一种基于角点与BP神经网络的文本检测及定位算法

Text detection and location algorithm based on angular-point and BP neural network

检测并提取视频图像中的文本信息对视频图像和内容的理解意义重大。以现有的文本检测算法为基础,提出一种基于角点与BP神经网络相结合的文本检测算法。该算法首先应用多尺度角点算法提取文本角点信息并初步定位文本行,接着提取文本特征,最后应用BP神经网络精确定位文本。实验结果表明,此算法与经典方法相比具有更高的正确率和鲁棒性,视频中文本的正确检测率达到90.3%。

The detection and extraction of text information in video image has important significance on understanding the video image and content. On the basis of the current text detection algorithm,an text detection algorithm based on the integration of angular-point and BP neural network is proposed. The multi-scale angular-point algorithm is used to extract the text angular-point information and primarily locate the text line,and then extract the text features. Finally the text is precisely located by BP neural network. The experimental results prove this method has higher accuracy and robustness in comparison with the classical method. The correct detection rate of text in the video can reach up to 90.3%.