CAJ | 학술논문

为得到立体图像对的全局最优匹配，将视差搜索范围离散化，与图像坐标一起构成三维空间网络.恰当定义网络各边的容量，使之兼顾立体匹配的相容性和光滑性约束，将立体匹配转化为网络优化问题.通过求解网络的最大流和最小切割，获得全局最优的视差分布数据.实验表明，算法生成的视差数据不仅连续稠密而且保留了细节信息.
위득도입체도상대적전국최우필배，장시차수색범위리산화，여도상좌표일기구성삼유공간망락.흡당정의망락각변적용량，사지겸고입체필배적상용성화광활성약속，장입체필배전화위망락우화문제.통과구해망락적최대류화최소절할，획득전국최우적시차분포수거.실험표명，산법생성적시차수거불부련속주밀이차보류료세절신식.
In order to get the global optimal matching of stereo image pairs, the algorithm discretized the searching range of disparity, then combined it with image coordinates to construct a 3D network. The capacity of net edge was properly defined to satisfy both smooth and compatible constraint. Thus, stereo matching was transformed to the problem of network optimizing. The global optimal disparity was derived from solving the maximum flow and minimum cut of the network. Experiments prove that the algorithm not only produces smooth and dense disparity but also keeps detail information. It establishes a good foundation for 3D recoinstruction.