一种基于改进动态规划的最佳拼接线搜索方法

AN OPTIMAL STITCHING SEAM FINDING ALGORITHM BASED ON IMPROVED DYNAMIC PROGRAMMING

寻找最佳拼接线是图像拼接中的重要一步,主要用于消除图像重叠部分的配准错位以及移动物体造成的拼接伪影。为了得到一条较理想的拼接线,根据重叠区域图像的颜色和纹理相似性提出一个新的能量函数。同时对传统查找最佳拼接线的动态规划算法进行了搜索方向上的改进,将原来只搜索当前点相邻行正对的三个点改为搜索相邻行中所有的点。利用此改进的搜索算法,在构造的能量函数上找出一条能量值之和最小的全局最佳拼接线。实验结果显示,改进的算法搜索到的拼接线可以成功地绕过重叠区域内的较大目标,保留了目标的完整性,从而实现视觉感观上更加自然的图像拼接。同时,实验数据表明,该拼接线的平均能量值之和比传统方法小10%。

Optimal stitching seam finding is an important step for image stitching, which is mainly used to eliminate misalignment artifacts in the overlapped area and stitching artifacts caused by moving objects. In order to find a better stitching line, we proposed a new energy function according to the color and texture similarity of images in the overlapped area. We also improved the traditional dynamic programming algorithm in the search direction for finding the optimal stitching seam. We searched for all points in the adjacent row rather than only three points right to the adjacent tow of the current point. A global optimal stitching line with minimum sum of energy values was found on the proposed energy function by using the improved search algorithm. The experimental results show that the stitching seam, searched by the improved algorithm, can successfully avoid larger objects in the overlapped area and maintain the integrity of the objects, so as to realize a more natural image stitching in visual sense. The experimental data illustrate that the sum of the average energy of the stitching seam is 10% smaller than that of the traditional method.