【作者】 张鹏飞；

【导师】 尉朝闻；

【作者基本信息】 西安科技大学 ， 计算机应用技术， 2012， 硕士

【摘要】 图像拼接的目的是将有重叠区域的若干张图像拼接成一张高分辨率全景图像，其是数字图像处理领域的一个研究热点，在虚拟现实、医学成像、遥感技术等诸多领域有着广泛的应用。本文主要研究内容和成果如下：1、总结了图像拼接的一般流程和相关算法的优缺点，介绍了图像采集、噪声抑制以及图像变换模型的相关知识，并对中值滤波进行实验仿真，仿真结果表明，中值滤波去噪效果良好。2、分析对比了常见的图像配准方法的特点、性能和实现步骤，重点研究了基于特征点的图像配准方法，对三种角点检测算子Moravec、SUSAN和Harris各自的原理、方法和优缺点进行了详细阐述，并对Harris角点检测算法和归一化互相关匹配算法进行了实验仿真，仿真结果表明算法存在一定数量的误匹配点对。3、提出了一种改进的特征点匹配算法，该算法首先采用归一化互相关方法进行特征点初始匹配，然后依据特征点对之间存在距离上的相似性，利用特征点对的整体匹配度进行精确匹配，从而剔除误匹配点对，实验结果表明，该算法在提高图像拼接的准确性和实时性方面具有显著效果。4、设计开发出一个图像拼接系统，该系统能够对输入的图像进行去噪、灰度化和提取边缘等预处理，采用Harris角点检测算法和本文提出的改进特征点匹配算法，最终实现两幅图像的无缝拼接。经过大量图像拼接实验验证，该图像拼接系统在准确性和实时性方面能够满足基本的需求。更多还原

【Abstract】 The purpose of image mosaic is to make several images which have overlapping regionsinto a high resolution panoramic images.It is a research focus in digital image processingfield,and image mosaic technology has a wide range of applications in virtual reality,medicalimaging,remote sensing and many other fields.In this paper,the main research contents and results are as follows:1.This paper summarizes the general process and related algorithms of imagemosaic,then it introduces the knowledge of image acquisition,image noise suppression andtransformation model.Besides,it completes the median filter simulation experiment,and thesimulation experiment results show that denoising effect of median filter is good.2.This paper analyzes and compares the features,performance and implementation stepsof several common image registrationg methods.It focuses on the study of image registrationmethods based on feature points,then it put forward three kinds of corner detection operatorMoravec,SUSAN and Harris,and each principle,methods,advantages and disadvantages arediscussed in detail.Besides,it completes simulation experiment of the Harris corner detectionalgorithm and normalized cross correlation matching algorithm.The simulation results showthat algorithm has a certain number of false matching points.3.This paper presents an improved feature point matching algorithm.This improvedalgorithm first uses the normalized cross correlation method for initial matching,and then onthe basis of distance similarity between feature points,this algorithm uses match degree on thewhole for precise matching,thereby it can eliminate the false matching points.The experimentresults show that this improved algorithm has a significant effect on improving imagestitching accuracy and real-time performance.4.This paper designs and develops an image mosaicing system.This system is able tocomplete image pretreatment,such as denoising,gray,edge extraction and so on.On the basis ofHarris corner detecting algorithm and the improved feature point matching algorithm,this system finally achieves image seamless splicing.After a large number of image stitchingexperiment,this image mosaic system in accuracy and real-time performance can satisfy thebasic needs.更多还原