项目名称: 大目标场景下的电镜图像拼接方法研究
项目编号: No.61201050
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 电子学与信息系统
项目作者: 陈曦
作者单位: 中国科学院自动化研究所
项目金额: 24万元
中文摘要: 对生物脑切片或集成电路芯片等纳米级的大目标场景进行重建,需要使用扫描电镜采集大量的序列图像,再通过图像拼接的方式合成为一幅完整的大图。全局配准是图像拼接中消除累积误差的重要步骤,常利用图像之间的拓扑关系或采用最小化误差函数的方法。针对大目标场景下的电镜图像拼接,如果采用常见的全局配准方法,拼接结果的稳健程度会受到限制。因此,本项目拟对此展开研究,具有较强的科学意义和现实意义。研究主要包括二个方面的内容:(一)扫描电镜图像畸变模型及畸变校正方法的研究;(二)利用马尔可夫随机场刻画全局配准模型及其求解方法的研究。项目预期可以在完成电镜图像畸变校正的基础上,通过马尔可夫随机场模型表征图像的全局配准位置之间的相关性,并开发出相应的并行算法,完成大目标场景的重建工作。研究结果扩展了马尔可夫随机场理论在图像拼接领域中的应用,丰富和完善了图像拼接系统理论,解决了图像拼接应用层面的实际问题。
中文关键词: 图像配准;序列切片;扫描电镜;神经回路;
英文摘要: For the purpose of reconstruction of large area scenes at nanometer-level, such as brain tissue in neuroscience or integrated circuit chip analysis, we first grab large sets of sequential images with Scanning Electron Microscope (SEM), and then obtain the result using image mosaic method. Global alignment is an important step to eliminate accumulation error in image mosaic, which usually adopts topology method or minimizes some error function. During the process of global alignment in SEM image mosaic for large area scene, the mosaic result would not be acceptable if we did not adopt the specific method. Thus, we focus our project on SEM image mosaic for large area scene, which has strong scientific meaning and realistic significance. The study mainly includes two aspects. The first is about SEM image distortion modeling and correction of the distortion. The second is global alignment using Markov Random Field (MRF) and corresponding inference method. In order to mosaic huge amounts of SEM images, we first correct SEM image distortion, and then characterize the relevance of images position in global alignment with MRF, and finally infer the mosaic result using parallel algorithm. The project is expected to expand application fields of MRF, enrich and perfect the framework of image mosaic, and resolve problems en
英文关键词: Image registration;Serial sections;SEM;Neural circuit;