一种基于多帧视频的文本图像质量增强方法

视频文本和视频内容高度相关,提供了理解视频内容的有用信息,然而文本往往位于复杂背景之中,从视频帧中定位到文本区域后,如果将其直接送入OCR软件,其识别效果较差。视频文本的时域信息提供了增强文本,消除背景的有用信息。因此,提出了一种利用视频文本的时域信息来消除背景,增强文本的方法。该方法首先利用边缘算子计算文本的轮廓特征,然后采用基于Hausdorff距离度量的匹配方法跟踪本文区域在相邻帧序列中的位置,利用多帧平均或帧间最小搜索法消去背景;其次,利用双线性插值技术调整文本尺寸,最终得到具有干净背景、合理分辨率的文本图像。不同测试视频序列的实验结果表明,该方法可以有效提高视频文本的OCR软件识别率。

Video Text Enhancement Using Multiple Frame Information

Text in video is a very compact and accurate clue for video indexing and summarization.But video texts are usually embedded in complex background,making it very difficultl for text separation from the background information.Hence the OCR accuracy was poor.This paper presents a multi-frames based technique to enhance video text image.After extracting a reference text block,we use Hausdorff distance based image matching technique to find and register the corresponding text block.Then the frames average or minimum pixel search method is applied to text blocks to obtain a new text block with a clean background.At last we apply a finite interpolation function to adjust the text block resolution.Experiments conducted on several video sequences show that our enhancement scheme can considerably improve the accuracy of OCR.