def otsu(source_image):

    total_pixel_count = source_image.shape[0] * source_image.shape[1]

    # build histogram: brightness -> count
    gray_histogram = dict()
    for i in range(source_image.shape[0]):
        for j in range(source_image.shape[1]):
            current_pixel_value = source_image[i, j]
            gray_histogram[current_pixel_value] = gray_histogram.get(current_pixel_value, 0) + 1

    total_weighted_sum = sum([brightness * count for brightness, count in gray_histogram.items()])

    # -1 for monochromatic images
    best_threshold = -1
    max_criterion_value = 0

    left_weighted_sum = 0
    left_pixel_count = 0

    for current_threshold, count in list(gray_histogram.items())[:-1]:
        left_weighted_sum += current_threshold * count
        left_pixel_count += count

        right_pixel_count = total_pixel_count - left_pixel_count

        left_mean = left_weighted_sum / left_pixel_count
        right_mean = (total_weighted_sum - left_weighted_sum) / right_pixel_count

        current_criterion_value = left_pixel_count * right_pixel_count * (left_mean - right_mean) ** 2

        if current_criterion_value > max_criterion_value:
            max_criterion_value = current_criterion_value
            best_threshold = current_threshold

    destination_image = None

    if best_threshold != -1:
        destination_image = np.zeros(source_image.shape, source_image.dtype)
        for i in range(destination_image.shape[0]):
            for j in range(destination_image.shape[1]):
                if source_image[i, j] > best_threshold:
                    destination_image[i, j] = np.iinfo(source_image.dtype).max
    else:
        destination_image = source_image

    return best_threshold, destination_image