Fastest Way to get Luma Rec.709 from RGB

[blue-j]

Right now I am:

# convert image to YCbCr to get the luma 709 values from Y' (709 coefficients are default)
img_ycbcr = RGB_to_YCbCr(unique_colors)
luma709_values = img_ycbcr[:, 0]

is there a faster way?

thank you! : )
J

[tjdcs]

It would be faster to take the Y row from the 709 matrix and use it directly with numpy. But if you later need the chroma data anyway, then you should compute it once.

If you share a characteristic script (with random RGB values) that you are trying to profile I might be able to help.

[blue-j]

i am iterating through tens of thousands of unique RGB values. i do not need the chroma data.
thanks for sharing your knowledge and helping me out! below is the function in which i am computing these values.

def process_colors(color_counter, total_pixels):
    color_data = []

    # convert entire image to HSV for efficiency
    unique_colors = np.array(list(color_counter.keys())) / 255  # Normalize RGB values
    hsv_colors = RGB_to_HSV(unique_colors)  # Convert all RGB to HSV at once

    # convert image to YCbCr to get the luma 709 values from Y' (709 coefficients are default)
    img_ycbcr = RGB_to_YCbCr(unique_colors)
    luma709_values = img_ycbcr[:, 0]

    # denormalize
    rgb_denormalized = unique_colors * 255
    # Convert entire array to HSV and scale the components
    hsv_scaled = hsv_colors * np.array([360, 100, 100])

    # Slice the array to get hue, saturation, and brightness components
    hue = hsv_scaled[:, 0]
    saturation = hsv_scaled[:, 1]
    brightness = hsv_scaled[:, 2]
    luma709 = luma709_values * 255

    for i in range(len(unique_colors)):
        rgb_tuple = tuple(rgb_denormalized[i].astype(int))
        count = color_counter[tuple(rgb_tuple)]
        percentage = 100 * count / total_pixels
        hexadecimal = "#{:02X}{:02X}{:02X}".format(*rgb_tuple)
        color_info = (rgb_tuple[0], rgb_tuple[1], rgb_tuple[2], hexadecimal, float(luma709[i]),
                      float(hue[i]), float(saturation[i]), float(brightness[i]), int(count), float(percentage))
        color_data.append(color_info)

    return color_data

[KelSolaar]

Yeah, I would look at colour.matrix_YCbCr definition.