基于活动轮廓模型的非同质图像分割算法研究

发布时间：2018-05-16 05:07

本文选题：活动轮廓模型 + 水平集方法　；参考：《西北农林科技大学》2017年硕士论文

【摘要】：图像分割是指将图像按照不同特征划分成若干个具有相似性或一致性的区域,并提取出感兴趣目标的过程。它是图形识别和计算机视觉领域至关重要的预处理,对图像的特征提取和研究起到了关键性的作用。随着新的理论和方法的发展以及研究人员对原有模型的不断改进,近年来涌现出不少分割算法。但是由于图像种类繁多,应用具有复杂性,目前还没有一种能够普遍适用于所有图像的分割方法。基于偏微分方程的活动轮廓模型是一类新兴的图像处理模型,它是依托曲线演化理论发展起来的,与传统的图像分割算法相比具有更强的自适应性、更高的分割精度等优点。本文围绕基于活动轮廓模型的图像分割方法展开研究。首先对图像分割算法进行了归纳和综述,根据活动轮廓模型的分类,详细介绍了几种经典的活动轮廓模型;然后对活动轮廓模型中的数学基础知识进行了描述;最后对基于活动轮廓模型的非同质图像分割算法进行了重点研究,主要内容如下:1.提出一种新的结合全局和局部信息活动轮廓的非同质图像分割模型。提出的模型充分利用图像的区域信息,消除非同质污染的影响,并利用交互熵准则得到局部自适应权重,有效地提高了非同质图像分割精度。2.对基于信息几何的快速分割算法进行改进。原有的快速算法利用黎曼最速下降法改变曲线演化的方向,极大地提高了算法收敛速度。但是原有算法没有考虑图像的局部特征,无法分割灰度非同质图像。因此对原有算法进行改进,充分利用图像区域信息,提高算法分割的准确性。实验中,采用Jaccard相似系数-JS(Jaccard similarity)和Dice相似系数-DSC(Dice similarity coefficient)两个指标对不同模型的分割结果进行定量分析,结果表明:1.与经典的活动轮廓模型相比,提出的模型JS和DSC的值最接近1,迭代次数不多于50次。此外,当设置不同初始轮廓线对图像进行分割时,传统的活动轮廓模型产生了不同的分割结果,而提出的模型结果是相似的。说明提出的模型对灰度非同质图像进行分割具有较高的计算效率和准确率,且降低了对初始轮廓位置的敏感度。2.与传统的活动轮廓模型相比,改进的快速算法利用黎曼最速下降法对曲线进行演化,其迭代次数不多于10次,可见改进的算法具有较快的收敛速度。另外与原有快速分割算法相比,改进的算法结合了图像全局信息和局部特征,不仅保留了原有算法在弱边界图像分割中的优势,而且对灰度非同质图像也能得到较好地分割结果。
[Abstract]:Image segmentation refers to the process of dividing the image into several similar or consistent regions according to different features and extracting the objects of interest. It is a very important preprocessing in the field of graphics recognition and computer vision, and plays a key role in feature extraction and research of images. With the development of new theories and methods and the improvement of existing models, many segmentation algorithms have emerged in recent years. However, due to the variety of images and the complexity of applications, there is not a general segmentation method that can be applied to all images. The active contour model based on partial differential equation (PDE) is a new kind of image processing model. It is based on the curve evolution theory and has the advantages of stronger self-adaptability and higher segmentation accuracy than the traditional image segmentation algorithm. This paper focuses on the method of image segmentation based on active contour model. Firstly, the image segmentation algorithm is summarized, and several classical active contour models are introduced in detail according to the classification of active contour model, and then the mathematical basic knowledge of active contour model is described. Finally, the non-homogeneous image segmentation algorithm based on active contour model is studied, the main contents are as follows: 1. A new heterogeneous image segmentation model combining global and local information active contours is proposed. The proposed model makes full use of the regional information of the image to eliminate the influence of non-homogeneous pollution, and the local adaptive weight is obtained by using the cross-entropy criterion, which effectively improves the segmentation accuracy of non-homogeneous image. The fast segmentation algorithm based on information geometry is improved. The original fast algorithm uses Riemann's steepest descent method to change the direction of curve evolution and greatly improves the convergence speed of the algorithm. However, the original algorithm does not consider the local features of the image, and can not segment gray heterogeneous image. Therefore, the original algorithm is improved to make full use of the image region information to improve the accuracy of the algorithm segmentation. In the experiment, Jaccard similarity coefficient -Jaccard similarity coefficient and Dice similarity coefficient -DSCDice similarity coefficient are used to quantitatively analyze the segmentation results of different models. The results show that: 1. Compared with the classical active contour model, the JS and DSC values of the proposed model are closest to 1, and the number of iterations is not more than 50. In addition, when different initial contour lines are set to segment the image, the traditional active contour model produces different segmentation results, but the proposed model results are similar. It is shown that the proposed model has high computational efficiency and accuracy for gray-scale heterogeneous image segmentation, and reduces the sensitivity of the initial contour position. 2. Compared with the traditional active contour model, the improved fast algorithm uses Riemann's steepest descent method to evolve the curve, and the number of iterations is not more than 10 times, which shows that the improved algorithm has a faster convergence speed. In addition, compared with the original fast segmentation algorithm, the improved algorithm not only preserves the advantages of the original algorithm in weak boundary image segmentation, but also combines the global information and local features of the image. In addition, the segmentation results of heterogeneous grayscale images can also be obtained.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】