摘要
相似病例对医生的诊断大有益处,快速、准确地从医学图像库中检索出相似病理图像是非常重要的。本文提出了一种结合图像纹理特征和形状特征的基于分级检索策略的医学图像检索新方法。该方法先计算图像的灰度共生矩阵,提取能量等纹理特征量,进行初检索,再通过Canny算子提取图像边缘,用投影法计算外边缘在垂直和水平方向上的投影,对初检索结果按形状特征进行二级检索。实验表明,本文方法有较高的查全率和查准率。
It is useful that a doctor consults analogical cases to diagnose for a patient. So it is very important for doctors to quickly and exactly search out the similar pathologica images from large numbers of images in clinic. This paper presents a new method for medical images retrieval based on graded retrieving policy. First, the four classical texture features, including texture energy, texture entropy, texture inertial moment and local texture calmness, are calculated based on the matrix of the image, and utilized to accomplish the elementary retrieval. Second, the method makes use of the projection vector of the object outline using Canny edge operator to breed out the false alarms of the first retrieval. Experiment results show that the method can achieve fairly high precision and recall.
出处
《北京生物医学工程》
2007年第1期36-39,64,共5页
Beijing Biomedical Engineering
关键词
分级检索
纹理特征
形状特征
共生矩阵
投影法
graded retrieval
texture feature
shape feature
matrix
projection method
