基于字符基元视觉短语的图像关键字识别

摘要

提出了一种利用字符基元视觉短语进行图像关键字识别的方法.该方法通过提取图像关键字的最大稳定极值区域,并进行归一化后得到字符基元.由于通常情况下每个关键字由若干字符基元构成,因此通过采用利用邻接的字符基元构造的视觉短语来提高图像关键字特征描述的可区分性；由于不同的字符基元组合结构可能构成不同的图像关键字,因此基于字符基元相邻关系判断短语几何结构的相似性.此方法不需要对图像进行二值化、布局分析和文本区域定位等预处理操作,具有更好的灵活性和鲁棒性.实验结果表明,此方法对于不同语言的图像关键字识别都具有较高的准确性.%This paper proposes a new approach for image keyword spotting using visual phrases of character primitives.The approach firstly extracts maximally stable extremal regions from a given image,and then normalizes them into character primitives.Considering that each image keyword is composed of several primitives,it adopts the visual phrases constructed by using the adjacent character primitives to improve the description of image keyword spotting.For different keywords may share the same character primitives but with the different geometric structure,it measures the similarity of geometric structures of phrases based on the adjacent relationship of character primitives.This method does not require the processes of image binarization,layout analysis and text area localization.And it is more flexibly and robust.The experimental results show that this method is with the high accuracy in spotting of different languages' image keywords.