基于多小波变换的文本图像文种识别

摘要

Considering the characters that the texture performance of text images is the orientation of text stroke and the quasi-periodic of conduct periodic.a manuscripts identification method of text images based on multi-wavelet transform is proposed,which extracts the energy features of sub-band images to form the feature vector and uses LIBSVM to classify the scripts.Through the experiments done on text images containing ten different scripts,the effectiveness of the manuscripts identification and its robust to the descendents of image quality has been confirmed.%针对文本图像的纹理表现为文字笔画的方向性和以行为周期的准周期性的特点,采用了基于多小波变换的文本图像文种识别方法,提取多小波变换各子带系数的能量特征,构造特征矢量,并采用LIBSVM进行多文种的分类。通过对10种语言文字的文本图像进行实验,表明该算法在文种识别方面的有效性和对图像质量退化的鲁棒性。