基于形状信息和结果反馈的多图谱图像分割方法

项目名称： 基于形状信息和结果反馈的多图谱图像分割方法

项目编号： No.61502002

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 汤振宇

作者单位： 安徽大学

项目金额： 20万元

中文摘要： 基于多图谱的图像分割方法（MAS）因其分割精度高和鲁棒性强在图像分割领域被广泛研究，其主要包含图谱选择，图像配准和标签融合三个部分，该三个部分依次顺序执行完成对目标图像的分割，即图像配准部分将图谱选择部分产生的图谱与目标图像进行配准，配准后的图谱通过标签融合得到目标图像的分割结果。目前大部分MAS的研究重点主要放在图谱选择和标签融合部分，缺乏针对MAS的图像配准方法的研究。另外传统MAS方法中包含的三个部分，由于缺乏结果反馈，任一部分存在误差都会造成后续部分误差增大，最终导致图像分割质量下降。因此本项目的研究重点主要包含两个方面:(1)基于MAS包含的信息特点提出适合MAS的图像配准算法；（2）通过加入结果反馈机制解决由于MAS流程中各部分误差导致的误差累计和误差放大的问题。项目拟通过以上两方面的改进，使得拟提出的方法较传统MAS方法达到更加精确的图像配准进而得到更高质量的图像分割。

中文关键词： 多图谱；主动轮廓模型；结果反馈；图像配准；图像分割

英文摘要： Due to the high segmentation accuracy and robustness, the Multi-Atlas based image Segmentation method (MAS) is currently a hot topic. It consists of three main components which are the atlas selection, the image registration, and the label fusion. These three main components execute one after another. The atlases coming from the atlas selection part are registered with the target image in the image registration part, and then the registered atlases are sent to the label fusion part where the final segmentation result of the target image is produced. Since a lot of generalized image registration methods are available, in MAS very few research focus on the image registration part. Moreover, as there is no feedback strategy applied in current MAS, when one of the three components has error, then it will lower the performance of subsequent components. In order to solve the problem mentioned above, in this project, a new image registration method for MAS will be proposed where information contained in MAS is utilized in the registration method to enhance the registration accuracy. Furthermore, a feedback strategy will be added into MAS to eliminate the problem caused by error in each component of MAS, which cannot be removed and is accumulated in the subsequent component.

英文关键词： Multi-atlas;Active shape model;Feedback;Image registration;Image segmentation