Temporal Visual Saliency Map and its Application

Autor: Zhi-Xiang Yang, 楊智翔
Rok vydání: 2012
Druh dokumentu: 學位論文 ; thesis
Popis: 100
This paper presents a novel method for detecting video saliency maps based on trajectory entropy. Our method utilizes long-term motions to alleviate short-term noises, and leads to spatially and temporally coherent saliency maps. Specifically, a set of collaborative trajectories, which are extracted by detecting and tracking salient points in the video frames, is used to represent a given video. We show that the object motions of various temporal lengths in the video can be faithfully captured by these trajectories, and the saliency maps can hence be established by taking the entropies of each trajectory with respect to velocity and acceleration into account. The experimental results demonstrate the effectiveness of our approach.
Databáze: Networked Digital Library of Theses & Dissertations