医学图像拼接

【作者】 徐波；

【导师】 汪卫；

【作者基本信息】 复旦大学 ， 计算机技术， 2011， 硕士

【摘要】 随着医学技术的发展,医学图像已经成为了医生不可或缺的参考资料。目前常用的X射线透视设备仅能提供局部范围的图像,为了更全面、更直观的了解患者病灶及其周围部位情况,需要将具有部分重合的医学图像进行无缝拼接得到超大视角的图像。我们开发了一个用于医学图像拼接的软件。医生可以根据需要,选择合适的拼接方法,将拍摄好的局部X光片图像拼接成一个完整的全景图像。软件使用微软Visual C++ 2008作为开发平台,中间调用了ITK、VTK、OpenCV等工具包。本文根据医学图像的特点,介绍了当前存在的医学图像拼接产品,总结了前人在医学图像拼接领域的研究成果,包括医学图像配准以及医学图像融合的方法,详细介绍了图像拼接的流程以及拼接过程中所采用的算法,重点介绍了医学图像配准算法以及医学图像融合算法。医学图像配准算法主要使用了基于像素灰度值的匹配方法以及基于特征点的匹配方法。在基于像素灰度值的匹配方法中,使用了归一化互相关模板匹配的方法；在基于特征点的匹配方法中,利用SURF算法找到相似特征点,进行匹配。医学图像融合算法主要使用了全局融合、局部融合以及图像平滑的方法。全局融合方法是通过计算两幅图像重合区域的像素灰度值的差异,将其中亮度较高的一幅图像的灰度值降低,使两幅图像的整体亮暗程度一致。局部融合方法主要是在两幅图像重合区域采用线性渐变的融合方法消除拼接缝。图像经过平滑处理后得到了最后拼接后的图像。在最后,我们介绍了本软件的设计以及最终的拼接效果。更多还原

【Abstract】 With the development of medical technology, medical images have become indispensable reference materials for doctors. At the present time, the widely used X-ray fluoroscopy equipment can only provide limited images of several parts. In order to gain a comprehensive and visualized knowledge of the focus as well as its surroundings of the patients, we need to make seamless stitching of medical images which overlap partly to get large-angled images.We developed a kind of software that can be used in medical image stitching. Doctors can choose appropriate ways of stitching according to their needs, and stitch the X-ray images of different parts into an integral panorama. This software uses Microsoft Visual C++2008 as its develop platform and also makes use of toolkits like ITK, VTK, OpenCV, etc.This essay introduces present medical image stitching products on the basis of the features of medical images. It generalizes former research achievements, which include ways of medical image registration and medical image fusion. This essay also presents procedures and algorithms of stitching. What are important in this essay are the algorithms of medical image registration and medical image fusion. Algorithm for medical image registration mainly uses matching methods based on the pixel value and characteristics. In the matching methods which based on the pixel value, ways of normalized cross-correlation template matching are used. While in the matching methods which based on characteristics, SURF is used to find out the similar characteristics to match. Algorithm of medical image fusion mainly uses methods of global fusion, local fusion and image smoothing. Global fusion calculates the differences of pixel value of the overlapped parts of two images and reduced the brighter and value-higher one so as to make consistent brightness of both the two images. Local fusion uses ways of linear gradient in the overlapped district to eliminate seams. Then the final image will be gained through smoothing. At last, the design and final results of this software are introduced.更多还原