基于Contourlet域树状系数的自组织神经网络图像分割

Image segmentation of self-organizing neural networks based on tree-type coefficients in Contourlet domain

为避免小波域隐马树模型分割算法中模型假设的不足,提出用SOM网络作为非参数概率密度函数估计器。用图像轮廓波变换域中的树状数据作为网络输入,以利用图像的几何特征来提高分割效果。由训练好的网络组可以得到待分割图像各个尺度下的条件概率密度函数值,应用最大似然分类准则得到相应尺度下的粗分割。通过多尺度粗分割结果的融合,得到像素级的分割结果。用合成纹理图像、航拍图像和SAR图像进行实验,并与小波域隐马树模型分割方法和基于SOM网络的多尺度贝叶斯分割方法进行比较。对合成纹理图像给出错分概率作为评价参数,实验结果表明所提算法分割效果更优。

To avoid improper model assumption in hidden Markov tree model segmentation method in wavelet domain, self-organizing feature map （SOM） neural networks are used as a nonparametric probability density function estimator. Tree type data in Contourlet domain of images are used as inputs of SOMs so as to utilize geometric features of images. Condition probability density function values at given scale for awaiting images to be segmentalized can be obtained by trained networks. The maximum likelihood classification criterion is used for raw segmentation of images. The segmentalized results at pixel level can be obtained by fusing the raw segmentation results. In experiments, syn thetic mosaic images, aerial images and SAR images are selected to evaluate the performance of the proposed method, and the segmentalized results are compared with the hidden Markov tree model method in wavelet domain and the multiscale Bayesian segmentation method based on SOMs. For synthetic mosaic texture images, the miss-classed probability is given as the evaluation parameter. The experiment results show the proposed method has better performance.