How do I calculate PNSR of an Image

Hello,

I have an image named "Foreman". I wish to estimate it's PSNR initial value in dB.

How do I achieve this ?

Thank you

[Rik]

Have a read here and here. It will greatly improve your chances of getting an answer.

Did you Google how to read an image? And how to calculate the PSNR?

Answer 1

[Shreeya]

For an m*n image "I" and it's noisy approximation "J" of 8-bit unsigned integer data type, the PSNR can be calculated using the following code implementation:

R = 255;
mse = mean((I-J).^2, "all");
if (mse ~= 0)
    psnr = 20*log10(R/sqrt(mse));
end

Apart from the editor, these computations can also be performed in Simulink through the PSNR block. You can refer to the link below to understand more about it:

https://www.mathworks.com/help/vision/ref/psnr.html

[DGM]

Even for uint8, this will be wrong unless both inputs were already mis-cast into a wider numeric class; otherwise truncation occurs in taking the MSE.

To avoid the whole problem, don't presume the input class. Expect only that it is properly-scaled for whatever class it has. Cast both images to floating point to take the MSE. Calculate the peak value based on whatever the class of the incoming images is.

% some test images
I = imread('cameraman.tif');
I = int16(I); % it's not uint8 anymore
J = imnoise(I,'gaussian',0.1);

% test it
R = diff(getrangefromclass(I));
mse = mean((double(I)-double(J)).^2, "all");
if (mse ~= 0)
    P1 = 20*log10(R/sqrt(mse))
end

P1 = 16.9947

% compare
P2 = psnr(J,I)

P2 = 16.9947

This makes the same assumptions that IPT psnr() makes. I and J must both have the same class and must be properly-scaled for their class. The supported classes are only those handled by IPT getrangefromclass(), so it's still not totally class-agnostic, but neither is psnr().

Answer 2

[Image Analyst]

There is a psnr function in the Image Processing Toolbox.

help psnr

 PSNR Peak Signal-To-Noise Ratio.
    PEAKSNR = PSNR(A, REF) calculates the peak signal-to-noise ratio for
    the image in array A, with the image in array REF as the reference. A
    and REF can be N-D arrays, and must be of the same size and class.
  
    PEAKSNR = PSNR(A, REF, PEAKVAL) uses PEAKVAL as the peak signal value
    for calculating the peak signal-to-noise ratio.
  
    [PEAKSNR, SNR] = PSNR(A, REF, __) also returns the simple
    signal-to-noise in SNR, in addition to the peak signal-to-noise ratio.
 
    [___] = PSNR(___,Name,Value) accepts name value pairs to control aspects
    of computation. Supported options include:
 
        'DataFormat'            Dimension labels of the input data A and REF
                                specified as a string scalar or character
                                vector. The format options 'S','C', and 'B' 
                                are supported. The options 'S', 'C' and 'B' 
                                correspond to spatial, channel, and batch 
                                dimensions, respectively. For data with a 
                                batch or 'B' dimension, the output PEAKSNR 
                                and SNR will contain a separate result for 
                                each index along the batch dimension.
                                As an example, input RGB data with two
                                spatial dimensions and one channel
                                dimension would have a 'SSC' DataFormat.
                                Default: All input dimensions in A treated as
                                spatial dimensions.
   
    Notes
    -----
    1. When dlarray input is labeled and contains a batch dimension, psnr
    yields a separate result for each element along the batch dimension.
 
    Class Support
    -------------
    Input arrays A and REF must be one of the following classes: uint8,
    int16, uint16, single, or double. Both A and REF must be of the same
    class. They must be nonsparse. PEAKVAL is a scalar of any numeric
    class. PEAKSNR and SNR are scalars of class double, unless A and REF
    are of class single in which case PEAKSNR and SNR are scalars of class
    single.
 
    Example
    ---------
   % This example shows how to compute PSNR for noisy image given the
   % original reference image.
  
    ref = imread('pout.tif');
    A = imnoise(ref,'salt & pepper', 0.02);
  
    [peaksnr, snr] = psnr(A, ref);
  
    fprintf('\n The Peak-SNR value is %0.4f', peaksnr);
    fprintf('\n The SNR value is %0.4f \n', snr);
 
    See also IMMSE, MEAN, MEDIAN, SSIM, SUM, VAR.

    Documentation for psnr
       doc psnr

    Other uses of psnr

       dlarray/psnr