作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2014, Vol. 40 ›› Issue (12): 214-219,224. doi: 10.3969/j.issn.1000-3428.2014.12.040

• 图形图像处理 • 上一篇    下一篇

一种强鲁棒性的稀疏NMF算法研究与应用

吴月,叶庆卫,王晓东,周宇   

  1. 宁波大学信息科学与工程学院,浙江 宁波 315211
  • 收稿日期:2014-01-24 修回日期:2014-03-05 出版日期:2014-12-15 发布日期:2015-01-16
  • 作者简介:吴 月(1988-),女,硕士,主研方向:图像处理;叶庆卫,副教授、博士;王晓东、周 宇,副教授、硕士。
  • 基金资助:
    国家自然科学基金资助项目(61071198);浙江省自然科学基金资助项目(LY13F010015)

Research and Application of a Sparse NMF Algorithm with Strong Robustness

WU Yue,YE Qingwei,WANG Xiaodong,ZHOU Yu   

  1. College of Information Science and Engineering,Ningbo University,Ningbo 315211,China
  • Received:2014-01-24 Revised:2014-03-05 Online:2014-12-15 Published:2015-01-16

摘要: 为提高稀疏非负矩阵分解(SNMF)算法对含噪声图像提取特征的有效性,引入噪声项,并结合SNMF设计新的稀疏优化目标函数,给出该目标函数的优化求解表达式,使提取出的特征具有稀疏性且能增强噪声抵抗能力。针对手机图像,提出一种强鲁棒性的SNMF算法,描述手机待分类界面图和模板子图集概念,以获取手机图像特征,并结合支持向量机实现分类识别。应用结果表明,该算法能够对图像数据进行大规模压缩获取手机图像特征,具有较强的鲁棒性,且以稀疏矩阵作为计算分类识别的目标矩阵,具备较高的识别率。

关键词: 非负矩阵分解, 稀疏约束, 鲁棒性, 手机图像, 特征提取, 特征识别

Abstract: In order to improve the effectiveness of Sparse Non-negative Matrix Factorization(SNMF) algorithm which is used in feature extraction of image data with noises,this paper adds a noise term and combines it with SNMF algorithm.It proposes a new sparse optimization objective function and works out its solution which can guarantee the sparseness of extracted feature and improves the algorithm’s immunity against noise.It uses this Robust Sparse Non-negative Matrix Factorization (RSNMF) algorithm on feature extraction and recognition of phone image,creates the concept of interface image and sub-graph of mobile phone,gets feature extraction of phone image and puts it in Support Vector Machine(SVM) to achieve classification recognition.Experimental results show that not only phone image data can be large-scale compressed through RSNMF algorithm with good robustness,but also it improves the recognition efficiency by generating sparse matrix as an intermediary target matrix to classification.

Key words: Non-negative Matrix Factorization(NMF), sparse constraint, robustness, phone image, feature extraction, feature recognition

中图分类号: