A framework for index point detection using effective title extraction from video thumbnails

For content based indexing of videos, numerous tools and techniques are pipe-lined. The major challenge that these techniques face is the accuracy of index points generated. This paper presents an efficient way to extract text from video frames along with its timestamps. Text extraction takes place in a three-step method which combines pre-processing of extracted Video Frames, similarity measurement for removing ambiguous frames and finally text extraction using PyTesseract Optical Character Recognition. The educational videos with presentations are prioritised. Text extraction is applied upon the headings of that presentation. These extracted keywords are referred to as Index Points through out the article.