Convex Shape Prior for Multi-Object Segmentation Using a Single Level Set Function

Shousheng Luo, Xue-Cheng Tai, Limei Huo, Yang Wang, Roland Glowinski; The IEEE International Conference on Computer Vision (ICCV), 2019, pp. 613-621

Abstract


Many objects in real world have convex shapes. It is a difficult task to have representations for convex shapes with good and fast numerical solutions. This paper proposes a method to incorporate convex shape prior for multi-object segmentation using level set method. The relationship between the convexity of the segmented objects and the signed distance function corresponding to their union is analyzed theoretically. This result is combined with Gaussian mixture method for the multiple objects segmentation with convexity shape prior. Alternating direction method of multiplier (ADMM) is adopted to solve the proposed model. Special boundary conditions are also imposed to obtain efficient algorithms for 4th order partial differential equations in one step of ADMM algorithm. In addition, our method only needs one level set function regardless of the number of objects. So the increase in the number of objects does not result in the increase of model and algorithm complexity. Various numerical experiments are illustrated to show the performance and advantages of the proposed method.

Related Material


[pdf]
[bibtex]
@InProceedings{Luo_2019_ICCV,
author = {Luo, Shousheng and Tai, Xue-Cheng and Huo, Limei and Wang, Yang and Glowinski, Roland},
title = {Convex Shape Prior for Multi-Object Segmentation Using a Single Level Set Function},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}