摘要
Segmenting a document image into text and picture areas is very important for compressing efficiently document images. This paper introduces an algorithm of multiscale image segmentation for document image compression,which uses of wavelet-domain hidden Markov tree model in order to directly calculate the parameter of model based on original image to be segmented,and to obtain multiscale classification and segmentation of image. The idea of the method is to combine several new technologies such as multilevel wavelet transform,multiscale decision,across-scale dependencies and joint probability density function. The paper describes in detail the concept of the dyadic block,the correspondency between wavelets and dyadic blocks based on quad-tree,the hidden Markov model and multiscale likelihood computation.
Segmenting a document image into text and picture areas is very important for compressing efficiently document images. This paper introduces an algorithm of multiscale image segmentation for document image compression, which uses of wavelet-domain hidden Markov tree model in order to directly calculate the parameter of model based on original image to be segmented, and to obtain multiscale classification and segmentation of image. The idea of the method is to combine several new technologies such as multilevel wavelet transform,multiscale decision, across-scale dependencies and joint probability density function. The paper describes in detail the concept of the dyadic block,the correspondency between wavelets and dyadic blocks based on quad-tree, the hidden Markov model and multiscale likelihood computation.
出处
《计算机科学》
CSCD
北大核心
2003年第2期168-171,共4页
Computer Science
基金
教育部优秀年轻教师基金(2000-1103)
关键词
文档图像
多尺度分割算法
图像处理
小波变换
概率模型
计算机
Image segmentation,Markov tree,Wavelet transform,Multiscale classification,Probability density function