摘要
利用核映射及非局部均值降噪特性构造相似性度量,即信息能度量。通过对细胞表型图进行特征映射并结合信息能度量得到图像特征的形态差异目标函数,依据梯度上升优化获取最优度量矩阵,建立基于核方法的形态差异学习模型。该模型特点在于:不仅考虑了各类样本的相似性,同时在降低噪声影响形态差异学习的过程中充分利用了图像的高阶统计量和非线性特征。实验结果表明,该核化算法灵敏度更高,且具有较好的鲁棒性,能有效应用于临床诊断。
Firstly,this paper defined the similarity measure by kernel mapping and non-local means,named as informative energy metric（IEM）.Secondly,combined the feature mapping of cell images with the IEM and an objective function for morphological differences was computed.Thirdly,it obtained the optimal metric matrix by gradient ascent algorithm and constructed a morphological difference learning model based on kernel methods.The characteristics of this model were that it could measure the similarity between each sample pairs and mine the high order statistics and nonlinear features.Experimental results show that this method is more sensitive and robust,it can be used in disease diagnosis.
出处
《计算机应用研究》
CSCD
北大核心
2012年第4期1341-1344,共4页
Application Research of Computers
基金
江西理工大学博士启动基金资助项目
江西省教育厅青年自然科学基金资助项目
国家自然科学基金资助项目(61070137)
关键词
图像分类
核方法
降噪
形态差异
image classification
kernel methods
denoising
morphological differences
