活动轮廓模型综述

基于活动轮廓模型的目标分割、物体跟踪方法是近十几年来图像和视频领域研究的热点,它可以将待处理问题的先验知识与各种图像处理算法有效地融合在一起,比以往的计算机视觉理论有更强的实用性.本文结合图像分割方法从指导思想和所用的数学方法两方面对活动轮廓模型特别是几何活动轮廓模型中基于水平集方法的C-V方法做了一定综述.     

An Active Contour Algorithm for Continuous-Time Cellular Neural Networks

A CNN-based algorithm for image segmentation by active contours is proposed here. The algorithm is based on an iterative process of expansion of the contour and its subsequent thinning guided by external and internal energy. The proposed strategy allows for a high level of control over contour evolution making their topologic transformations easier. Therefore processing of multiple contours for segmenting several objects can be carried out simultaneously.     

基于EMD的快速活动轮廓图像分割算法

C-V模型等传统基于区域的几何活动轮廓模型仅将灰度同质作为区域相似性的测度,致使其在分割噪声分布大、灰度复杂的自然及遥感图像时难有较好的分割结果。为此,该文提出基于Earth Movers Distance (EMD)的快速活动轮廓图像分割算法。首先,给出了基于EMD的区域相似性测度,并引入到模型的定义;接着,提出了基于过分割的规则化和快速曲线演化方法,很好地克服了传统模型的冗余轮廓、计算复杂等问题。对合成图像和遥感图像的实验结果证明了算法的有效性。     

基于局部图划分的多相活动轮廓图像分割模型

几何活动轮廓模型是图像分割领域的强有力工具。最近,一种基于成对相似性的图划分活动轮廓(GPAC)模型被提出,并有效应用于均质图像分割。但是,该模型的连接权函数仅与图像光谱相关,使得模型在低对比度模糊图像的应用存在较大局限,同时,成对相似性的计算量大,模型的数值实现效率不甚理想。针对这些问题,该文引入测地核函数定义连接权函数,结合多相水平集,提出了基于局部图划分的多相活动轮廓图像分割模型。自然图像的实验结果证明了该模型的有效性。     

一种具有向心力的新型Snake算法

动态轮廓模型（Snake算法）是一种较好的目标轮廓检测方法。但是传统的Snake算法在对其能量函数进行优化时,不能检测多目标、血管内轮廓和凹形轮廓。文章首先对图像进行自动Snake初始化,然后在Snake 的成长过程中加入向心力因子。实验结果表明,新的算法能够检测多目标、血管内轮廓和凹形轮廓,优于传统的Snake算法。     

引入判别准则的主动轮廓分割模型

基于区域信息的主动轮廓模型应用在图像分割中,难以使初始轮廓线的鲁棒性和分割强度异质图像的能力实现有效统一.针对这一缺陷,根据Fisher判别准则,在基于全局区域信息和局部区域信息的主动轮廓模型的基础上,对局部区域信息项进行了变换以及引入了判别准则,得到了一种改进的基于区域信息的主动轮廓模型.改进的模型不仅增强了初始位置的鲁棒性,而且可以有效处理强度异质图像,通过实验检验了该模型的性能.     

利用局部离散度活动轮廓模型的强度非均匀图像分割

强度非均匀现象在真实图像中普遍存在,采用常规基于强度的分割算法会导致严重的误分割。针对强度非均匀图像分割,提出了基于局部离散度的活动轮廓模型分割算法。首先定义基于类内类间距离的离散度,然后利用核函数提取局部区域信息,同时加入边缘指示函数加权的轮廓线长度项能量,建立基于局部离散度的活动轮廓模型。最后引入水平集函数惩罚项,避免水平集方法在演化求解时需要不断初始化的问题。合成图像和真实图像实验结果证明本文算法性能稳定,适应于强度非均匀图像的分割。      

基于活动轮廓模型的左心室MR图像分割

本文提出一种基于局部与全局特征的活动轮廓模型左心室MR图像分割算法.该算法融合了图像局部信息和全局信息.局部信息包含了图像局部均值和方差信息,来克服图像灰度不均匀的影响.全局信息则较好地提高模型处理图像弱边界的能力,并防止模型陷入局部最优,实验结果表明,改进算法分割出较为精确的心脏左心室MR图像.     

