- Can I tell that different versions of the same video are correlated to each other?
- and if so, how can I show this correlation? for example, high-quality and poor-quality versions of the same video.
- Can we say various episodes of the same tv series are correlated since they mostly share the same background, characters, and scene?
- and if so, how I can extract this similarity?
- Maybe it's better to ask, how can I compare two videos and show that they are statistically similar or they are correlated?
-
$\begingroup$ Those are five questions, so I removed the meaningless "I have two questions" preamble ;) Anyway, understand you video as multidimensional signal and then just throw the definition of correlation at it. That solves parts of your questions $\endgroup$– Marcus MüllerCommented Aug 13, 2018 at 20:52
-
$\begingroup$ I just wanted to make my question more clear by dividing it into 2 parts. Thank you for editing it anyway. Actually, I need more details for doing the process would you please tell me more explanation $\endgroup$– BonnieCommented Aug 13, 2018 at 21:52
-
1$\begingroup$ Do you think that you could make the question a bit more specific? Most of the questions you pose have positive answers. Yes you can quantify similarity between two videos and yes, videos of the same TV series are "similar". The point is in how you estimate this similarity (which is the thing that makes this question broad) and this depends on what you are trying to do. So, what are you trying to achieve? $\endgroup$– A_ACommented Aug 14, 2018 at 7:40
-
$\begingroup$ Dear @A_A thank you for your answer. what I want to do is to cluster these correlated files, (among all other files which they may belong to other TV series or other categories so they can make other clusters), based on the correlation or similarity between them as the similarity measure.I would be grateful if you have any idea that helps me to understand how I can compare two videos and say the correlation(or similarity) between them is %x (or maybe we should make this comparison frame by frame and say their corresponding frames are %x similar,) $\endgroup$– BonnieCommented Aug 14, 2018 at 14:05
1 Answer
The video is actually a sequence of images (frames, for a PAL standard that is 25 frames) and in such a sequence you can apply objective measures of quality to the certain number of frames. The most frequently used quality measures are PSNR (Peak Signal to Noise Ratio)1 and SSIM (Structural Similarity Index) 2. In the matlab, there is a built-in fuction that calculate PSNR and SSIM:
psnr(image, ref_image) PSNR in Matlab
ssim(image, ref_image) SSIM in Matlab