基于LLE特征降维及改进SVM的肝脏图像识别

Liver Image Recognition Based on LLE Feature Dimensionality Reduction and Improved SVM

为了提高肝脏CT图像正常和异常的识别率,提出了一种基于LLE特征降维及改进SVM的肝脏图像识别方法。在对采集的CT图提取感兴趣区域的颜色特征、形状特征和纹理特征,利用标准差变换和极差转换把这些特征规格化到0与1之间后,采用LLE算法对特征数据进行降维融合,并使用改进的SVM对待识别图像进行分类。实验结果表明,对多类特征进行降维融合比仅用单类特征能更好地表达感兴趣区域的内容信息,LLE算法较其他流形学习算法表现出更强的鲁棒性,改进的混合核函数SVM较单一核函数SVM识别率要高。该方法可以为医生辅助诊断提供参考。

In order to improve the recognition rate of normal and abnormal images,a liver image recognition method based on LLE feature dimensionality reduction and improved SVM is proposed.The basic idea is to extract the color features,shape features and texture features of the interested region for the collected CT images,normalize these features to between 0 and 1 by using standard deviation transform and range transform,use LLE algorithm to reduce the dimension of the feature data,and then use improved SVM for classification and recognition.The experimental results show that the dimension reduction fusion of multi class features can better express the content information of the region of interest than only using single class features.LLE algorithm shows stronger robustness than other manifold learning algorithms,and the recognition rate of the improved hybrid kernel SVM is higher than that of single kernel SVM.The results of this paper can provide reference for doctors to assist diagnosis.