中文摘要：

由于高斯核的参数敏感性,导致在传统谱分割中往往无法识别图像数据的部分特征信息,图像分割准确率下降。为提高图像分割准确率,构建了一种基于小波函数的核相似度量函数,利用小波函数的多分辨率特性,对图像进行相似度量;证明了该核函数是可容许核,分析了该函数作为相似度量函数的可行性及其多分辨率分析特性;最后,提出了基于该小波核相似度量函数的谱分割算法。在Berkeley数据库的实验结果显示,该算法能够提高图像分割结果的精度,识别图像中的细节信息。

英文摘要：

In classical spectral segmentation,some feature information is missed because of parameter sensitive of Gaussian kernel,which leads to a declined accuracy of image segmentation. To improve the accuracy of image segmentation,a wavelet-based kernel similarity function（ WKSF） is constructed,which takesthe advantage of the multi-resolution analysis of wavelet to measure similarity. The proposed kernel is proved to be an admissible kernel and suitable for being used as a kind of similarity measure function. The property of multi-resolution analysis of wavelet is also inherited. Then,based on the constructed WKSF,a new spectral segmentation method is proposed.Some experiments on Berkeley dataset are carried out,the results show that the accuracy of image segmentation is improved with this method and some detail features of image are recognized.