Full reference image quality assessment using a statistical model and Machine Learning Regression (VIF)
This is the implementation of a Full Reference Image Quality Assessment based on Information fidelity criteria (Visual Information Fidelity). This implementation focuses on the Visual Information Fidelity (VIF) Index, or Sheikh-Bovik Index, which is a specific and quite successful implementation of the information fidelity–based approach.
For the theoretical part, you can refer to the original paper.
The output of the method is a value between 0 and 1. We take two images for comparison
- Original image (img2o.jpg)
- distorted image (img2c.jpg) Note that the second image is the distorted image we get when we lose some information in the communication channel. Hence, one can note that the image size (img2c.jpg) is less than the size of the original image.
Then we run our main function, which calculates the VIF index.
Top-down Approaches for Full Reference Image Quality Assessment (Information Theoretical Approach)
-
We calculate the visual information fidelity index (VIF) of the reference image and distorted image a. Main file: full_ref_tech.py b. sub files: utils.py c. Reference Image: img2o.jpg d. Distorted Image: img2c.jpg e. Rotated Image: img2cr.jpg
-
Libraries installed in python: sciy, numpy, skimage (scikit-image).
-
First, go to the reference folder in which you have all your code files and images for testing, and then go to Python shell and run these commands. a. from skimage import io b. i = io.imread('img2o.jpg') c. i1 = io.imread('img2c.jpg') d. from full_ref_tech import vif e. vif(i,i1) %this will return VIF index of above 2 images
you can try your own images but ideally size shoild be same. rotation is prohibited.
Thank you.