基于SVD和曲波变换的图像特征提取研究  

Study on Image Feature Extraction Using Singular Value Decomposition and Curvelet Transform

作　　者：张云强[1] 张培林[1] 徐超[1] 王国德[1] 

机构地区：[1]军械工程学院一系,河北石家庄050003

出　　处：《计算机仿真》2012年第12期303-306,共4页Computer Simulation

基　　金：国家自然科学基金项目(50705097);清华大学摩擦学国家重点实验室开放基金资助项目(SKLTKF09B06)

摘　　要：关于图像特征提取优化问题,为有效地描述图像特征,提出了一种奇异值分解(SVD)和曲波变换的特征提取方法,首先对图像进行奇异值分解和曲波变换,分别获得图像的奇异值和不同尺度的曲波系数,根据识别成功率选取一组较大的奇异值,并计算各尺度曲波系数的均值、标准差、能量和熵等统计特征。最后利用选取的奇异值和曲波系数的统计特征构造特征集描述图像特征。将提取的特征集应用于纹理图像识别,平均识别率达到了94%。仿真结果表明,改进方法提取的特征集能很好地刻画图像特征,应用于图像识别可获得较高的识别成功率。To depict images features effectively, an image feature extraction method based on singular value decomposition（SVD） and Curvelet transform was proposed. For each image, singular values and Curvelet coefficients of different scales were firstly obtained utilizing SVD and Curvelet transform. Then, a group of big singular values was selected according to image recognition rates of success. Meanwhile, the mean, standard deviation, energy and entropy of Curvelet coefficients of different scales were calculated. Finally, a feature set was constructed by the selected singular values and the statistic features of Curvelet coefficients to describe image features. When the extracted feature set was applied for image recognition, an average recognition rate of 94% was reached. Simulation experiment results show that the feature set extracted by the proposed method can describe image features perfectly. A high recognition rate of success can be achieved if the feature set is employed for image recognition.

关 键 词：图像识别 特征提取 奇异值分解 曲波变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]