一种新的SAR图像局部拟合活动轮廓模型

本文基于I-divergence-TV(IDT)去噪模型和变分水平集方法提出一个新的局部拟合活动轮廓模型。该模型能有效解决SAR图像的分割问题,并且不需要周期性地更新水平集函数。通过对合成图像和实测Envisat SAR图像的分割实验,提出的模型与局部拟合RSF模型相比具有更准确的边缘定位能力和噪声抑制能力。     

Adaptive Energy Weight Based Active Contour Model for Robust Medical Image Segmentation

The performance of the traditional active contour model is subject to the energy weight parameters and initial level set functions, which significantly affect the accuracy of segmentation results. This paper proposes a new robust active contour model to eliminate the above limitations for both 2D single-object and multi-object medical image segmentation. The optimal values of energy weight parameters are defined with adaptive energy weight functions to adjust the contribution of each external energy term dynamically. Thus, the energy functional will not be controlled by the large external energy terms, otherwise boundaries leakage would occur. The initial level set functions are optimized with the coarse results obtained by fuzzy C-means clustering method. The evolution of level set functions starts from the locations near the true boundaries. Therefore, the deformable curves could converge to the true boundaries robustly. The proposed algorithm is verified using both synthetic images and medical images from different modalities. The experimental results demonstrate that the proposed algorithm could realize accurate and robust segmentation for medical images even in the presence of noises and weak boundaries.     

A New Active Contour Model Based on Distance-Weighted Potential Field

Snakes, or active contours, have been widely used in various image processing applications. Typical problems of snakes, including limited capture range, poor convergence to concavities, noise sensitivity, and initialization sensitivity, have limited their applications. For solving these problems, we propose a new potential for the active contour model. In this proposed potential field, each location’s potential is computed by integrating the feature information from all the pixels in the image with the distances as the weights. The external forces are computed as the gradients of this proposed potential field, and the computed external forces are static and have global capture range. Experiments and also the comparisons with the snake using gradient vector flow (GVF) as external forces are conducted to examine the performances of this proposed snake. The results show that the proposed snake has a large capture range and an excellent convergence to boundary concavities, and also the proposed snake is more robust to noise, more time efficient, and less sensitive to the initialization, compared with the GVF snake.     

融合先验形状知识的主动轮廓模型研究

融合先验形状知识的主动轮廓模型（HACM）能够解决目标存在遮挡、局部形变、相似变换等实际问题,广泛应用于图像分割、轮廓提取等领域。本文以主动轮廓模型（ACM）为基础,从先验形状知识的提取以及知识的融合方法两个方面,总结近年来ACM在融合先验形状知识的目标轮廓提取领域的一些最新研究成果。首先研究基于区域不变矩描述子、样条函数以及水平集的形状知识提取方法,用于提取刚体性、非刚体性目标样本集所包含的轮廓信息,该信息具有一定的泛化能力,即针对样本集以外的目标轮廓具备一定的表达能力。然后以单目标轮廓提取问题为研究对象,从距离函数的角度分析各种先验形状知识与原始轮廓模型的全局性融合方法,并介绍了一种基于标识函数的局部融合方法。局部融合方法可以解决单知识、多目标情况下的轮廓提取问题。最后对文中所介绍模型进行总结分析,对该领域的研究方向进行了展望。      

基于视网膜血管模型的图像分割与血管提取

本文采用非线性图像处理方法,对眼底图像进行去噪增强和分析,提取血管中心线,以便于对视网膜血管网络进行定量描述和定量分析。实验表明该方法能好地从非荧光眼底图像中提取出膜血管网络。     

一种基于活动轮廓模型的PET-CT肺肿瘤分割方法

针对PET-CT肺肿瘤分割中存在的没有充分将医生临床经验融入到算法设计的问题,该文利用PET高斯分布先验,结合区域可伸缩拟合(RSF)模型和最大似然比分类(MLC)准则,提出一种基于变分水平集的混合活动轮廓模型RSF_ML。进一步,借鉴人工勾画肺肿瘤过程中融合图像的重要价值,提出了基于RSF_ML的PET-CT肺肿瘤融合图像分割方法。实验表明,所提出方法较好地实现了有代表性的非小细胞肺肿瘤(Non-Small Cell Lung Cancer, NSCLC)的精确分割,主客观结果优于对比方法,可为临床提供有效的计算机辅助分割结果。     

