Hi,
I am trying to detect changes in text in a video. The video is slide over powerpoint so I just want to extract the times when the slide transitions in the video.
Initially I tried using changes in average intensity from frame to frame to identify a transition/change but sometimes the amount of text on two slides is the same intensity even though the text changes.
Then I thought of calculating the center of mass of each image, which didn't work but did lead me to the notion of higher order moments. So what I wound up doing was using the third moment divided by the second moment and detecting a change in value over a threshold (.05) and it worked great! I was using ONLY the blue portion of the image to do this (so I am aware that some changes where only the green or red change would be undetected but I am guessing this is rare with ppt slides).
So now I am wondering 2 things: 1) Is there a better or more standard way to detect changes in text in a video (faster or more accurate)? 2) What exactly did I actually wind up measuring???
The formula I am using is x = m03/m02 and y = m30/m20. I am guessing some sort of measure of skewness? Or is it something else?