一种背景抑制改进的显著性目标检测方法

计算机工程与科学

一种背景抑制改进的显著性目标检测方法

崔丽群，赵越，吴晓冬，魏可飞，刘晨

（辽宁工程技术大学软件学院，辽宁葫芦岛 125000）

收稿日期:2017-01-03 修回日期:2017-05-08 出版日期:2018-08-25 发布日期:2018-08-25
基金资助:
国家自然科学基金（61172144）；辽宁省教育厅项目（L2012113）

A salient objects detection method based on

background suppression improvement

CUI Liqun,ZHAO Yue,WU Xiaodong,WEI Kefei,LIU Chen

(School of Software,Liaoning Technical University,Huludao 125000,China)

Received:2017-01-03 Revised:2017-05-08 Online:2018-08-25 Published:2018-08-25

摘要/Abstract

摘要：

针对显著性目标检测在复杂背景下准确率低的问题，提出超复数傅里叶变换改进的条件随机场显著性目标检测方法。首先，建立图像无向图并提取节点特征；然后重构超复数傅里叶变换得到平滑振幅谱与相位谱，获得无向图节点背景抑制权值，从而初步确定多尺度高斯核背景抑制图；最后输入到训练后的条件随机场中，通过增强目标表示得到最终显著性目标区域。实验表明，本文方法在准确率上较现有流行方法有显著提高，且能够在抑制复杂背景的同时，准确锁定指定目标位置区域。实验验证本文方法在复杂背景下显著性目标检测具有较好的准确性和鲁棒性。

关键词: 显著性目标检测, 超复数傅里叶变换, 背景抑制, 条件随机场

Abstract:

Aiming at the problem that the salient objects detection has low accuracy under complex background, an improved Conditional Random Field (CRF) salient objects detection method is proposed by using Hypercomplex Fourier Transformation (HFT). Firstly, this method builds unoriented graphs and extracts the node features on the image; Then, the HFT is reconstructed to obtain the smooth amplitude spectrum and phase spectrum, and the background suppression weights of unoriented graph nodes are obtained. Thus, the multi-scale Gaussian kernel background suppression graphs are preliminarily determined. Finally, they are inputted into the trained Conditional Random Field, the final significant target area is obtained by enhancing target representation. Experimental results show that the proposed method has obviously higher accuracy than the existing methods and can restrain the complex background and lock the specified target location region accurately at the same time. Experiments verify that salient objects detection has good accuracy and robustness under complex background.

Key words: salient object detection, hypercomplex Fourier transformation(HFT), background suppression, conditional random field(CRF)

崔丽群，赵越，吴晓冬，魏可飞，刘晨. 一种背景抑制改进的显著性目标检测方法[J]. 计算机工程与科学.

CUI Liqun,ZHAO Yue,WU Xiaodong,WEI Kefei,LIU Chen.

A salient objects detection method based on

background suppression improvement

[J]. Computer Engineering & Science.

[1]	杜洁, 骆力明, 孙众. 基于ALBERT预训练模型的事件抽取技术研究[J]. 计算机工程与科学, 2023, 45(04): 711-717.
[2]	仇增辉, 赫明杰, 林正奎. 基于深度学习的网购评论命名实体识别方法[J]. 计算机工程与科学, 2020, 42(12): 2287-2294.
[3]	王豪聪1,2，赵晓叶1,2，彭力1,2. 基于前景增强与背景抑制的显著性物体检测[J]. 计算机工程与科学, 2018, 40(06): 1119-1124.
[4]	王恒升1,2，李熙印2. 机器人自然语言导航的层叠式条件随机场模型[J]. 计算机工程与科学, 2017, 39(08): 1538-1545.
[5]	孙东普，朱鸣华，林鸿飞. 中文专利属性值对抽取技术及应用[J]. J4, 2016, 38(04): 800-806.
[6]	王文珂1，文雅玫2，蔡喆2. 基于条件随机场模型的数据异常检测算法[J]. J4, 2015, 37(09): 1756-1760.
[7]	吕品1,2,3，钟珞1，蔡敦波2,3，吴云韬2,3. 基于CRF的中文评论有效性挖掘产品特征[J]. J4, 2014, 36(02): 359-366.
[8]	何中市,刘莉,邢欣来,毛小丽. 基于语义角色的中文事件识别[J]. J4, 2013, 35(4): 181-185.
[9]	张朝胜1，郭剑毅1,2，线岩团1,2，余正涛1,2，雷春雅1, 王海雄1. 基于条件随机场的英文产品命名实体识别[J]. J4, 2010, 32(6): 115-117.