基于语义概念的视频检索系统的设计与实现

Design and Implementation of Semantic Concept Based Video Retrieval System

设计并实现了一种基于语义概念的视频检索系统,该系统包括视频镜头分割与关键帧提取、语义概念检测和用户检索3个部分。系统采用镜头分割与关键帧提取对视频进行层次分割,并对关键帧图像提取有效的图像低层特征,再使用支持向量机(SVM)进行概念的检测,最后针对概念内容进行视频检索。在概念检测中,提出了一种基于验证平均准确率的线性加权方法对SVM的分类结果进行后融合。实验结果表明,该方法可以达到较高的检索准确率。

This paper introduces the design and implementation of a semantic concept based video retrieval system, which consists of shot boundary detection and key frame extraction subsystem, semantic concept detection subsystem and user retrieval subsystem. First, digital video is divided into hierarchical structure for retrieval. Then, efficient low level feature of key frames are extracted. Support Vector Machine is used to detect concepts in such key frames, and the video retrieval is based on those concepts. In the procedure of concept detection, we take a linearly weighted fusion method based on validation precision to improve the average precision. Experiments show that the Mean Average Precision of our system is as high as the best one of all submissions.