EM算法在纹理织物图像分割中的应用

提出了一种基于多特征值高斯混合模型（Gaussian Mixture Model）期望最大化（Expectation Maximization,EM）聚类的图像分割算法.该算法采用YCbCr彩色空间提取每个像素点的颜色特征,选择像素点邻近的一个方块计算每个像素点的纹理特征,然后采用基于高斯混合模型的EM算法对图像每个像素进行聚类,根据聚类结果进行区域合并得到纹理织物图像分割的最终结果.通过和其他分割算法进行对比,此算法具有较好的分割效果.

The application of EM algorithm in texture textile image segmentation

For the texture textile images,a segmentation algorithm of Expectation Maximization clustering based on Gaussian Mixture Model is presented.The algorithm firstly extracts color features of each pixel by adopting YCbCrcolor space and computes texture feature of each pixel by choosing the neighboring square patch around the pixel.And then each pixel of image is clustered by using the algorithm of Expectation Maximization based on Gaussian Mixture Model.At the end,according to the clustering of image,the last result of segmentation is got by merging regions.Compared with other algorithms,the new algorithm has better effect on the texture textile image segmentation.