图像的扩散界面无监督聚类算法

图像的无监督聚类就是基于图像数据,在无任何先验信息的情况下将整个图像集合划分成若干子集的过程。由于图像的本征维度很高,在图像处理中会遇到“维数灾难”问题。针对图像无监督聚类的特点,提出了一种图像的扩散界面无监督聚类算法,将图像编码成高维观测空间中的点,再通过投影变换映射到低维特征空间,在低维特征空间中构建扩散界面无监督聚类模型,并在模型中引入维度约简算子,采用循环迭代算法优化扩散界面模型的能量函数。基于最优的扩散界面,将整个图像集合聚类成不同的子集。实验结果表明,扩散界面无监督聚类算法优于传统聚类算法中的K-means算法、DBSCAN算法和Spectral Clustering算法,能够更好地实现图像的无监督聚类,在相同条件下具有更高的准确度。

Diffuse Interface Based Unsupervised Images Clustering Algorithm

Unsupervised clustering of images aims to partition the whole image set into several subsets on the basis of image data itself,while without any priori information.As dimensionality of an image is usually very high,curse of dimensionality arises du-ring the image processing.Having analyzed the problem of images clustering,a novel unsupervised image clustering algorithm is proposed.The proposed algorithm is based on diffused interface model on graph.Images were encoded as the data points in high dimensional observing space,and then were projected into low dimensional feature space.Diffuse interface model based unsupervised clustering algorithm was constructed in feature space,and dimension reduction operator was introduced into the model.Loop iterative algorithm was employed to optimize the energy function of diffuse interface model.The optimized diffuse interface was adopted to cluster images into different subsets.Experimental results show that the proposed algorithm is superior to traditional K-means,DBSCAN and Spectral Clustering algorithm.It achieves better clustering results and lower error rates.