基于全变分的运动分割模型及分裂 Bregman 算法

提出了一种基于全变分的运动分割模型,可以适用于2D/3D视频.首先, 通过活动轮廓模型将分割与估计融合在同一能量函数中, 该模型能够同时进行分割曲面的演化和运动参数的估计. 其次,通过凸松弛方法将原始问题转化为等价的全变分模型, 克服了局部最小值问题.最后,采用分裂Bregman快速算法进行求解. 多组实验证明了本文方法对2D/3D视频的通用性和算法的高效性.

Motion Segmentation Model Based on Total Variation and Split Bregman Algorithm

A general motion segmentation model for 2D/3D videos based on total variation is presented in this paper. Firstly, a spatiotemporal energy functional based on active contour model is established to perform motion segmentation and estimation simultaneously. Secondly, a convex optimization technique is introduced to convert the original energy functional into a global convex functional based on total variation, which overcomes the local minimum problem. Finally, the Split Bregman algorithm is applied to solve the segmentation surface. Various experiments indicates the universality for both 2D and 3D videos, and demonstrates the feasibility and effectiveness of the proposed algorithm.