摘要
针对乳腺病理图像分类,提出一种非相干字典学习及其稀疏表示算法.首先针对不同类别的图像,基于在线字典学习算法分别学习各类特定的子字典;其次利用紧框架建立一种非相干字典学习模型,通过交替投影优化字典的相干性、秩与紧框架性,从而有效地约束字典的格拉姆矩阵与参考格拉姆矩阵的距离,获得判别性更强的非相干字典;最后采用子空间旋转方法优化非相干字典的稀疏表示性能.利用乳腺癌数据集BreaKHis进行实验的结果证明,该算法所学习的非相干字典能平衡字典的判别性与稀疏表示性能,在良性肿瘤与恶性肿瘤图像分类上获得了86.0%的分类精度;在良性肿瘤图像中的腺病与纤维腺瘤的分类上获得92.5%的分类精度.
An incoherent dictionary learning and sparse representation algorithm is proposed for histopathological image classification in this paper.Class-specific sub-dictionaries are firstly learned from each class training samples by exploring online dictionary learning.Furthermore,a novel incoherent dictionary learning model is designed by introducing tight frame.This model can optimize the difference between Gram matrix and reference Gram matrix by alternating projection on coherence,rank and tight frame of dictionary.The high-quality discriminative incoherent dictionary is obtained.To obtain the preferable sparse representation performance,subspace rotation method is utilized to optimize the sparse representation performance of incoherent dictionary.Experimental results on BreaKHis dataset show that the learned incoherent dictionary can trade-off the discriminative ability and sparse representation.The proposed method achieves 86.0%of classification accuracy on benign and malignant tumors image,and 92.5%of classification accuracy on adenosis and fibroadenoma image.
作者
汤红忠
王翔
郭雪峰
刘婷
Tang Hongzhong;Wang Xiang;Guo Xuefeng;Liu Ting(College of Information Engineering,Xiangtan University,Xiangtan 411105;Key Laboratory of Intelligent Computing&Information Processing of Ministry of Education,Xiangtan University,Xiangtan 411105)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2019年第8期1368-1375,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61573299,61602397)
湖南省自然科学基金(2017JJ3315,2017JJ2251)
关键词
非相干字典学习
紧框架
稀疏表示
组织病理图像分类
incoherent dictionary learning
tight frame
sparse representation
histopathological image classification
