图像匹配快速算法的研究与实现

发布时间：2018-07-02 11:53

本文选题：图像匹配 + HOG特征　；参考：《华中科技大学》2014年硕士论文

【摘要】：在日益发展的信息世界中，图像匹配技术已经成为图像处理技术中一个非常重要的研究领域。新一代互联网的图像搜索引擎技术、军用飞行器的目标识别技术、工业机器人视觉技术、医学图像处理技术无一不用到图像匹配技术。目前，，研究者已经提出了多种图像匹配算法。这些匹配算法需要用统一的指标对其进行评价，衡量匹配算法性能的指标主要包括匹配精度、匹配实时性、匹配概率。本文采用HOG特征作为特征描述子，实现了一种基于HOG特征的图像匹配快速算法。该算法对图像进行分割后，利用积分直方图的方法快速构建梯度方向直方图，用HOG特征表达特征向量，衡量待匹配图像和模板图像的相似性，得到最佳匹配位置。该算法能够在保证匹配精度的前提下，相对于NCC算法，较大的提高图像匹配算法的实时性。此外，基于空间降维的思想，本文提出了一种快速匹配算法。该算法将常规的二维图像空间投影到一维图像空间，再将二维变量的相关系数公式延伸到一维变量的相关系数，并将它作为相似性测度。该算法不仅可以保持良好的匹配精度和匹配概率，而且在实时性上较NCC算法有了两个数量级的提高。最后，针对图像匹配算法性能评价指标，在基于MFC的图像匹配算法软件平台上，对本文提出的快速匹配算法和实现的其他匹配算法做了对比实验，并分析实验结果。
[Abstract]:In the information world, image matching technology has become a very important research field in image processing technology. The new generation of Internet image search engine technology, military aircraft target recognition technology, industrial robot vision technology, medical image processing technology does not need to image matching technology. At present, researchers have proposed a variety of image matching algorithms. These matching algorithms need to be evaluated with a unified index. The performance of the matching algorithm is mainly measured by matching accuracy, real-time matching, and matching probability. In this paper, a fast image matching algorithm based on hog features is implemented by using hog features as feature descriptors. After image segmentation, the gradient direction histogram is constructed quickly by the method of integral histogram, the feature vector is expressed by hog feature, the similarity between the image to be matched and the template image is measured, and the best matching position is obtained. This algorithm can greatly improve the real-time performance of the image matching algorithm compared with the NCC algorithm on the premise of ensuring the matching accuracy. In addition, based on the idea of space dimension reduction, this paper proposes a fast matching algorithm. In this algorithm, the conventional two-dimensional image space is projected into one-dimensional image space, and then the correlation coefficient formula of two-dimensional variable is extended to the correlation coefficient of one-dimensional variable, which is regarded as similarity measure. The algorithm can not only maintain good matching accuracy and probability, but also improve the real-time performance of NCC algorithm by two orders of magnitude. Finally, aiming at the performance evaluation index of image matching algorithm, a comparison experiment between the fast matching algorithm and other matching algorithms proposed in this paper is made on the MFC based image matching algorithm software platform, and the experimental results are analyzed.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41

【参考文献】