CODEWITHSUNDEEP
pythonimage
Silviu Petrut
Silviu Petrut

Reputation: 89

Calculate SNR of an image in db(decibel)

I create a script that adds gauss noise to an image, how can I calculate SNR of this image?

def noisy(img,sigma): # noise function
   # ...
   img = np.asarray(img,'double')   # numpy-array of shape (N, M);
   embeadead_line = img[0] # first line/row should not be changed


   img = np.delete(img, (0), axis=0)

   mean = 0.0   # some constant
   std = sigma    # some constant (standard deviation)double
   noisy_img = img + np.random.normal(mean, std, img.shape).astype(np.double)
   noisy_img_clipped = np.clip(noisy_img, 0, 255).astype(np.uint8)  # we might get out of bounds due to noise

   noisy_img_clipped = np.insert(noisy_img_clipped, 0, embeadead_line, 0) # insert embedead line back


   return noisy_img_clipped

Upvotes: 0

Views: 2163

Answers (1)

Flash Thunder
Flash Thunder

Reputation: 12036

SNR = μ/σ

where: μ is the average or expected value and σ is the standard deviation

So...

image_sd = np.std(noisy_img_clipped);
image_mean = np.mean(noisy_img_clipped);
image_snr = image_mean / image_sd;

If you want to get it in dB, that could be tricky, as you need additional constant depending of what you are measuring. If you don't know what it is, you could use 1, but generally showing SNR in dB for unknown source is considered as a bad approach.

import math    
image_snr_db = 1 * math.log(image_snr,2);

If you are asking about PSNR, the equasion is different and it needs 2 images. The original one and the blurred one. With OpenCV, you could calculate it like this:

import cv2
img1 = cv2.imread(original_image)
img2 = cv2.imread(noisy_image)
psnr = cv2.PSNR(img1, img2)

Upvotes: 1

Related Questions