Some Aspects of Text Recognition from Video Document in Education 4.0

Optical character recognition in the video document or videos from is a challenging task because flipping in the unconstrained environment and at any speed can cause difficulties specially in selecting a frame that contains the required image (OBI) for readability. In this paper, we stress on the problem of recognizing the readable and whole frame by selecting the frames with optical flow. Fundamentally, optical flow is used here for segmentation of the required area from the frame. Due to the unavailability of such a dataset, we have also created a dataset of video document containing book-flipping videos. Experiments are executed on the sample video documents recorded using mobile camera. The use of optical flow on video document gives proper track of motion as it detects the boundary of the document. Here in this case a normal Indian publication textbook is used. The need to develop this technique is to help various educational reforms in Indian universities related to paper evaluation. A light-weight model and accurate results makes this technique is cost-efficient and sustainable.