Hi,
I know in MATLAB 'imnoise' command scale the image between [0,1], so the noisy image that generates using the following command "noisy_image = imnoise(Original_im,'gaussian',0.1)" with variance '0.1' is more degraded than "noisy_image = original_im + sqrt(0.1)*randn(n,m) " command. Now my questions are:
1. Why image is scaled between [0,1]?
2. How the variance is scaled in "noisy_image = original_im + sqrt(variance)*randn(n,m) " command? It means if i have a variance of '200', what should be the value that corresponds to MATLAB command 'imnoise' variance.
The power of a noise process is proportional to its variance, so if you choose a Gaussian noise with the variance of 0.1, its power is equal to 0.1*C.
On the other hand the power of a signal is proportional to the mean square of the gray level of its pixels (or in terms of peak signal power, its power equals the maximum value of its gray level values, squared).
Now, suppose you want to add noise to an image and get an image having a given peak SNR (or SNR). You will need to know the peak signal power to calculate the required noise power. A shortcut to this is normalizing the image before adding noise. In this way, the peak signal power will be equal to 1. Therefore, peak SNR of the noisy image will be reversely proportional to the variance of the noise.
% An example MATLAB script:
clc, clear all, close all
I = imread('eight.tif');
SIG = 0.5;
J1 = imnoise(I,'gaussian', 0, SIG^2);
J2 = uint8(255*mat2gray(double(I) + SIG * randn(size(I))));
[PSNR1, SNR1] = psnr(J1, I);
[PSNR2, SNR2] = psnr(J2, I);
subplot 131, imshow(I), title('Original image')
subplot 132, imshow(J1), title(['imnoise, PSNR = ', num2str(PSNR1)])
subplot 133, imshow(J2), title(['your noise, PSNR = ', num2str(PSNR2)])
You can open imnoise in MATLAB and see how it works, use ctrl + D
You can do it generally with other functions.
All the best
I have the same question. I checked both methods of introducing Gaussian noise into an image and obtained different results.
- Badges
- Science topic