CAJ | 학술논문

在小样本情况下，传统的2DPCA 算法采用的训练样本的平均值不一定就是训练样本分布的中心，而矩阵广义低秩逼近（GLRAM）算法需要多次迭代求解左右投影变换矩阵，复杂度高。为了解决这些问题，利用基于样本中间值的2DPCA 算法（M2DPCA），通过协方差矩阵获得右变换矩阵，进一步对其投影特征矩阵降维获得左投影变换矩阵，提出一种改进的 GLRAM算法的掌纹识别方法。在 PolyU 掌纹库上实验表明：改进的 GLRAM算法在节省了大量训练时间的同时，取得了比 GLRAM算法更好的重构效果和识别率。
재소양본정황하，전통적2DPCA 산법채용적훈련양본적평균치불일정취시훈련양본분포적중심，이구진엄의저질핍근（GLRAM）산법수요다차질대구해좌우투영변환구진，복잡도고。위료해결저사문제，이용기우양본중간치적2DPCA 산법（M2DPCA），통과협방차구진획득우변환구진，진일보대기투영특정구진강유획득좌투영변환구진，제출일충개진적 GLRAM산법적장문식별방법。재 PolyU 장문고상실험표명：개진적 GLRAM산법재절성료대량훈련시간적동시，취득료비 GLRAM산법경호적중구효과화식별솔。
Under the condition of small sample,the average of all training samples used in traditional two-dimensional principal component analysis (2DPCA)is not always the scatter centre of the samples.In addition,the generalised low rank approximation of matrix (GLRAM) algorithm has to iterate many times for seeking the solution of the left and right projection transformation matrix,resulting in high degree of complexity.In order to solve these problems,we get the right projection transform matrix by making use of sample median-based 2DPCA algo-rithm (M2DPCA)and through covariance matrix,and obtain the left projection transform matrix by further reducing the dimensionality of pro-jection feature matrix of M2DPCA.Then the palmprint recognition algorithm of the generalised low rank approximation of matrix is proposed. Experiments on PolyU palmprint database indicate that while the improved GLRAMsaves a lot of training time,it also gets better performance than GLRAMin image reconstruction and recognition rate.