6-11 領域分割・投票によるコストボリュームフィルタリングの高速化(第6部門 メディア工学(画像認識))

抄録

We present a fast method to solve discrete labeling problems based on cost volume filtering. By region splitting and voting, the number of sub-regions in cost slices are reduced. As a result, our method can reduce computational time to 60%〜80% of the original algorithm, while keeping the accuracy.