结合投影误差校正的快速SIFT图像拼接

摘要

Aiming at the problems of limited instantaneity in current image stitching methods,a fast image stitching method through combining feature points detection on part of image with correction of projection error in the overlapping area in the process of image stitching was proposed.Firstly,the scope of the feature detection for the method only focused on part of image blocks in the overlapping area of the image which were needed to be stitched for obtaining information of feature points for Scale Invariant Feature Transform(SIFT).Secondly,after the feature matching,the correction method was used for projection error to make full use of limited matching points to figure out high-precision projection transformation matrix,in order to avoid unnecessary feature detections and matching searching for greatly accelerating the speed to stitch images.Finally,combining with quality evaluation method for image stitching,quality analysis was made on image stitching result for reflecting the performance of the improved method.The result of the experiment shows:compared with the fast image stitching method mentioned in references,on the premise of maintaining the stitching quality of the image,the method in the thesis significantly reduces the time for the stitching process.Among three experimental images,the average stitching speed is increased approximately by 54%,which shows the feasibility and effectiveness of the proposed method.%针对现有图像拼接方法实时性受限的问题,提出了一种通过对图像拼接中重叠区域部分图像特征检测和投影误差校正两部分相结合的快速图像拼接方法.首先,该方法特征检测范围仅集中在待拼接图像重叠区域的部分图像块,从中获取尺度不变特征变换(SIFT)的特征点信息.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,避免了不必要的特征检测和匹配搜索,继而大大加快了图像拼接速度.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:对比文献所提出的快速图像拼接方法,本文方法在保证图像拼接质量的前提下显著降低了拼接时间;在所给的3组实验图像中,拼接速度平均提升了54%左右,证明了本文方法的可行性和有效性.