基于多颜色空间与累计直方图的视频场景分类.pdfVIP

第十五届全国图像图形学学术会议 基于多颜色空间和累计直方图的视频场景分类 刘林 ， 李金屏 ， 王真 （济南大学信息科学与工程学院，山东济南 250022） 摘 要：视频分析中的一个热点和难点是研究视频的语义，例如场景分类、视频检索、移动目标行为分析等。其 中场景分类是视频语义研究中的一个基础问题。在对视频场景进行分割的基础上，通过分析场景多个颜色分量的 统计直方图，发现大多数典型场景总有一个或者多个颜色分量的累计直方图具有独特性，即多个累计直方图的形 状是与特定场景相关的。提取高斯拟合后累计直方图各峰的峰高、峰位和峰宽等特征，利用最小二乘法，提出了 一种基于多颜色空间和累计直方图的视频场景分类方法，并以常见的五种视频场景为例进行了初步应用。实验表 明，该方法具有较好的分类效果。 关键词 ：视频场景分类；多颜色空间；累计直方图；高斯拟合；最小二乘法 Video Scene Classification Based on Multiple Color Spaces and Cumulative Histogram LIU Lin, LI Jin-ping, WANG Zhen （School of Information Science and Engineering ，University of Jinan，Shandong, 250022) Abstract: Research on video semantic is a hot and difficult spot in video analysis, such as scene classification, video retrieval, moving target’s behavior analysis and so on. Therein, scene classification is a fundamental problem. Uniqueness of most typical scene with one or more color components cumulative histogram was found by analyzing the statistics histogram of several color components, namely many cumulative histograms are related to particular scene. Using least square method, a novel video scene classification algorithm based on multiple color spaces and cumulative histogram method is proposed via extracting the height of peak, the position of peak and the width of peak after Gaussian fitting. Preliminary application has been demonstrated with commonly used five video scenes. Empirical results show that the proposed method yields good classification effect. Keywords: video scene classification; multiple color spaces; cumulative histogram;Gaussian fitting;least squares method 类、视频检索和移动目标行为分析等。对其研究具 0 引 言