基于领域关系广义混合模型图像分割的研究

Generalized mixture model with spatial neighborhood relationship for image segmentation

针对高斯混合模型不能有效处理复杂噪声图像分割问题,提出了基于领域关系广义混合模型。在高斯混合模型基础上引入形状参数r提高混合模型对不同噪声适应能力,另外该方法结合图像中像素点邻域关系,融入像素点间的互动信息。与混合模型通常采用EM算法估计参数不同,该模型参数估计采用梯度方法,通过最小化负似然对数优化参数。实验结果表明,广义混合模型在处理高斯噪声,重尾噪声,混合噪声以及脉冲噪声图像分割问题都取得了很好的效果。

To solve the problem that Gaussian mixture model can not deal with image segmentation with complex noise, a gener- alized mixture model is proposed. By introducing shape parameter r into Gaussian mixture model, the ability of the model adap- ting to multiple noises is improved, besides, the spatial relationship between neighboring pixels is also taken into account. In or- der to estimate the model parameters from observations, instead of utilizing an expectation-maximization algorithm, the gra-dient method is employed to minimize a higher bound on the data negative log-likelihood. The experimental results show that the pro- posed generalized mixture model has achieved good results in dealing with image segmentation with impulse/Gaussian noise, heavy-tailed and even mixed noise.