电子学报 ›› 2016, Vol. 44 ›› Issue (2): 453-460.DOI: 10.3969/j.issn.0372-2112.2016.02.029

• 学术论文 • 上一篇    下一篇

基于二维压缩感知和分层特征的图像检索算法

周燕, 曾凡智   

  1. 佛山科学技术学院 计算机系, 广东佛山 528000
  • 收稿日期:2015-08-10 修回日期:2015-11-03 出版日期:2016-02-25
    • 作者简介:
    • 周燕 女,1979年12月生于江西抚州.2004年毕业于东华理工大学信息工程学院,硕士.现为佛山科学技术学院计算机系副教授.主要研究方向:图像处理.E-mail:zhouyan791266@163.com;曾凡智 男,1965年1月生于湖北洪湖.1984年、1987年和2009年分别在厦门大学、中国科学院计算中心和华中科技大学获学士、硕士和博士学位.现为佛山科学技术学院计算机系教授.主要研究方向:图像处理、数据挖掘.E-mail:coolhead@126.com
    • 基金资助:
    • 广东省自然科学基金 (No.2015A030313635); 广东省科技计划 (No.2014A010103037); 佛山市科技创新专项资金 (No.2015AG10008,No.2014AG10001); 佛山科学技术学院优秀青年教师培养计划 (No.fsyq201411); 佛山科学技术学院优秀青年人才培育项目

An Image Retrieval Algorithm Based on Two-Dimensional Compressive Sensing and Hierarchical Feature

ZHOU Yan, ZENG Fan-zhi   

  1. Department of Computer Science, FoShan University, Foshan Guangdong 528000, China
  • Received:2015-08-10 Revised:2015-11-03 Online:2016-02-25 Published:2016-02-25
    • Supported by:
    • National Natural Science Foundation of Guangdong Province,  China (No.2015A030313635); Science and Technology Project of Guangdong Province (No.2014A010103037); Foshan Science and Technology Innovation Project (No.2015AG10008, No.2014AG10001); Excellent Young Teachers Training Program of Foshan University (No.fsyq201411); Excellent Young Talent Training Program of Foshan University

摘要:

为了保留图像分析时的像素点位置关系及降维处理,把一维压缩感知理论推广到二维,建立了二维可稀疏信号的压缩测量模型,研究了一种二维信号的自适应梯度下降重构AGDR(Adaptive Gradient Descent Recursion)算法,由此提出了一种图像分层特征提取与检索方法.首先对图像在RGB颜色空间上进行网格离散划分,通过分层算子对图像进行分层映射,定义一种基于颜色网格空间的扩展灰度共生矩阵,采用二维测量模型获取图像的分层测量特征、纹理特征与分层颜色统计特征,图像分层测量特征综合反映出图像的颜色及像素点位置的关系,扩展灰度共生矩阵反映纹理特征.其次用AGDR算法计算检索图像之间的原始信号差量及其稀疏值.最后结合两类分层特征差量、稀疏值和颜色统计特征,融合计算图像间整体相似度度量指标.仿真实验表明,应用分层二维压缩感知测量与AGDR算法的图像检索方法在检索时间、查全率和查准率等指标上具有优越性能,为图像检索提供了新思路.

关键词: 二维压缩感知, 图像检索, 图像分层特征, 纹理特征, 自适应梯度下降重构

Abstract:

To retain the position relationship of pixels when image analyzing and dimension reducing, we extend the one-dimensional compressive sensing theory to two-dimensional, and establish a two-dimensional compressive measurement model for sparse signal.We study an adaptive gradient descent recursion algorithm for two-dimensional signal, and propose an image hierarchical feature extraction and retrieval method.Firstly, it conducts grid discrete division on the RGB color space, and mapping to the image by hierarchical operator.It defines an extended GLCM based on color grid space, and extracts the hierarchical measurement feature, texture feature and hierarchical color statistical feature by the two-dimensional measurement model.The hierarchical measurement feature of image reflects the position relationship between the image color and pixel, and the extended GLCM reflects the texture feature.Secondly, it calculates the original signal difference and sparse value between images by the AGDR algorithm.Finally, it calculates the overall similarity metrics between images by combining the two hierarchical feature difference, the sparse value and the color statistical feature.The simulation results show that the image retrieval method which applying hierarchical two-dimensional compressive sensing measurement and AGDR algorithm has superior performance on retrieval time, recall and precision, it provides a new idea for the image retrieval.

Key words: two-dimensional compressive sensing, image retrieval, image hierarchical feature, texture feature, adaptive gradient descent recursion

中图分类号: