基于Laplace谱的图像分类

利用谱能够反映图像的结构特性,提出了一种运用Laplace谱进行图像分类的算法。首先对图像中的特征点构造Laplace矩阵,通过SVD分解得到该矩阵的特征值,再由协方差矩阵,将高维的Laplace特征值投影到低维的特征空间中,最后分别采用BP算法和SVM算法对图像进行分类。通过模拟实验和真实实验,表明不同类序列图像的结构不同,其Laplace谱也不同,而同类序列图像的结构相似,其Laplace谱也相近,因此,Laplace谱与图像的结构之间存在着直接联系,能够作为图像分类的特征样本。实验结果说明,利用Laplace谱表示的图像特征对图像进行分类,具有较高的识别率。

Image Classification Based on Laplacian Spectra

A new algorithm for image classification using Laplacian spectrum is presented,utilizing spectrum to reflect the structural characteristics of the images.Firstly,the eigenvalue of Laplacian matrix constructed on the feature points of images is obtained by using the method of singular value decomposition.Secondly,the high-dimensional Laplacian eigenvalue is projected onto the lowdimensional eigenspace by means of covariance matrix.Finally,image classification is performed by adopting back propagation algorithm and support vector machine algorithm.Experiments of synthetic data and real image demonstrate that different classes of the sequence images have different structure,and their Laplacian spectrum are also different,whereas similar classes of the sequence images have similar structure,and their Laplacian spectrum are similar,too.Therefore, there have directly relationship between Laplacian spectrum and the graph structure,and Laplacian spectrum enables to regard as feature samples of image classification.Experimental results indicate that it is of comparatively high recognition rate to use the image characteristics denoted by Laplacian spectrum for classifying images.