基于隐马尔可夫模型边缘表示的多尺度边缘检测

Multi-Scale Edge Detection Based on Representation of Edge Characterized by Hidden Markov Model

基于增强图像特征空间的特性分析 ,提出一种新的图像边缘表示模型———隐马尔可夫模型 ,该模型把边缘作为一个可观测序列 ,认为它是人们根据感官知识 (不可观测的状态量 )所观察到的具有一定空间关系的像素集合 ;通过小波变换 ,得到多尺度增强图像 ;将包含在低分辨率图像边缘中的全局信息进行综合 ,作为高分辨率图像中隐马尔可夫边缘模型的训练数据 ,使用Viterbi算法完成高分辨率图像中边缘点的搜索 ,从而 ,使得多分辨率边缘检测对先验知识的依赖大大降低 ,同时降低了计算复杂度 .边缘的隐马尔可夫模型表示也为多分辨率数据融合提供了一种有效的工具 .实验表明该模型是合理的。

Based on the properties of an enhanced image, a novel approach to edge detection is developed by modeling the edges as hidden Markov models. The model views an edge as an observable data series, with orders in space domain. In order to reduce the dependency on the prior knowledge and the complexity of algorithms, an image is represented in the frame of the multi scale analysis by wavelet transformation. It leads to guiding search of the edge points in higher resolution by edge information in lower resolution. At the same time, the hidden Markov model of an edge provides an efficient computational tool for multi scale data fusion. The experiment results show the proposed approach is reasonable and feasible.