Concept‐Based and Event‐Based Video Search in Large Video Collections
Foteini Markatopoulou, Damianos Galanopoulos, Christos Tzelepis, Vasileios Mezaris, Ioannis Patras · 2019
This chapter analyses the literature and presents the research efforts for improving concept-based and event-based video search. It focuses on feature extraction using hand-crafted and deep convolutional neural networks (DCNN)-based descriptors, dimensionality reduction using accelerated generalised subclass discriminant analysis (AGSDA), cascades of hand-crafted and DCNN-based descriptors, multi-task learning to exploit model sharing, and stacking architectures to exploit concept relations for concept-based video search. The chapter also focuses on methods which exploit positive examples, when available, again using DCNN-based features and AGSDA for video event detection. It presents a pseudo-relevant feedback mechanism that relies on AGSDA. Video understanding is the overall problem that deals with automatically detecting what is depicted in a video sequence. Video concept detection is a multi-label classification problem that is typically treated as multiple independent binary classification problems, one per concept. The chapter also presents a method that builds a fully automatic zero-example event detection system.