摘要
视频中的文本如果直接送入OCR软件,识别率较低,因为文本往往叠加在复杂背景中,所以需要先将文本从背景中分割出来。背景像素可能具有和文本像素相似的颜色,并且由于解压缩的影响,文本像素颜色分布可能具有渐变性,给分割带来一定的困难。针对这些问题,提出一种基于文本边缘和颜色特征的文本分割方法,该方法首先利用文本边缘的高频特性沿文本轮廓对图像的颜色分布进行采样;其次使用K-均值空间聚类方法从采样点集合得到图像分割的种子点和分割半径,从而分割文本图像得到不同的分割结果;最后,利用文本笔画的连通域特征挑选出正确的分割结果。实验表明,该方法较好的解决了视频文本和背景的分离问题,分割结果具有较高的OCR识别率。
Before video text images are input to OCR software, text should be separated from background, because video texts often are embedded in complex background. There are two problems: one is that the color of background pixels may be similar with the color of text pixels, the other one is that the color of text pixels has variance caused by video compression and decompression. To solve the two problems, a new text image segmentation algorithm was introduced based on text edge and color features. First, the sample pixels set was got according to high frequency edge information of text. Second, K-means clustering method was applied to get segmentation seed pixels and radius, then segment text image into several text candidate images. Last, false text candidate images were excluded according to connected component property of text strokes Experimental result shows that this method can separate text from background easily, and gets good OCR result.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2008年第23期6498-6501,共4页
Journal of System Simulation
基金
航空支撑科技基金(05E551010)
关键词
视频检索
文本检测
文本分割
K-均值聚类
video indexing
video text detection
text segmentation
K-means clustering