基于多尺度图像融合的新闻视频文字区域检测定位算法  

Automatic Text Detection in News Video Based on Multi-Resolution Image Fusion

作　　者：章慧[1] 赵丽娟[1] 

机构地区：[1]淮阴工学院计算机工程学院,江苏淮安223003

出　　处：《贵州大学学报（自然科学版）》2012年第6期86-90,共5页Journal of Guizhou University:Natural Sciences

基　　金：国家自然科学基金(No.60973113);淮安市工业科技支撑项目(HAG2010069)

摘　　要：针对新闻视频中的文字区域检测定位问题,研究了如何快速有效地检测新闻视频中的文字区域,用以实现自动视频检索。为克服视频中文字大小不一的难题,根据视频图像中文字区域特征有别于背景区域的特点,提出了一种基于多尺度图像融合的新闻视频文字区域检测定位算法。算法主要采用训练和学习两个步骤,首先对人为收集的含字符样本与不含字符样本进行小波特征和局部二值模式等特征提取,并完成SVM分类器训练,获取分类器;然后对测试视频帧进行多尺度的遍历检测,并融合检测结果,获取每帧的文字区域。实验结果表明,与前人提出的基于边缘检测的方法相比,算法具有明显优越性,在定位准确度上有较大提高,同时还能克服视频帧之间的快速变换,具有一定的实用意义。As to the problem of automatic text detection in news video, an efficient algorithm was proposed for text location and video searching. In order to overcome the challenge of the different size of text in news video frames, an algorithm which was based on multi-resolution The method includes two steps ： firstly, the wavelet feature image fusion and text block feature was presented. and LBP feature of positive samples and negative sam- ples were extracted which can be trained by support vector machine （SVM）. And then, the test video for text detection should be ergoticly detected by multi-resolution method. Finally, the result image of text detection can be gained by multi-resolution image fusion. The experimental results show that this method has the superiority of accuracy rating compared with the traditional method based on edge detection, so that the video frames are trans- formed quickly

关 键 词：小波特征提取 局部二值模式特征 文字定位 多尺度融合 

分 类 号：TP317.2[自动化与计算机技术—计算机软件与理论]