基于超像素和光流引导的正射影像拼接方法

Orthophoto Splicing Method Guided by Superpixel and Optical Flow

影像拼接是生成大规模数字正射影像的关键技术之一,但现有的影像拼接方法在进行多个影像拼接时存在拼接线穿过明显地物导致的鬼影现象。光流是观察者和场景间相对运动引起的影像边缘等的相对运动,其中,大光流对应影像间的变化区域,可用于检测正射影像间的明显地面区域。提出一种基于光流引导的新型影像拼接方法,通过超像素的密集光流提取影像中明显的地物信息,以避免接缝穿过明显的地面物体。采用由粗到细的接缝线优化策略,并在超像素级别上利用Dijkstra算法进行最佳拼接区域检测,从而提高接缝线检测的效率。在此基础上,结合归一化互相关成本函数在像素级别上进行拼接线的像素级优化,获得最优的接缝线。实验结果表明,该方法从主观视觉上能够生成高质量的接缝线,在保证拼接效率的情况下,SSIM质量评价指标较Dijkstra方法、图割方法以及商业软件OrthoVista得到明显提高。

Image splicing is one of the critical technologies to generate a large-scale digital orthophoto map.However,when existing methods split multiple images,they often produce visible parallax as seamline crosses objects.To address this issue,we propose a novel image splicing method based on optical flow.Optical flow is the relative motion of image edges caused by the relative motion between the observer and the scene,among them,the large optiac flow corresponds to the change area between images,which can be used to detect the obvious ground object area of orthophoto images.By using superpixel-based dense optical flow to extract obvious object information from the image,we can prevent the seamline from passing obvious objects.In addition,we adopt a coarse-to-fine seamline optimization strategy,and use the Dijkstra’s algorithm to detect the optimal splicing region at the superpixel level,so the efficiency of seamline detection can be improved.On this basis,a Normalized Cross Correlation(NCC)cost function is used for superpixel-level seamline optimization to obtain the optimal seamline.Experimental results show that the proposed method can produce visually high-quality seamlines.Compared with Dijkstra,state-of-the-art image segmentation method and commercial software OrthoVista,the proposed method displays a significant improvement in the SSIM quality indicator while the splicing efficiency is ensured.