基于BRISK图像配准的全景图像拼接研究

发布时间：2018-03-10 13:41

本文选题：图像拼接　切入点：BRISK算法　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：图像拼接技术在图像处理领域有着重要意义,也是图像处理领域中的研究热点。为了将同一场景拍摄下的几张图像拼接成一个具有较宽视角、无明显拼接缝隙的、信息度较高的全景图像就需要应用全景图像拼接技术。全景图像拼接在计算机识别、医学图像分析以及虚拟现实技术等领域均有重要应用。本文的主要内容分为图像配准以及图像融合技术两部分,在深入研究现有算法的基础上,对图像配准以及图像融合技术均提出了改进算法。本文首先阐述了图像拼接技术相关的研究背景及意义,并且分析了国内外研究现状,介绍了图像拼接过程中的基本流程,实现了基于归一化积相关的BRISK图像配准,以及基于彩色共生矩阵的多分辨率图像融合算法的全景图像拼接。对于图像配准,本文提出了基于归一化积相关的BRISK图像配准算法。该算法首先对图像应用AGAST算法提取特征点,再应用归一化积相关算法对图像进行块匹配,对匹配成功的重叠区域,采用特征点的主方向以及描述子向量进行特征点的匹配,对于匹配成功的特征点,应用随机样本一致性方法进行误匹配点的去除。对三组有旋转及视角变化的图像进行仿真实验,通过对仿真得到的图像和数据分析表明:本文提出的改进BRISK图像配准算法,与原算法、传统SIFT算法和先进算法ORB相比较,特征点匹配准确率与匹配速度均有提高。在图像配准完成之后,通过柱面投影算法计算出图像之间的几何变换关系,然后对图像序列之间进行矩阵变换,实现多幅图像的拼接。在拼接完成后,进行图像融合,以消除拼接缝隙和鬼影等现象。对于图像融合,本文提出了基于彩色共生矩阵的多分辨率图像融合算法。该算法首先对图像进行彩色共生矩阵计算,提取图像重叠区域的纹理特征,利用提取的纹理特征对图像进行多分辨率融合。图像融合实验应用融合评价标准对实验结果进行分析表明:本文提出的改进多分辨率图像融合算法比原算法具有更好的融合效果,消除了拼接缝隙和鬼影,实现了全景图像的无缝拼接。
[Abstract]:Image mosaic technology is of great significance in the field of image processing, and it is also a research hotspot in the field of image processing. Panoramic images with higher information level need to apply panoramic image mosaic technology. Panoramic image stitching is recognized by computer. Medical image analysis and virtual reality technology have important applications. The main content of this paper is divided into two parts: image registration and image fusion technology. An improved algorithm for image registration and image fusion is proposed. Firstly, the research background and significance of image mosaic technology are described, and the current research situation at home and abroad is analyzed, and the basic process of image stitching is introduced. BRISK image registration based on normalized product correlation and panoramic image mosaic based on multi-resolution image fusion algorithm based on color co-occurrence matrix are realized. In this paper, a BRISK image registration algorithm based on normalized product correlation is proposed. Firstly, the AGAST algorithm is applied to extract the feature points, then the normalized product correlation algorithm is applied to the block matching of the image. The principal direction of the feature points and the description subvectors are used to match the feature points. The method of random sample consistency is used to remove the mismatch points. Three groups of images with rotation and angle of view change are simulated. The analysis of the simulated images and data shows that the improved BRISK image registration algorithm is proposed in this paper. Compared with the original algorithm, the traditional SIFT algorithm and the advanced algorithm ORB, the accuracy and speed of feature point matching are improved. After image registration is completed, the geometric transformation relationship between images is calculated by cylindrical projection algorithm. Then the matrix transformation is carried out between the image sequences to realize the mosaic of multiple images. After the stitching is completed, the image fusion is carried out in order to eliminate the phenomena of stitching gaps and ghost images, etc. For image fusion, In this paper, a multi-resolution image fusion algorithm based on color co-occurrence matrix is proposed. Using the extracted texture features to fuse the image with multi-resolution. The experimental results of image fusion are analyzed by using the fusion evaluation standard. The results show that the improved multi-resolution image fusion algorithm proposed in this paper is better than the original image fusion algorithm. Better fusion, The stitching gap and ghost image are eliminated and the panoramic image is jointed seamlessly.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】