基于矩形子块法的快速全局立体匹配算法

A Fast Global Method for Correspondence Computing Based on Rectangular Subregioning

由于计算量大,采用割图法(GC,GraphCuts)来最小化立体匹配的能量函数,计算速度慢,难以满足许多方面应用的要求。为了提高立体匹配能量函数最小化的速度,文章提出了一种新的算法:矩形子块割图法(RSRGC,RectangularSubregioningGraphCuts),即将大图像分割为小矩形子块,然后再在各小矩形子块上分别利用GC最小化能量函数。实验表明,采用该方法可以大大提高立体匹配的速度,而不影响立体匹配的效果。

Due to the expensive computing cost,the speed of energy minimization by graph cuts in stereo correspondence computing is not fast enough for many applications.In order to improve the speed of the energy minimization for stereo correspondence computing,we propose a new algorithm：rectangular subregioning graph cuts（RSRGC）.The essential of RSRGC is dividing the image into small rectangular regions（similar to Changming Sun＇s Rectangular Subregioning method） and minimizing energy by graph cuts in them respectively.The experiments with real images show that this method improving the speed of computing correspondence greatly while at the same time keeping the quality of the computing result.