一种具有旋转鲁棒性的文本图像文种识别方法

A Robust Rotation-invariant Script Identification Method of Document Images

针对目前用于文本图像文种识别的纹理特征描述子对文字行倾斜缺乏不变性,采用可控金字塔变换提取文本图像的纹理特征,通过对特征空间元素重新排列,提出一种对文字行倾斜具有鲁棒性的文本图像文种识别方法。不同倾斜角度文本图像的文种识别结果表明,该算法具有较高的识别准确率并对文字行倾斜具有较强的鲁棒性。

Script identification is significant for attaining information from document images. Most algorithms on texturc feature extraction from document images for script identification are inadaptable to the skew of text line presently. For the skew of text line is inevitably, a new algorithm robust to the skew of text line is proposed. Steerable Pyramid transform is used on the document images and the energy statistical features of sub-bands is extracted. Through the realignment of features, the algorithm implements robustness to rotation. Libsvm is used as a classifier. The experiments are eonducted on image database containing ten scripts that are scanned from books or magazines. The test samples are rotated with different angles and the results confirm that the algorithm can identify scripts accurately and is robust to the skew of text line simuhaneously.