文本图像信息的提取与识别

文本是计算机视觉的许多应用中的一项重要特征，图像中的文本往往包含着比较丰富的信息，将文本图像信息里的文字进行提取和识别，对于图像内容的分析、理解、信息检索等方面具有重要的意义。文本图像的识别分为预处理，文字的切分，细化，特征选择与提取，最后对候选文字进行识别。在文字的切分方面提出了一种改进的投影算法，该算法能在很大程度上提高文字切分的准确度，采用基于数学形态学算法对文字进行细化处理，并在特征选择方面引用了多级分类的算法。

Feature Extraction and Recognition of Information in Document Image

Document is an important feature in many computer vision applications. The document image tend to contain more abundant information. Feature extraction and recognition of characters in document image have the vital significance in image content analysis and un- derstanding, even in information retrieval. The recognition of document image includes the following steps： characters preprocessing, seg- mentation, thinning, feature selection and extraction, finally the recognition of candidate words. An improved projection algorithm is pro- posed. This algorithm can greatly improve the segmentation accuracy. Mathematical morphology proposed is used to characters thinning, and a multi-stage classification algorithm is introduced in feature selection.