基于小波域隐马尔科夫模型的文本图像子带分割方法

A New Document Segmentation Based on Subbands by Wavelet-Domain Hidden Markov Tree Models

本文在已有文献的基础上 ,通过分析不同子带小波系数之间的相关性 ,提出了一类基于小波域HMT(HiddenMarkovTree)模型文本图像分割方法 .其基本思想是先在子带分类的基础上 ,综合考虑不同尺度上的分类 ,进行多尺度文本图像分割 ,最后根据后验像素信息对上述方法所得分割结果进行修正 ,得到优于已有文献的分割效果 ,而且在一定程度上减少了分割算法的计算量 .

We introduce a kind of new wavelet-domain HMT segmentation method,a finer to coarser HMTseg,which combine with the classification results of the three subbands for the 2-D wavelet transform.We demonstrate that the new method's performance is somewhat better than the raw segmentation in prevenient document.The advantage to our segmentation algorithms is that they can offer improved segmentation accuracy with smaller computational burden.Finally we introduce a new HMTseg method by updating the classification constrainedly,and then we get a better segmentation result for document image.