An accurate lecture video segmentation method by using sift and adaptive threshold

Abstract

Much research has been done in the past for segmenting lecture videos by detecting slide transitions. However, they do not perform well on certain kinds of videos recorded under nonstationary settings: the changes of a camera position or focus during a lecture. Since such non-stationary settings greatly affect visual properties of slides, the existing approaches utilizing global features and a global threshold, often have trouble in computing similarities between slides. In this paper, we propose a highly accurate method for lecture video segmentation by using SIFT and an adaptive threshold. By using SIFT, we can reliably match two slides whose contents are the same but are visually different. We also propose an adaptive threshold selection algorithm that detects slide transitions accurately by considering characteristics of features. Through various experiments that use real lecture videos, we show that our method provides 30% improvement in the average -score over other existing methods.

Spatio-Temporal Residual Networks for Slide Transition Detection in Lecture Videos

“…Whereas, the frames appearance will change with the camera motion. Hyun et al [5] and Jeong et al [6] extracted SIFT features, which are invariant to image scaling and rotation, and match them between two adjacent frames. The slide transition is detected when the similarity based on SIFT similarity is smaller than a threshold.…”

Lecture Vdeo Indexing Using Boosted Margin Maximizing Neural Networks

“…As a result, many redundant segments and indexes will be created. Jeong et al proposed a lecture video indexing method using Scale Invariant Feature Transform (SIFT) features and the adaptive threshold [8]. In their work SIFT features are applied to measure slides with similar content, and an adaptive threshold selection algorithm is used to detect slide transitions.…”

Content Based Video Retrieval

“…And can calculate Frame transition quite accurately by using Adaptive threshold. [11] OCR was initially developed for high contrast data images, taken from metal and other surfaces with uneven roughness and reflectivity. Content-based gathering within video data requires textual metadata that has to be provided manually by the users or that has to be extracted by automated analysis.…”

“…After that the similarity between the features of the query video and the stored feature vectors is determined. That means that computing the similarity between two videos can be transformed into the problem of computing the similarity between two feature vectors [11]. This similarity measure is used to give a distance between the query video and a candidate match from the feature data database V. CONCLUSION This work present content based approach to retrieve data in format of text automatically over videos.…”