invocations of np.where rely on NumPy's broadcasting.

edited Dec 3, 2022 at 8:52

answered Dec 3, 2022 at 6:03

kmkurn

7154 silver badges13 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

Phinnik · Accepted Answer · 2022-12-03 05:41:01Z

0

You can apply numpy filtering to image. In your scenario it would be:

mask_gray = (
    (150 <= image[:, :, 0]) & (image[:, :, 0] <= 180) & 
    (150 <= image[:, :, 1]) & (image[:, :, 1] <= 180) & 
    (150 <= image[:, :, 2]) & (image[:, :, 2] <= 180)
)

image[mask_gray] = 0

mask_green = (
    (image[:, :, 0] == 0) &
    (image[:, :, 1] == 169) &
    (image[:, :, 2] == 0)
)

image[mask_green] = 0

mask_gray and mask_green here are boolean masks

answered Dec 3, 2022 at 5:41

Phinnik

516 bronze badges

Comments

Christoph Rackwitz · Accepted Answer · 2022-12-03 22:57:25Z

0

If you insist on writing your own loops...

You could just use numba. It's a JIT compiler for Python code.

from numba import njit

@njit
def your_function(input):
    ...
    return output

input = cv.imread(...) # yes this will shadow the builtin of the same name
output = your_function(input)

The first time you call your_function, it'll take a second to compile. All further calls are blazingly fast.

answered Dec 3, 2022 at 22:57

Christoph Rackwitz

16.4k5 gold badges42 silver badges56 bronze badges

Collectives™ on Stack Overflow

How to effectively loop through each pixel for saving time with numpy?

3 Answers 3

Comments

Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related