基于小波域多状态隐马尔科夫树模型的自适应文本图像分割算法

Context-Adapted Document Segmentation Based on Multi-State Hidden Markov Tree Models in the Wavelet Domain

本文针对文本图像首先提出了一种基于小波域多状态隐马尔科夫树模型的自适应文本图像分割算法(Context-Adapted wavelet-domain Hidden Markov Tree,简称为CAHMT),该算法具有较高的分割质量和较低的计算复杂度.其次,为了进一步提高CAHMT算法分割的效果,将该算法与微分算子、尺度系数相结合提出了两种新的文本图像分割算法.最后通过实例阐明了这些算法的有效性.

This paper presents a new document segmentation algorithm, called context-adapted wavelet-domain hidden Markov tree （CAHMT） model, which extends a recently emerged wavelet-domain hidden Markov tree （HMT） model[1]. The proposed CAHMT can achieve more accurate quality in document segmentation with low computation complexity. In addition to further improving the segmenting performance, we combine differential operator and the lowest frequency subband （called scale coefficients in wavelet transform） with CAHMT and produce much better visually segmentation quality than the HMT does.