3D videos are usually saved either as Side-By-Side images, or Over-Under images.
My TV can automatically detect when 3D content is being played - even when there's no specific file metadata. It does this by analysing the images and recognising their stereo nature.
I'm aware of how to calculate a disparity map once I have a stereo pair - but how do I detect whether the image is HSBS or OU?
One tactic would be to separate the image into top/bottom and left/right - then compute the differences between the pairs. Whichever is shorter, wins.
I could use Mean Squared Error or Structural Similarity Measure
Or I could compare the histograms
But these feel a bit hacky and they only give me an approximation of whether two sides are similar.
Is there any way to detect the vertical or horizontal split?