新疆地方性肝包虫CT图像纹理特征的分类研究

Classification of Xinjiang Local Liver Hydatid CT Image based on Texture Feature

纹理特征是图像分析的重要线索,灰度共生矩阵法提取的纹理特征具有很好的鉴别和分类能力。本文运用灰度共生矩阵分析新疆地方性肝包虫CT图像并进行特征提取,对图像进行尺寸归一、去噪和增强的预处理,计算0°,45°,90°和135°方向的能量、熵、对比度、相关性和逆差矩的均值,构成特征向量,并进行统计分析,用最大类间距法获取图像分类的主要特征,同时使用判别分析法对特征的分类能力进行评价。实验结果表明,灰度共生矩阵法提取的特征在统计分析中存在差异,最大类间距计算获得的特征能提高图像分类的准确率,一定程度上有助于对肝包虫病CT图像进行分类和检索。

Texture feature is an important clue of image analysis.Texture features extracted by GLCM has a good ability to identify and classify.In this paper,for CT images we normalizing scale,removing the noise by median filter,enhancing by contrast limited adaptive histogram equalization,and then we use the gray level co-occurrence matrix（GLCM） to calculation the mean of angular second moment,entropy,inertia moment,correlation and inverse difference moment in 0 °,45 °,90 ° and 135 ° directions to constitute the texture feature vector.To get main features of the image classification,we use statistical and maximum classification distance analyze the feature vector,and then evaluate the feature＇s classification ability by discriminant analysis.The result show that feature extraction by gray level co-occurrence matrix is significant in statistical analysis,features calculated by maximum classification distance can improve the accuracy of image classification and to some extent contribute to the CT images of liver hydatid disease classification and retrieval.