What is the best matrces to compare improvment in compare to Target image?

Hi all,

I have a question regarding the best metrics for comparing improvements to a target image. I've already looked into PSNR, SSIM, MS-SSIM, MSE, and similar metrics, but they don't seem to show significant improvement, even though the improvement is visually evident see in the image.

image.

Could you suggest other metrics or methods that might better capture the visual differences and improvements? I’ve attached an example image for reference.

Additionally, if possible, could you provide sample code to implement the suggested metrics?

Thank you

Answer 1

[Drishti]

Hi Dynamic,

To measure the visual differences and improvement in image quality compared to target image, you can refer to Difference Structural Similarity (DSSIM) and Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) methods.

DSSIM is used to calculate the dissimilarity between the images and depicts sensitivity to visual changes.

BRISQUE is a no-reference image quality assessment method which compares image to a default model computed from images of natural scenes with similar distortions.

These methods can be directly put into use by following the below given code snippet:

%DSSIM
[ssim_value, ~] = ssim(img1, img2);  
% Calculate DSSIM
dssim_value = (1 - ssim_value) / 2;
%brisque method syntax
brisque_score = brisque(img);

For more information, you can refer to the MATLAB documentation of ‘ssim’ and ‘brisque’ functions:

https://www.mathworks.com/help/releases/R2024a/images/ref/ssim.html

https://www.mathworks.com/help/releases/R2024a/images/ref/brisque.html

I hope this helps