Novel Active Contour Model for Image Segmentation Based on Local Fuzzy Gaussian Distribution Fitting

A novel active contour model is proposed, which incorporates local information distributions in a fuzzy energy function to effectively deal with the intensity inhomogeneity. Moreover, the proposed model is convex with respect to the variable which is used for extracting the contour. This makes the model independent on the initial condition and suitable for an automatic segmentation. Furthermore, the energy function is minimized in a computationally efficient way by calculating the fuzzy energy alterations directly. Experiments are carried out to prove the performance of the proposed model over some existing methods. The obtained results confirm the efficiency of the method.     

基于类间方差参数活动轮廓模型图像分割

提出了基于类间方差参数活动轮廓模型图像分割法.该方法将气球力参数活动轮廓模型中的恒定气球力替换为包含区域信息的变力,最大化目标和背景两区域类间方差,引导轮廓曲线进化.实验结果表明:对于初始轮廓位置不论是处于目标区域内部,或者是处于背景区域内部,还是与目标和背景区域相交,该模型都能获得正确分割结果.     

一种非参数估计的活动围道图像分割方法

该文采用Parzen窗非参数估计的活动围道方法进行图像分割,不必假定图像区域符合特定的概率分布,大大增加了该算法的应用范围。通过动态调整Parzen窗口参数,更好地利用了图像灰度区域一致的信息,使之更好地适应图像分割的任务。通过合成图像和真实图像的实验证明此方法具有很好的合理性,具有很好的分割精度和抗噪性能。     

Segmentation of Multi-Channel Image with Markov Random Field Based Active Contour Model

Segmentation is an important research area in image processing and computer vision. The essential purpose of research work is to achieve two goals: (i) partition the image into homogeneous regions based on certain properties, and (ii) accurately track the boundary for each region. In this study, we will present a novel framework that is designed to fulfill these requirements. Distinguished from most existing approaches, our method consists of three steps in the segmentation processes: global region segmentation, control points searching and object boundary tracking. In step one, we apply Markov Random Field (MRF) modeling to multi-channel images and propose a robust energy minimization approach to solve the multi-dimensional Markov Random Field. In step two, control points are found along the target region boundary by using a maximum reliability criterion and deployed to automatically initialize a Minimum Path Approach (MPA). Finally, the active contour evolves to the optimal solution in the fine-tuning process. In this study, we have applied this framework to color images and multi-contrast weighting magnetic resonance image data. The experimental results show encouraging performance. Moreover, the proposed approach also has the potential to deal with topology changing and composite object problems in boundary tracking.     

Laser Doppler Technique for Absolute Measurement of Blood Speed in Retinal Vessels

We describe an electrooptical laser Doppler system and technique of data analysis that provides absolute measurements of the speed of red blood cells flowing at discrete, selectable sites in the retinal vasculature. We present in vitro test measurements of the instrumentation as well as an example of an in vivo measurement from a patient with retinal vascular disease. We also present experimental data leading to the derivation of the relationship between the blood speeds measured in retinal arteries during the minimum diastolic and maximum systolic phases of the cardiac cycle and the time-averaged blood speed. Mean blood flow rate is calculated using the time-averaged speed and the cross-sectional area of the vessel at the measurement site. We discuss the criteria for selection of the measurement sites and assess the reproducibility of the measurements. We conclude that measurements on retinal arteries are less susceptible to experimental artifacts and provide more information than do measurements on retinal veins. The system is currently being used clinically in studies of retinal circulatory alterations in patients with diabetic retinopathy, arterial occlusive disease, retinal detachment, and carotid artery disease.     

一种基于参数活动轮廓模型的高能闪光照相图像分割算法

针对高能闪光照相系统成像质量较差的特点,提出了一种基于参数活动轮廓模型(Snake模型)的闪光照相图像分割算法.该算法在传统高斯力Snake模型中引入包含图像区域信息的变力,以目标和背景两区域具有最小方差为准则,构建兼顾边缘和区域信息的外部能量函数.数值实验结果表明,该算法对初始轮廓位置不敏感,较好地解决了客体凹陷区域分割问题,能够实现对含噪声的弱边界闪光图像的自动分割.