I am performing video classification using CRNN model. I have extracted the frames from the videos and subtracted their background. Which has created alot of blank frames for leading and trailing number of frames. I want to reject the blank frames. Can anyone suggest a suitable method to reject the blank frames?