面向人体参数测量的非闭合Snake模型局部轮廓提取

摘要

The traditional Snake model is not suitable for local contour extraction, and is easily affected by complex background and human dressing. To solve these problems, a shape-constrained unclosed Snake model was proposed and applied to the local contour extraction for anthropometry. Firstly an unclosed Snake model was realized by analyzing the solving process of the Snake model and resetting the differentiability of each point. Then the shape constraint based on prior knowledge of anthropometry was added to the unclosed Snake model, so that the Snake curve can keep the specified contour shape in iterations. Finally, a gradient-based automatic contour initialization method was adopted to avoid tedious manual settings. Experimental results show that the proposed method can extract local body contours effectively, reduce the interference caused by complex background and human dressing, and basically meet the requirements of anthropometry.%针对传统 Snake 模型不适用于局部轮廓提取,且在提取人体轮廓时易受背景和人体着装影响的问题,提出一种带形状约束的非闭合Snake模型.首先对Snake模型求解过程进行分析,重新设置每个点的可导性,实现了非闭合Snake模型; 然后针对人体参数测量,根据先验知识为非闭合Snake模型添加形状约束力项,使得Snake曲线在迭代过程中能够保持指定轮廓形状; 最后采用一种基于梯度的初始轮廓自动设置方法避免人工设置的烦琐.实验结果表明,该模型能有效地提取人体局部轮廓,降低复杂背景和人体着装带来的干扰,基本满足人体参数测量的要求.