基于集成学习思想的深度图像遮挡边界检测方法

Occlusion Boundary Detection Method for Depth Image Based on Ensemble Learning

针对现有深度图像遮挡检测方法不能有效地检测出深度信息变化不明显的遮挡边界点的状况，提出了8邻域总深度差特征和最大面积特征，并定义了计算方法。在此基础上，提出一种新的基于集成学习思想的深度图像遮挡边界检测方法，该方法结合所提新特征及现有遮挡相关特征训练基于决策树的AdaBOOst分类器，完成对深度图像中遮挡边界点及非遮挡边界点的分类，实现对深度图像中遮挡边界的检测。实验结果表明，同已有方法相比，所提方法具有较高的准确性和较好的普适性。

The existing occlusion detection method for depth image can not effectively detect the occlusion boundary point with less obvious depth change, this status should be changed. The eight neighborhood total depth difference feature and maximal area feature are proposed firstly, and then the calculation methods for these two new features are defined. On this basis, a new occlusion detection approach based on ensemble learning is proposed, which combines the proposed features and existing occlusion related features to train the decision tree-based AdaBoost classifier to classify the pixel of depth image into occlusion boundary point or non-occlusion boundary point. The experimental results show that, compared with the existing methods, the proposed approach has higher accuracy and better universality.