Efficient Perceptual Region Detector based on Object Boundary 3
meaningful regions. The main contributions in this paper can be concluded as
follows:
– This paper proposes a unified work to find perceptual-homogeneous regions
in nature images. Once generated by CAR, these regions can be useful to
many computer visual applications, such as image retrieval, object recogni-
tion and segmentation.
– Inspired by general object proposal methods [14], we improve SLIC [9] su-
perpixel method by clustering pixels considering not only in color and image
plane space, but also the contour influence. The improved superpixel seg-
mentation is fast, regular and more close to the boundaries.
– A hierachical region tree is built with superpixels merge process using water-
shed algorithm [20] and an objectness function is defined to select perceptual
regions.
– Extensive exp eriments on the Berkeley benchmark dataset [8] show that
the superpixels generated by our method evidently outperform the state-of-
art method, as measured by boundary recall and under-segmentation error.
what’s more, our method can find representative meaningful regions which
exhibit promising repeatability.
The rest of this paper is organized as follows. Section 2 briefly review existing
methods about superpixels generation and local detector. Section 3 describe
the key stages in CAR detector. Experiments results including boundary recall,
under-segmentation error and repeatability are shown in Section 4. At last, we
conclude our paper in Section 5.
2 Related Work
In this section, we briefly introduce some related work and discuss the differences
between these methods and ours.
2.1 Superpixel
According to the grouping strategies, these methods can be broadly divided
into two categories: graph-based algorithms and Meanshift [10] like algorithms.
The graph-based algorithms usually treat each pixel of the image as a normal
node in a graph, and two special nodes are added as foreground and background
respectively. The cost function is defined considering the similarity between each
connected normal nodes and the similarity between normal nodes and special
nodes. Superpixels are generated by minimizing the cost function on that graph.
The well known algorithm in this category is Normalized Cuts [1]. However the
complexity of these algorithms grows fast as the number of sup erpixels increases,
because the superpixel generation scheme depends on the eigen-based solution.
Due to the nature of graph-based algorithm, the methods in this category always
generate irregular superpixels both in size and shape, and suffer from efficiency
issue.
剩余11页未读，继续阅读
weixin_38739919
- 粉丝: 4
- 资源: 903
我的内容管理
展开
最新资源
- 最优条件下三次B样条小波边缘检测算子研究
- 深入解析：wav文件格式结构
- JIRA系统配置指南：代理与SSL设置
- 入门必备：电阻电容识别全解析
- U盘制作启动盘：详细教程解决无光驱装系统难题
- Eclipse快捷键大全：提升开发效率的必备秘籍
- C++ Primer Plus中文版：深入学习C++编程必备
- Eclipse常用快捷键汇总与操作指南
- JavaScript作用域解析与面向对象基础
- 软通动力Java笔试题解析
- 自定义标签配置与使用指南
- Android Intent深度解析：组件通信与广播机制
- 增强MyEclipse代码提示功能设置教程
- x86下VMware环境中Openwrt编译与LuCI集成指南
- S3C2440A嵌入式终端电源管理系统设计探讨
- Intel DTCP-IP技术在数字家庭中的内容保护
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
