Abstract: Feature extraction is a key problem for the mass CT image segmentation, a novel features extraction algorithm of CT image is proposed based on Non-Subsampled Contourlet Transform（NSCT） and Gray Level Co-occurrence Matrix（GLCM） in this paper. Firstly, CT image is multi-scale, multi direction decomposed by the NSCT, and the co-occurrence features of sub-images are extracted by GLCM, and then the redundant features are eliminated by the principal component analysis and feature vectors are composed, finally CT image is segmented by the support vector machine based on multi-feature vector space. The experimental results show that the proposed algorithm can extract features of CT image, and has improved the segmentation accuracy of CT images, can provide assisted information for the doctor diagnosis.

Key words: image segmentation, non-subsampled contourlet transform, gray level co-occurrence matrix, feature extracted, features fusion

摘要： 针对海量CT图像分割中特征提取的难题，提出一种非下采样轮廓变换（NSCT）和灰度共生矩阵（GLCM）相融合的CT图像特征提取算法。首先采用NSCT对CT图像进行多尺度、多方向分解，并采用GLCM提取子带图像的共生特征量，然后对共生特征量进行主成分分析，消除冗余特征量，构成多特征矢量，最后利用支持向量机完成多特征矢量空间的划分，实现CT图像分割。实验结果表明，NSCT-GLCM能够较好地提取CT图像特征，提高了CT图像分割准确率，可以为医生诊断提供辅助信息。

关键词: 图像分割, 非下采样轮廓变换, 灰度共生矩阵, 特征提取, 特征融合