基于主动轮廓模型的光谱图像分割算法研究

发布时间：2018-12-24 12:36

【摘要】：在图像处理与计算机视觉领域,图像分割是最基础最重要的问题之一。近年来,国内外学者对基于主动轮廓模型的图像分割算法研究比较积极和深入,但现有的主动轮廓分割算法还缺乏普适性,分割精度也有待提高,因此,还需进一步探索研究针对光谱图像进行处理的分割算法。基于此,本文针对几何主动轮廓模型中水平集方程的改进开展相应研究,并将其成功应用于二维单波段红外图像及三维多光谱图像的分割。本文主要提出并研究了两种新型的光谱图像主动轮廓分割模型:(1)自适应的基于多特征的红外图像分割模型。该模型引入自适应权重系数整合全局符号压力函数和基于多种局部特征信息的符号压力函数,以实现水平集方程的改进;局部项利用高斯核函数嵌入多种统计特征信息和纹理信息,能够更加全面地表征图像各类区域。对不同场景下的红外图像进行实验,并与传统的主动轮廓模型以及边缘检测算法进行对比,结果表明:该模型受背景噪声影响小,且能够实现对灰度不均匀、边界模糊和对比度低的红外图像的有效分割。(2)基于空间-光谱信息的多光谱图像分割模型。其核心在于构造一个新型基于空间—光谱信息的符号压力函数。一方面,通过计算比较轮廓内外主成分大小,作为判断轮廓演化方向的准则;另一方面,利用权重系数将距离测度和光谱形状测度综合,作为向相应方向演化时的演化力大小。对AOTF多光谱成像系统采集的多光谱图像进行实验,结果表明:相比于对单一谱段进行处理的传统主动轮廓模型,该模型充分利用丰富的光谱信息,分割精度更高;相比于经典多光谱图像非监督分类算法,该模型不受细节信息干扰,能得到更突出的多光谱图像目标轮廓。
[Abstract]:In the field of image processing and computer vision, image segmentation is one of the most basic and important problems. In recent years, scholars at home and abroad have been active and in-depth research on image segmentation algorithm based on active contour model, but the existing active contour segmentation algorithm is still lack of universality, and the segmentation accuracy needs to be improved. It is also necessary to further explore the segmentation algorithm for spectral image processing. Based on this, the improvement of the level set equation in the geometric active contour model is studied in this paper, and it is successfully applied to the segmentation of two dimensional single band infrared image and three dimensional multispectral image. In this paper, two novel active contour segmentation models for spectral images are proposed and studied: (1) an adaptive infrared image segmentation model based on multiple features is proposed. In order to improve the level set equation, the adaptive weight coefficient is introduced to integrate the global symbolic pressure function and the symbolic pressure function based on a variety of local characteristic information. Using Gao Si kernel function to embed a variety of statistical feature information and texture information, local terms can more comprehensively represent various regions of the image. The infrared images in different scenes are tested and compared with the traditional active contour model and edge detection algorithm. The results show that the model is less affected by background noise and can achieve uneven gray scale. Efficient segmentation of infrared images with blurry boundary and low contrast. (2) Multi-spectral image segmentation model based on space-spectral information. Its core is to construct a new symbolic pressure function based on spatial-spectral information. On the one hand, the magnitude of the principal components inside and outside the contour is calculated and compared as the criterion to judge the evolution direction of the contour; on the other hand, the distance measure and the spectral shape measure are synthesized by using the weight coefficient as the evolutionary force when the contour evolves in the corresponding direction. The experimental results of multispectral images collected by AOTF multispectral imaging system show that compared with the traditional active contour model which processes a single spectral segment, the model makes full use of abundant spectral information and has higher segmentation accuracy. Compared with the classical unsupervised multispectral image classification algorithm, the model can get more prominent multi-spectral image target contour without the interference of detail information.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】