Computational Video: Can we just extend all image-based approaches to video by applying image analysis to all the frames?

Irfan Essa

Georgia Institute of Technology and Google Research, US

Abstract

In this session, we will dive into the deep dark abyss of video analysis. We will explore a series of image analysis methods that have been extended to video analysis by considering video as a simple collection of frames. We will discuss the validity of these approaches and see what are the strengths of these approaches and what are weaknesses. This will allow us to then dive deeper into the issues of how best to represent video and consider a series of applications. We will look into low-level approaches from video compression to video registration and alignment. We will also look at topics of video segmentation and tracking in video. Then we will explore issues of categorization in video, all the way from object recognition to action/activity recognition. As explore categorization in video, we will also explore the importance of language models.