汕头大学学报:自然科学版
汕頭大學學報:自然科學版
산두대학학보:자연과학판
Journal of Shantou University(Natural Science Edition)
2012年
1期
65-73
,共9页
小波变换%双向二维主成分分析%加权%人脸识别
小波變換%雙嚮二維主成分分析%加權%人臉識彆
소파변환%쌍향이유주성분분석%가권%인검식별
wavelet transformation%two-directional two-dimensional principle component analysis%weight%face recognition
人脸识别过程中,针对二维主成分分析(2DPCA)算法在特征提取和数据降维上存在的问题,本文首先引入双向二维主成分分析(2D2DPCA)算法.该算法同时考虑图像行与列方向上的信息.考虑到人脸冈像存在信息冗余而影响识别率的问题,于是本文提出一种基于小波加权双向二维主成分分析(WT—W2D2DPCA)的人脸识别算法.该算法首先采用二级小波分解对人脸图像进行预处理,提取其低频部分;然后根据人脸图像的特性,将低频部分进行奇偶分解,并引入加权思想,重组低频人脸图像,最后在ORL人脸数据库上进行双向二维主成分分析.实验结果表明,该方法不仅克服了传统2DPCA系数矩阵大的问题,而且得到了比传统的2DPCA、2D2DPCA算法更好的识别效果.
人臉識彆過程中,針對二維主成分分析(2DPCA)算法在特徵提取和數據降維上存在的問題,本文首先引入雙嚮二維主成分分析(2D2DPCA)算法.該算法同時攷慮圖像行與列方嚮上的信息.攷慮到人臉岡像存在信息冗餘而影響識彆率的問題,于是本文提齣一種基于小波加權雙嚮二維主成分分析(WT—W2D2DPCA)的人臉識彆算法.該算法首先採用二級小波分解對人臉圖像進行預處理,提取其低頻部分;然後根據人臉圖像的特性,將低頻部分進行奇偶分解,併引入加權思想,重組低頻人臉圖像,最後在ORL人臉數據庫上進行雙嚮二維主成分分析.實驗結果錶明,該方法不僅剋服瞭傳統2DPCA繫數矩陣大的問題,而且得到瞭比傳統的2DPCA、2D2DPCA算法更好的識彆效果.
인검식별과정중,침대이유주성분분석(2DPCA)산법재특정제취화수거강유상존재적문제,본문수선인입쌍향이유주성분분석(2D2DPCA)산법.해산법동시고필도상행여렬방향상적신식.고필도인검강상존재신식용여이영향식별솔적문제,우시본문제출일충기우소파가권쌍향이유주성분분석(WT—W2D2DPCA)적인검식별산법.해산법수선채용이급소파분해대인검도상진행예처리,제취기저빈부분;연후근거인검도상적특성,장저빈부분진행기우분해,병인입가권사상,중조저빈인검도상,최후재ORL인검수거고상진행쌍향이유주성분분석.실험결과표명,해방법불부극복료전통2DPCA계수구진대적문제,이차득도료비전통적2DPCA、2D2DPCA산법경호적식별효과.
To solve the problem of feature extraction and dimensional reduction of 2DPCA algorithm in face recognition, two-directional two-dimensional principal component analysis (2D2DPCA) algorithm is adopted simultaneously in the row and column directions of face images. The recognition rate may be influenced by information redundancy of face images. An algorithm is proposed based on the wavelet transibrmation-weighted two-directional two-dimensional principle component analysis (WT-W2D2DPCA) for face recognition. Firstly, the two-level wavelet transformation is used to preprocess human facial image, and low frequency subband is extracted. Considering the even and odd symmetrical characteristics of human face, it is decomposed into even and odd symmetrical images, and the new low frequency facial image is reconstructed. Finally, the WT-W2D2DPCA is used in ORL face database. Experimental results show that not only the problem of the great coefficient matrix of the traditional 2DPCA is overcome, but also better performance than 2D2DPCA algorithm and 2DPCA algorithm after the weighted processing is achieved.