Multi-view Object Segmentation Based on Epipolar Plane Image Analysis

In order to segment interest objects from multi-view images accurately, an epipolar plane image (EPI) analysis based object segmentation algorithm is proposed. Firstly, the straight lines in EPI are extracted by an improved Hough transform. Then, according to the depth of object, the corresponding straight lines are converted into original image space to get an initial contour. Following that, we design a boundary growing algorithm to shorten the gaps between broken edge segments. Finally, a contour linking algorithm is applied to obtain a closed segmentation curve. Experimental results show that, the proposed method can extract frontal objects from complicated scene correctly.