I have the following python function below which takes in an image and a noise coefficent as input parameters such that the percentage of total pixels converted to noise is proportionate to the coefficient (represented as the noise matrix). This noise matrix is then multiplied to the original image to obtain the noisy image.
def dot_salt_pepper(img, noise_coeff):
height = img.shape[0]
width = img.shape[1]
noise_matrix = np.ones([height, width])
num_noise_pixels = int(height * width * noise_coeff)
for i in range(num_noise_pixels):
noise_matrix[np.random.randint(height)][np.random.randint(width)] = 0
new_img = np.multiply(img, noise_matrix)
new_img = np.uint8(img)
return new_img
I understand that in order to add dot noise to an existing image, the image must first be converted to grayscale. However the image I am working with came in grayscale format, thus when I read it in OpenCV using cv2.imread('img.jpg'), calling print(img.shape) outputs the tuple (3000, 4000, 3) which suggests that the BGR channels are still present in the image (thus I cannot multiply the 2D noise matrix to the 3D image matrix). How do I go around this issue?
>Solution :
The image is in RGB datatype, even though it has no colour
Convert to greyscale
This can be done on read of the image cv2.imread(IMAGE_FILE, cv2.IMREAD_GRAYSCALE)