Signal to Noise Ratio (SNR)

Hi everone!

I wanted to add gaussian noise to an image. I used the command like noisy=imnoise (image, 'gaussian', 0, 0.05), it makes the image so noisy. In different Journal papers different researchers are claiming that they are adding gaussian noise with the power such as 20dB, 25dB etc. moreover their reported images are also in good quality. I am unable to solve this mystery, as this command gives some floating type result but not as a whole number. I am also attaching the published papers. Please anyone have a look on them and reply, Thanks.

Answer 1

[Thorsten]

20dB, 25dB is not the power of the noise, but the signal to noise ratio (SNR) in decibel.

It is computed as SNRdb = 10*log10(sigma_signal^2/sigma_noise^2); For imnoise you provide the variance of the noise, i.e., sigma_noise^2. You have to transform the above equation to determine sigma_noise for the desired SNRdB and the sigma_signal of your image.

This little demo creates noise of different SNR using either randn or imnoise; both give same results.

I = im2double(imread(('lena'));
varI = std2(I)^2;

SNRdB = 5:5:30;

for i=1:numel(SNRdB)
  sigma_noise = sqrt(varI/10^(SNRdB(i)/10));
  N = sigma_noise*randn(size(I));
  IN1 = I+N; % using randn
  IN2 = imnoise(I, 'Gaussian', 0, sigma_noise^2); % using imnoise
  imshow([IN1 IN2])
  title(['SNR = ' int2str(SNRdB(i)) 'dB' ...
    ', \sigma_{noise} = ' num2str(sigma_noise)]);
  disp('Press any key to proceed')
  pause

end

[Varoun Hanooman]

This was very helpful, thank you very much

[AMIR KHAN]

Very Helpful

[Mahmoud Yassin]

Very helpful, thanks a lot

Answer 2

[Toqeer Mahmood]

Thanks Thorsten for you kind reply and help.

[Thorsten]

Toqeer, please be so kind to formally "Accept" my answer.