基于SIFT特征检测与匹配的快速图像拼接方法研究

发布时间：2019-06-27 15:43

【摘要】：随着计算机技术的快速发展,人类对于图像的质量要求也不断提升。图像拼接技术因其可以获取宽视角、高分辨率的图像一直成为计算机视觉领域的研究热点,被广泛应用在实际生活中的各个领域,包括虚拟现实、遥感图像、视频监控以及无人机航拍等。图像拼接技术指的是将两幅或者多幅部分重叠的图像序列经过空间匹配对准,融合拼接成一幅包含全部图像序列信息的宽视角和高分辨率的图像,从而解决图像视野与分辨率不能同时满足的矛盾。尽管目前不断有专家学者对图像拼接方法进行改进,但当前图像拼接技术仍然存在计算复杂度高、拼接速度慢的问题,也因此限制了它的实时应用。本文主要围绕如何加快图像拼接速度展开深入研究。本文首先介绍了图像拼接技术的研究背景、研究意义以及国内外研究现状。然后详细介绍了图像拼接技术各个流程的相关理论以及SIFT(Scale Invariant Feature Transform)特征点检测方法。通过阅读大量相关的中外文献,分析图像拼接过程的特点及特殊性,本文提出了三种加快图像拼接速度的方法,分别为基于SIFT特征矢量图的快速图像拼接方法、图像局部特征自适应的快速SIFT图像拼接方法以及结合投影误差校正的快速SIFT图像拼接方法。鉴于传统拼接方法中检测到的SIFT特征点数量过多、匹配过程较复杂的问题,提出一种基于SIFT特征矢量图的快速图像拼接方法。该方法首先结合相位相关算法,确定待拼接图像的重叠区域,限定SIFT特征点检测范围;然后考虑特征点的空间位置信息,构建SIFT特征矢量图像,以便在特征匹配时限制匹配点的搜索范围,快速获得匹配点对。实验结果表明:该方法能够在保证图像拼接质量的前提下减少大量的不必要搜索,提高图像拼接速度。针对SIFT特征点维数较高、特征点检测过程计算复杂度较高的问题,提出一种图像局部特征自适应的快速SIFT拼接方法。首先,对待拼接图像分块,确定图像局部块的特征类型;接着自适应采用不同的简化方法检测各局部块的特征点。然后,通过特征匹配求出变换矩阵,并结合RANSAC算法去除伪匹配对。最后,通过图像融合得到最终的拼接图像。从实验结果可以看出:这种方法能够有效提高图像拼接的效率,克服图像拼接中计算复杂度高的问题。分析图像拼接过程的特点和特殊性,提出一种结合投影误差校正的快速SIFT图像拼接方法。首先,该方法特征检测范围仅集中在待拼接图像重叠区域中的部分图像块,从中获取SIFT特征点信息。然后,在特征匹配后应用投影误差校正方法,达到充分利用有限匹配点对计算出高精度投影变换矩阵的目的,避免不必要的特征检测和匹配搜索,继而大大加快图像拼接速度。最后,结合图像拼接的质量评价方法对图像拼接结果做出质量分析,以反映改进方法的性能。实验结果表明:与当前快速图像拼接方法相比,该方法显著降低了拼接过程的时间,并且拼接结果具有较好的视觉效果,证明了这种方法的可行性和有效性。
[Abstract]:With the rapid development of computer technology, human's quality requirements for images are also rising. Because of its wide viewing angle, the image stitching technology has been a hot topic in the field of computer vision, and is widely used in all fields of real life, including virtual reality, remote sensing image, video monitoring and aerial photography of unmanned aerial vehicle. Image splicing technology refers to the contradiction that two or more partial overlapping image sequences are aligned by space matching, and fused and spliced into a wide-angle and high-resolution image containing all the image sequence information, so as to solve the contradiction that the image field and resolution cannot be met at the same time. Although there are many experts and scholars to improve the image stitching method, the current image splicing technology still has the problems of high computational complexity and slow splicing speed, and therefore the real-time application of the image splicing method is limited. This paper mainly focuses on how to speed up the research on the speed of image stitching. This paper first introduces the research background, significance of the image stitching technology and the current research situation at home and abroad. Then, the correlation theory and SIFT (Scale Invariant Feature Transform) feature point detection method of the image stitching technology are introduced in detail. By reading a large number of relevant Chinese and foreign literature and analyzing the characteristics and particularity of the image stitching process, three methods of accelerating the image stitching speed are proposed in this paper, which are the fast image stitching method based on the SIFT feature vector diagram, The invention relates to a fast SIFT image splicing method for local characteristic of an image and a quick SIFT image splicing method combined with projection error correction. In view of the problem of too many SIFT feature points and complex matching process, a fast image stitching method based on the SIFT feature vector diagram is proposed. The method comprises the following steps of: firstly, combining a phase correlation algorithm, determining an overlapping area of an image to be spliced, and defining a SIFT feature point detection range; then taking into account the spatial position information of the feature point, constructing an SIFT feature vector image so as to limit the search range of the matching point when the feature matches, And the matching point pair is quickly obtained. The experimental results show that the method can reduce the large amount of unnecessary searching under the premise of ensuring the image splicing quality, and improve the image splicing speed. Aiming at the problem that the SIFT feature point dimension is high and the calculation complexity of the feature point detection process is high, a quick SIFT splicing method of the local feature self-adaptation of the image is proposed. The method comprises the following steps of: firstly, segmenting the spliced image, and determining a characteristic type of an image local block; and then, detecting the characteristic points of each local block by adopting different simplified methods. Then, the transformation matrix is obtained by the feature matching, and the pseudo-matching pair is removed by combining the RANSAC algorithm. And finally, the final spliced image is obtained by image fusion. It can be seen from the experimental results that the method can effectively improve the efficiency of image stitching and solve the problem of high computational complexity in image stitching. This paper analyzes the characteristics and particularity of the image stitching process, and proposes a fast SIFT image stitching method combined with projection error correction. First, the method characteristic detection range is only concentrated in a partial image block in the overlapped area of the image to be spliced, and the SIFT feature point information is obtained from the partial image block. Then, the projection error correction method is applied after the feature matching, so that the purpose of calculating the high-precision projection transformation matrix by fully utilizing the limited matching point is achieved, and the unnecessary feature detection and matching search are avoided, and then the image splicing speed is greatly accelerated. In the end, the quality analysis of the image mosaic is made based on the quality evaluation method of image stitching to reflect the performance of the improved method. The experimental results show that the method significantly reduces the time of the splicing process compared with the current fast image splicing method, and the splicing result has good visual effect, and the feasibility and the effectiveness of the method are proved.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】