基于MG-HMT模型的正交有限脊波域图像分割

摘要

针对图像分割中的过分割问题,提出了一种基于图像混合高斯-隐Markov树(Mixture Gaussian-hidden markov tree,MG-HMT)模型的正交有限脊波分析的图像分割算法.正交有限脊波变换处理信息时具有检测信号线奇异的能力,在图像分割中为准确定位信息的边缘、轮廓提供了有力的支持.其次,对图像的小波系数建立了混合高斯-隐Markov树(MG-HMT)模型来描述其在不同尺度子带间的相关性,并利用小波系数自身的传递性和同层小波系数相关性进行补偿处理.仿真结果表明,采用本文算法实现的图像分割,有效地检测出图像信息的线奇异,从而减小了由于干扰在梯度图中造成虚假的局部极值而产生的过分割现象,准确地定位了图像的区域信息.%Against the over-segmentation problem, an image segmentation algorithm based on mixture Gaussian-hidden Markov tree (MG-HMT) model in orthonormal finite ridgelet domains is proposed. Orthonormal finite ridgelet transform is used to detect the signal line singularity and accurately located the edge and contour in the image segmentation; then a MG-HMT model is built to describe the subband correlationship of different scales by the wavelet coefficients. Moreover, the edge and contour are compensated with the wavelet coefficient self-transitivity and the correlationship of the same scale. Simulation results show that the proposed image segmentation algorithm effectively detects the image line singularity and reduces the oversegmentation phenomenon produced by the false local extremum of the interference in the gradient map. Finally, the image regions are accurately located.