Massive Open Online Courses (MOOCs) have emerged as new learning phenomenon in the field of e-learning. Over recent years, it has attracted a significant number of learners as well as researchers. A wide range of researches is being carried out across multiple aspects of MOOCs. Video lectures are the most fundamental component in a MOOC. There are standard video styles that are normally used across several MOOC platforms, such as, talking head, demonstration, slides, animation etc. This paper presents an Image-Based classification approach of the video styles where a single video is split into multiple image frames, and then each frame is classified into one of the video style-category. Different classifier models built on top of each state-of-the-art deep neural architectures, including VGG16, InceptionV3, and ResNet50 are evaluated and the comparison of results is shown. Furthermore, the paper also discusses a numeric method to calculate the composition level of a single video style in multi-style filed videos based on the classification results.
Paper
Full text
Using Pre-trained Models As Feature Extractor To Classify Video Styles Used In MOOC Videos
Semantic Scholar · Computer Science · 2018
Abstract
Massive Open Online Courses (MOOCs) have emerged as new learning phenomenon in the field of e-learning. Over recent years, it has attracted a significant number of learners as well as researchers. A wide range of researches is being carried out across multiple aspects of MOOCs. Video lectures are the most fundamental component in a MOOC. There are standard video styles that are normally used across several MOOC platforms, such as, talking head, demonstration, slides, animation etc. This paper presents an Image-Based classification approach of the video styles where a single video is split into multiple image frames, and then each frame is classified into one of the video style-category. Different classifier models built on top of each state-of-the-art deep neural architectures, including VGG16, InceptionV3, and ResNet50 are evaluated and the comparison of results is shown. Furthermore, the paper also discusses a numeric method to calculate the composition level of a single video style in multi-style filed videos based on the classification results.