Author Login Editor-in-Chief Peer Review Editor Work Office Work

Computer Engineering

Previous Articles     Next Articles

Image Foreground Segmentation Method Based on Significance Prior

SUN Jun  1,2,HUANG Zhiyong  3,CHEN Yilin  1   

  1. (1. School of Computer Science,Wuhan University,Wuhan 430072,China;2. Huanggang Polytechnic College,Huanggang 438002,China;3. College of Computer and Information Technology,China Three Gorges University,Yichang 443002,China)
  • Received:2014-03-13 Online:2015-05-15 Published:2015-05-15

基于显著先验的图像前景分割方法

孙 俊1,2,黄志勇3,陈壹林1   

  1. (1. 武汉大学计算机学院,武汉430072; 2. 黄冈职业技术学院,湖北黄冈438002;3. 三峡大学计算机与信息学院,湖北宜昌443002)
  • 作者简介:孙 俊(1972 - ),男,博士研究生,主研方向:图形图像处理;黄志勇,副教授、博士;陈壹林,博士研究生。

Abstract: In order to study the influence on object contour caused by significance prior information and appearance information in image,this paper presents an image foreground segmentation method. The method is obtained by considering the spectrum based significance probability map and codebook based appearance priori,and puts them into a unified probabilistic framework,the foreground probability distribution is obtained. For the test image,through the spectrum of significance,it calculates the foreground probability at different positions,and the probability of appearance model as foreground inside the region. Synthetically it obtains the probability of the target area for foreground. When exceeds a certain threshold,it can be considered as foreground. This method only needs a small amount of learning,and can get results very similar to true value image segmentation. Experimental results on standard image set show that the method is simple,fast,and effective.

Key words: image segmentation, significance, codebook model, spectrum significance, probability model, appearance model

摘要: 为研究图像中物体的显著先验信息和外观信息对物体轮廓所造成的影响,提出一种图像前景分割方法。通过将 频谱余量获得的显著概率先验与基于码书模型的外观先验结合,在一个概率框架下学习得到统一的图像前景概率分布。 对于测试图像,通过基于频谱的显著性计算其不同位置处出现前景的概率,计算基于区域内外观模型为前景的概率,综合 得到目标区域为前景的概率,该值超过一定阈值即可认为是前景。该方法仅需要较少量的学习,就能够得到一个近似于 真值图像的分割结果。在图像分割标准库上进行测试,结果表明,该方法计算简单,速度快,图像分割效果较好。

关键词: 图像分割, 显著性, 码书模型, 频谱显著性, 概率模型, 外观模型

CLC Number: