How can I remove pixels from a 3D numpy array where the V channel equals 255?

Below is an example showing how you can avoid looping by using boolean masks. In your case you want to remove every pixel (indexed by h and s) if all channels equal a specific value (in your example, if the V channel equals 255). Instead of iterating, you can build a mask and then index the array directly.

For example, if you want to remove pixels from your HSV image where V equals 255, you can do the following:

PYTHONCopied
import cv2
import numpy as np

IMAGE_FILE = './images/2024-11-17/00.png'

# Load image (if you have transparency, make sure to load the correct number of channels)
image = cv2.imread(IMAGE_FILE, cv2.IMREAD_UNCHANGED)

# If your image has an alpha channel (4 channels), you may want to ignore those pixels.
# Otherwise, if you're just converting BGR to HSV, continue as below:
if image.ndim == 3 and image.shape[2] == 3:
    hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
elif image.ndim == 3 and image.shape[2] == 4:
    # If the transparency info is in the 4th channel you might use it as a mask.
    # Here we're assuming that fully transparent pixels have an alpha of 0.
    alpha = image[:, :, 3]
    bgr = image[:, :, :3]
    # Create a mask to keep only opaque or partially opaque pixels.
    valid_pixels = alpha != 0
    # Convert only the valid pixels to HSV (note that this will break the spatial structure)
    hsv_valid = cv2.cvtColor(bgr, cv2.COLOR_BGR2HSV)
    # Then apply the transparency mask to the HSV image
    hsv = hsv_valid[valid_pixels]
else:
    raise ValueError("Unexpected image shape.")

# Now create a mask to remove pixels where the V channel is 255.
# This mask will be True only for pixels NOT having V==255.
# If you've kept the spatial structure in hsv (height, width, 3), do:
mask = hsv[:, :, 2] != 255

# There are two common approaches:
#
# 1. If you want a 2D list of valid pixels (flattening the spatial structure):
hsv_removed_transparency = hsv[mask]

# 2. If you want to keep the same spatial shape, you might want to mask out the pixels:
hsv_masked = np.copy(hsv)
hsv_masked[~mask] = 0  # or any value indicating "removed"
#
# At this point hsv_removed_transparency is a 2D array (n_pixels, 3) containing only valid pixels.
# You can then compute your histogram on hsv_removed_transparency.

Explanation

• Loading the Image:
  If your PNG image has transparency you may load it with cv2.IMREAD_UNCHANGED to preserve the alpha channel.

• Creating a Boolean Mask:
  Using mask = hsv[:, :, 2] != 255, you create a mask that is True for pixels whose V channel is not 255.

• Applying the Mask:
  Using hsv[mask] returns a flattened array of HSV values for which the mask is True. If you need to retain the 2D structure, you may choose to work with the mask differently (for example, setting unwanted pixels to zero).

• Removing Transparent Pixels:
  If your image’s transparency was indicated with an alpha channel, you might want to combine an alpha mask and your HSV V-channel mask depending on your needs.

This vectorized approach is more efficient than looping through each pixel. Adjust the logic according to your specific needs regarding the transparency representation.