This is the second part of a final project for the freshmen course "Introduction to Computing"(Honor Track). The project mainly focused on image processing using OpenCV.

Specifically, SpeechDivider suggests the division of a long video(record of a class or presentation) based on the contents on the screen. Ideally, it would divide a presentation into small videos of every slides.

• Extract the screen position as ROI using Canny edge detector
• Then computes the similarity between neighboring ROI
  • Several methods are applied to improve robustness
  • Computing similarity using HSV distance/SIFT/perceptual hashing algorithm
  • Using dHash as similarity function performs best
• Suggest division when the similarity is below threshold

A typical frame of the video:

Extracted ROI