一种基于表面法向矢量的深度图像分割方法

A SURFACE NORMAL BASED SEGMENTATION APPROACH FOR 3-D RANGE IMAGES

针对深度图像中存在的各种噪声干扰,本文提出了一种基于表面法向矢量的多面体分割方法。通过对深度图像中表面法向矢量及其直方图的分析,利用一个四邻域生长算法实现任意形状区域的聚合,并通过融合处理消除噪声对深度图像分割结果的影响。利用这种分割方法在真实深度图像上取得了较好的实验结果,该方法具有实现简单,抗噪性能好的特点。

A surface normal based segmentation approach is proposed for polyhedron range images with presence of measurement noise. Based on the surface normal distribution features and their corresponding histogram, a four-neighborhood grouping algorithm is applied to segmented regions with arbitrary shapes. A merging processing is then used to eliminate the noise effets on the segmentation results. This proposed segmentation approach is less sensitive to random noise than many existing segmentation methods and it is also easy to implement. The experimental results show that this surface normal based segmentation approach works well in analyzing 3-D surfaces consisting of polyhedral components.