基于马尔科夫随机

发布时间：2018-04-02 01:23

本文选题：医疗图像分割　切入点：GRF　出处：《昆明理工大学》2017年硕士论文

【摘要】：医学图像分割在临床医学的研究和实际应用中具有重要的作用。借助于医疗图像分割,使得临床医生对疾病部位能够更直接、更清楚、更便捷地进行诊疗。由于医疗图像受到噪声污染、算法、设备、显示技术等因素的影响,使得医疗影像具有模糊性等。为了解决这些医疗影像处理中的图像分割精度、分割效率等问题,采用MRF理论、GRF理论、知识、模糊理论、D-S理论等多种融合算法进行三维图像分割,实现医疗图像分割的最佳效果。医疗图像分割是对图像的提取、分割、复原、重构、融合、识别等的过程。综合利用多次成像或多种成像设备的图形信息,补偿数据丢失、局部数据不精确以及不明确造成的局限性,使用MRF模型使图像的分割更精确。分析在脑部NMRI图片等分割中的具体应用效果。采用模糊聚类分割算法,使用灰度图信息和加入空间图像信息,无手动设置参数,实现了 NDFCM法比FCM法更好的图像分割效果,具有更强的抗噪性能和极高的医疗价值。利用最新FCM进行分类的MRF复原来获取更高精度的脑核组织图片,再用Markov和模糊聚类分别进行脑部图像分割并把这个作为D-S证据理论基本概率统计的条件,最后用D-S理论对脑部组织融合与分割,提高了医疗图形分割速度和质量,实现对脑部组织进行定性定量地分析研究。采用最新的图像模糊C均值(FCM)法,对医疗图形进行更精确分割。创新了高效的二维的间距程度测量技术,用相应的二维的直方图来定义邻域相关性。给出聚类中心v同一时间在像素值和邻域像素值二维的坐标上刷新的观点,实现了用邻域空间图像信息的二维FCM分割算法。采用D-S理论在融合两个或多个图像信息时,利用现有的的多种先验摸型进行融合,解决数据丢失、部分数据不精确、模糊或不明确因素造成的缺点。根据图像模糊聚类C-均值(FCM)理论和MRF理论,由条件概率服从高斯信号分布及先验概率,设置概率值M,采用D-S理论进行多种数据融合,利用D-S决策准则进行决策划分,提高了图像的分割精确。医疗图像分割方法主要以智能化、高精度、复杂度、抗噪能力、高速度、鲁棒能力、自适应能力等多个方面作为研究对象。传统的分割技术与最新的先进分割技术相融合是未来医疗图像分割技术的发展趋势。
[Abstract]:Medical image segmentation plays an important role in the research and practical application of clinical medicine. Because of the influence of noise pollution, algorithm, equipment, display technology and other factors, medical images have fuzziness. In order to solve the segmentation accuracy of these medical images, In order to realize the best effect of medical image segmentation, MRF theory, knowledge, fuzzy theory and D-S theory are used to realize the best effect of medical image segmentation. The process of reconstruction, fusion, recognition, etc., using the graphical information of multiple imaging or multiple imaging devices to compensate for the limitations caused by data loss, local data imprecision, and uncertainty. The MRF model is used to make the image segmentation more accurate. The concrete application effect in the brain NMRI image segmentation is analyzed. The fuzzy clustering segmentation algorithm is adopted, the gray image information and the spatial image information are added, and the parameters are not set manually. The image segmentation effect of NDFCM method is better than that of FCM method, and it has stronger anti-noise performance and higher medical value. The new FCM is used to restore the classified MRF to obtain more accurate brain tissue images. Then Markov and fuzzy clustering are used to segment the brain image, which is regarded as the basic probability and statistics condition of D-S evidence theory. Finally, the D-S theory is used to fuse and segment the brain tissue, which improves the speed and quality of medical image segmentation. To achieve qualitative and quantitative analysis of brain tissue. The latest image fuzzy C-means (FCM) method is used to segment medical images more accurately. The neighborhood correlation is defined by the corresponding two-dimensional histogram, and the view that the cluster center v refreshes at the same time on the coordinates of the pixel value and the neighborhood pixel value is given. A two-dimensional FCM segmentation algorithm using neighborhood spatial image information is implemented. When two or more image information is fused by D-S theory, existing prior models are used to solve the problem of data loss and partial data imprecision. According to the theory of image fuzzy clustering C-means FCM) and MRF theory, according to the conditional probability from Gao Si signal distribution and a priori probability, the probability value M is set, and the D-S theory is used to perform a variety of data fusion. The D-S decision criterion is used to divide the image to improve the accuracy of image segmentation. The medical image segmentation methods are mainly intelligent, high precision, complexity, anti-noise, high speed, robust ability, the main methods of medical image segmentation are intelligent, high precision, complexity, anti-noise ability, high speed, robust ability. The combination of traditional segmentation technology and the latest advanced segmentation technology is the development trend of medical image segmentation technology in the future.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R310;TP391.41

【参考文献】