统一贝叶斯框架下的活动轮廓模型OCT心管图像序列分割

项目名称： 统一贝叶斯框架下的活动轮廓模型OCT心管图像序列分割

项目编号： No.61272237

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 董方敏

作者单位： 三峡大学

项目金额： 80万元

中文摘要： 针对光学相干层析(OCT)心管图像的强噪声特性、以及目标形变的多样性和背景的复杂性，本项目提出统一贝叶斯框架下活动轮廓模型图像序列分割方法：采用层次化设计，通过统一贝叶斯框架实现特征提取与活动轮廓模型两个模块的有机结合。该方法既可有效融合现有图像及图像序列所有可用特征，又简化活动轮廓模型。主要内容有：①分析OCT系统噪声形成机制并对噪声建模，构建有效去噪方法；②分析OCT心管图像的特点，特别是目标形变的多样性与背景的复杂性，研究在统一贝叶斯框架下融合多特征的活动轮廓模型图像分割；③研究心管图像序列的时间和空间相关性，据此构建融合时间或/和空间特征的活动轮廓模型图像序列分割方法；④研究自动化及快速分割方法。其成果旨在探索复杂环境下的活动轮廓模型图像分割，最终实现OCT心管图像序列自动、快速分割；同时为广泛的OCT图像处理提供一般模型和方法；对广域目标轮廓跟踪也具有很好的支撑作用。

中文关键词： 图像分割；图像去噪；活动轮廓模型；光学相干层析；心管图像

英文摘要： It is difficult to segment the optical coherence tomography (OCT) heart tube image because of the noise in the image, the diversity of the object and the complexity of the background. To tackle the problem, the image sequence precision segmentation approach based on the active contours model (SNAKE) in the generalization Bayesian framework is proposed. It is designed as hierarchy, where two modules of the system, featrue extraction and SNAKE, are linked with the the Bayesian framework. By doing so, not only all the features of the image and image sequence can been incorporated into the system, but the SNAKE is simplified and effectively. The main sub-projects are as follows: a) Investigate the principle of OCT and the forming mechanism of noise, and construct the model of noise and filter approach; b) Investigate the feature of the OCT image, especial the diversity of the object and the complexity of the background and propose the image segmentation approach based on the SNAKE in the generalization Bayesian framework; c) Investigate the spatial and temporal relationship of object in the OCT image sequence and proposed the image sequence segmentation approach based on the SNAKE in the generalization Bayesian framework with the spatial and temporal features; d) Propose the automatic and fast segmentation approach.

英文关键词： image segmentation；image denoise；active contours model；optical coherence tomography(OCT)；heart tube image