基于HOG特征和MLP分类器的印刷体维吾尔文识别方法

A Printed Uyghur Recognition Method Based on HOG Feature and MLP Classifier

针对印刷体维吾尔文的有效识别问题,提出了一种基于梯度方向直方图(HOG)特征和多层感知器(MLP)神经网络的印刷体维吾尔文识别方案。对维吾尔文图像进行预处理,获得去除噪声后的二值化图像。利用水平投影积分对文本执行行切分,利用垂直投影积分方法执行单词切分和字母切分,获得独立的字母。基于HOG方法提取字母的特征。通过训练好的MLP神经网络分类器,根据提取的HOG特征对字母进行识别。实验结果表明,提出的方法能够精确地从图像中识别出维吾尔文字母。

For the effective recognition issues of printed Uyghur, a printed Uighur recognition scheme based on histogram of gradient oriented （HOG） and multi-layer pereeptron （MLP） neural network is proposed. Firstly, the Uighur image is preprocessed to remove noise and obtain a binarized image. Then, the text is linearly segmented by the horizontal projection integral, and the vertical projection integral method is used for word segmentation and letter segmentation, so as to obtain some independent letters. After that, the character of letter is extracted based on the HOG method. Finally, a trained MLP neural network classifier is used to identify the letter according to the extracted HOG features. Experimental results show that the proposed method can accurately identify Uighur alphabets from image.