计算机应用研究
計算機應用研究
계산궤응용연구
APPLICATION RESEARCH OF COMPUTERS
2010年
4期
1557-1559
,共3页
特征抽取%线性鉴别分析%对称散布子空间%小样本问题%人脸识别
特徵抽取%線性鑒彆分析%對稱散佈子空間%小樣本問題%人臉識彆
특정추취%선성감별분석%대칭산포자공간%소양본문제%인검식별
feature extraction%linear discriminant analysis%symmetrical scatter subspace%small sample size problem%face recognition
为有效解决小样本问题,从线性子空间的角度出发,构造了一种矩阵变换,得到了类内散布矩阵的另一个对称线性子空间;通过对两个子空间的分别求解,从而得到样本有效的鉴别信息.该方法有效地解决了传统Fisher鉴别分析方法中的最终特征维数受类别数限制的问题.在NUST603和ORL人脸数据库上的实验结果验证了算法的有效性.
為有效解決小樣本問題,從線性子空間的角度齣髮,構造瞭一種矩陣變換,得到瞭類內散佈矩陣的另一箇對稱線性子空間;通過對兩箇子空間的分彆求解,從而得到樣本有效的鑒彆信息.該方法有效地解決瞭傳統Fisher鑒彆分析方法中的最終特徵維數受類彆數限製的問題.在NUST603和ORL人臉數據庫上的實驗結果驗證瞭算法的有效性.
위유효해결소양본문제,종선성자공간적각도출발,구조료일충구진변환,득도료류내산포구진적령일개대칭선성자공간;통과대량개자공간적분별구해,종이득도양본유효적감별신식.해방법유효지해결료전통Fisher감별분석방법중적최종특정유수수유별수한제적문제.재NUST603화ORL인검수거고상적실험결과험증료산법적유효성.
To solve the small sample size problem efficiently, proposed a matrix transform on the basis of linear subspace theory, by which constructed a new linear symmetrical subspace of within-class scatter matrix. By the Fisher's discriminant crite-rion, derived two solution spaces from the within-class scatter matrix and respectively utilized its corresponding symmetrical subspace to obtain the efficient discriminatory information of the samples. Therefore, overcame the shortcoming of final dimensionality of features obtained by Fisher's discriminant analysis was limited by the number of classes. Experimental results conducted on the NUST603 and ORL face databases demonstrate the effectiveness of the proposed method.
