图像矩阵降维压缩的一种新方法  

A new method of image matrix reducing dimension and compress

作　　者：王建辉[1] 崔连延[1] 顾树生[1] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110004

出　　处：《控制与决策》2007年第12期1408-1410,1416,共4页Control and Decision

基　　金：国家自然科学基金项目(60474040);沈阳市科技计划项目(10220360-1-07)

摘　　要：由于矩阵的SVD分解方法反映了图像的内在属性和本质特征,在图像处理中得到了广泛应用.但在矩阵降维压缩过程中对奇异值的最小数目还没有定量的确定方法,这给SVD在图像特征提取中的应用带来很大困难.为此,在研究确定奇异值可压缩到最小数目的基础上,提出一个奇异值压缩的降秩准则,并给出了具体的证明过程和计算方法.计算机仿真研究验证了所提出计算方法的正确性.The method of the singular value decomposition （SVD） of matrixes reflects the inwardness and essence character of the images, so it has wideapplication in the image processing. However, there is no a quantitative calculational standard to get the minimum quantity of the singular value in image matrix reducing dimension and compress, which brings much trouble to the application of SVD in the extraction of image characters. Therefore, it is researched that the singular value can be compressed to minimum, based on which, a rule of reducing dimension is proposed, and a specific proof procedure and calculation are given also. Computer simulation shows the effectiveness of calculation method.

关 键 词：奇异值分解 降秩准则 人脸识别 保真度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]