摘要
目的提出一种基于特征检测、特征匹配、空间坐标变换和图像混合等4个步骤的图像拼接技术,用于脊椎和长骨等X线图像的自动且无需标尺的全景成像。方法通过Haar小波分解和Sobel算子检测图像特征点,使用归一化互相关和一致性条件来实现特征点之间的匹配,根据匹配的特征点信息估计仿射变换模型参数,最后将重叠区域的像素通过加权求和实现合成。结果经开放式X线图像数据库和同济大学附属同济医院临床骨科X光图像数据的实验,结果显示对重叠面积在50%以上的X光图像,系统能够自动进行拼接。主观问卷调查中86.7%的被访者接收系统的拼接结果。结论该方法能够有效应用于连续X光片全景成像。
Objective To provide a method of making medical panorama from X-ray images, comprised of feature detection, feature registration, spatial transformation model and image blending, which can be used for X-ray images of vertebra and long bone. Method Feature points were detected from images by Haar wavelet decomposition and Sobel operation, and then cross-correlation and condition for consistency detection were used to match those feature points. After that, spatial transformation was implemented by using affine matrix, and then image blending was compounded with the overlapped area of the images. Result This method was tested by the data from the Open Source X-Ray Gallery（COX） and clinical orthopedic x-ray images in Shanghai TongJi hospital. The present system succeeded in mosaicing X-ray images with 50% overlap. Among the interviewees who had finished the questionaire, and 86.7% accepted the mosaicing result. Conclusion This method is effective for building long bone and vertebra panorama from fluoroscopic X-ray images.
出处
《航天医学与医学工程》
CAS
CSCD
北大核心
2005年第5期360-364,共5页
Space Medicine & Medical Engineering
基金
上海市重点学科建设项目资助(T0102)
关键词
医学图像
图像拼接
长骨
脊椎
全景成像
medical images
image mosaicing
long bone
vertebra
panorama
