基于Contourlet域HMT-3S模型的激光主动成像图像分割

Laser active image segmentation based on Contourlet-domain hidden Markov trees-3S model

基于Contourlet系数分布统计特性,结合隐马尔科夫树(HMT)-3S模型和贝叶斯准则,提出了一种基于Contourlet域HMT-3S模型的图像分割算法(CHMT-3Sseg),并将其用于激光主动成像图像的分割。CHMT-3Sseg方法首先利用HMT-3S模型充分挖掘子带间的相关性,通过HMT-3S模型参数训练和似然值计算得到了可靠的初始分割。为了获得较好的区域一致性和边缘准确性,对初始分割结果进行了基于邻域背景的多尺度融合。对合成图像和激光主动成像图像的实验结果表明:CHMT-3Sseg不但在边缘和方向信息保持上有明显改进,而且错分概率显著降低,对真实图像得到了理想的分割效果。

Based on the statistics characteristics of Contourlet coefficients,a new image segmentation method combining hidden Markov model（HMT-3S） with multiscale Bayesian approaches was presented for the segmentation of laser active images.In this algorithm,the cross correlation across subbands were well exploited by HMT-3S,and reliable initial segmentation results were obtained by training the parameters of HMT-3S and computing the likelihood of each scale.In order to obtain more exact edges and better region information,multiscale fusion based on a contextual model was applied to initial segmentation results.Experimental results of the segmentation for both synthetic image and laser active image show that the method not only have better performance in edges and anisotropy information detection,but also have lower missed classed probability,and it can achieve satisfied segmentation results for real images.