计算机应用 ›› 2014, Vol. 34 ›› Issue (12): 3536-3539.

• 虚拟现实与数字媒体 • 上一篇    下一篇

基于姿势估计与显著性目标检测的衣物提取算法

何妮,赵波   

  1. 西南交通大学 信息科学与技术学院,成都 610031
  • 收稿日期:2014-07-03 修回日期:2014-08-08 出版日期:2014-12-01 发布日期:2014-12-31
  • 通讯作者: 何妮
  • 作者简介:何妮(1990-),女,四川遂宁人,硕士研究生,主要研究方向:图像分割、计算机视觉;赵波(1987-),男,江苏常州人,博士研究生,主要研究方向:图像分割、信息检索。
  • 基金资助:

    国家自然科学基金资助项目;四川省科技创新苗子工程资助项目

Clothing extraction algorithm based on pose estimation and salient object detection

HE Ni,ZHAO Bo   

  1. School of Information Science and Technology, Southwest Jiaotong University, Chengdu Sichuan 610031, China
  • Received:2014-07-03 Revised:2014-08-08 Online:2014-12-01 Published:2014-12-31
  • Contact: HE Ni

摘要:

针对衣物识别对服饰购物图像搜索的影响,分析了网络服饰购物图像的特点,将姿势估计和显著性检测结合,提出了一种基于姿势估计与显著性目标检测的衣物提取算法。该算法对图像进行姿势估计,实现姿势的适应性,将姿势估计融入显著性目标检测的区域检测部分,将二者优势互补,得到结合姿势估计的显著性检测图,衣物区域得以自动定位,通过迭代的图割方法提取出衣物。实验结果表明,所提算法可以较为准确地提取出复杂背景中的衣物,说明了衣物提取中引入姿势估计和显著性检测的有效性;同时适用于大部分服饰购物图像,具有较好的通用性。

Abstract:

Considering the influence of clothing recognition on clothing shopping image search, the characteristics of online clothing shopping images were analyzed and a novel clothing extraction algorithm was proposed on the basis of pose estimation and salient object detection, which combined pose estimation and salient object detection. Implementing pose estimation on images, the presented method realized adaptability to poses, and introduced that integrating pose estimation into the region detection part of the salient object detection to obtain the salient detection map combining pose estimation, which took two complementary advantages. The clothing region was automatically located. Clothing was extracted by adopting the graph cut principle iteratively. The experimental results demonstrate that the proposed algorithm can accurately extract the clothing in complex background, illustrating the effectiveness of the introduction of pose estimation and saliency detection to clothing extraction. Besides, it can be applied to most clothing shopping images, and has good universality.

中图分类号: