I have a few sequences of images, taken at fixed camera position, that address a moving object, which means the object moves towards the camera. But all the sequences of images are the cropped images of the object and all images are in the same size (grayscale images). From naked eyes, we can see the images that are closer to the camera are better than the ones that taken at far distance. How can I show scientifically or show a proof that the images have varied quality? I also want to compare the quality between sequences.
You are apparently in the context of no-reference, reference-free or blind image quality assessment. The topic is quite active, and I am not sure people have already a completely accepted framework for that. Multiple distortions may affect images: random noise, compression artifact, static blur, motion blur, etc. They require different metrics (benchmark data here). Some references to start with, and help you for more focused questions:
- No-Reference Image Quality Assessment in the Spatial Domain, 2012
- Making a “Completely Blind” Image Quality Analyzer, 2013
- No-reference image quality assessment based on spatial and spectral entropies, 2014
- MDID: A multiply distorted image database for image quality assessment, 2017
The first three papers come from the same group, an hint that the paint is still wet.