基于文字混合特征的视频文本定位研究  

Research of video text location based on multiple text character

作　　者：赵超[1,2] 方向忠[1,2] 

机构地区：[1]上海交通大学图像通信与信息处理研究所,上海200240 [2]上海市数字媒体处理与传输重点实验室,上海200240

出　　处：《微计算机信息》2009年第18期161-162,176,共3页Control & Automation

摘　　要：针对视频中文本信息在视频序列和视频索引中的重要性,本文提出了一种基于文字混合特征的文本定位算法。该算法首先对视频序列中每隔25帧的单帧图像进行边缘检测和投影处理来提取文本块,然后用支持向量基进行筛选,排除非文本块的干扰,最后利用视频序列中相邻帧之间的相关性来搜索剩余帧中的文本块。本文的算法在提高检测速度的同时保证了较高的检测准确度。Since the text information is important in the video stream and video indexing, this paper provides an algorithm based on muhiple character of the text to locate text block in video stream. This algorithm first extracts every 25 frames in the video stream, then uses text edge information, projection technology and support vector mechanic to check text block and exclude non-text block. Finally this algorithm furth, er checks the text block in the remaining video frames based on the checked frames. This algorithm greatly increases the check speed and has a high text block checked rate at the same time.

关 键 词：文本块 支持向量基 视频序列 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]