CAJ | 학술논문

针对(2D)2PCA无法保存某些重要局部特征的问题，提出了一种分块双向二维主成分分析融合局部特征方法。首先，将图像分解为互不重叠的子块，每个子块包含重要的局部信息，利用(2D)2PCA对子块进行特征提取并投影到特征子空间；然后，对每个子块分别设计一个分类器并在一定置信度范围内判别测试样本所属类别。最后，根据所有子块所属类别的置信度之和完成人脸分类。在四个人脸识别数据库上的实验结果表明，相比其它几种人脸识别算法，所提方法取得了更高的识别精度。
침대(2D)2PCA무법보존모사중요국부특정적문제，제출료일충분괴쌍향이유주성분분석융합국부특정방법。수선，장도상분해위호불중첩적자괴，매개자괴포함중요적국부신식，이용(2D)2PCA대자괴진행특정제취병투영도특정자공간；연후，대매개자괴분별설계일개분류기병재일정치신도범위내판별측시양본소속유별。최후，근거소유자괴소속유별적치신도지화완성인검분류。재사개인검식별수거고상적실험결과표명，상비기타궤충인검식별산법，소제방법취득료경고적식별정도。
A fusion method of blocked (2D)2PCA and local feature is proposed because of (2D)2PCA unable to preserve some essential local features. Firstly, the image is decomposed into non-overlapping image sub-blocks;each sub-block contains important local information which uses (2D)2PCA projecting sub-blocks into feature subspace and then design a classifier for each sub-block with a certain degree of confidence voting for sample identification category, the final classification result is the sum of confidence with all the sub-blocks. The proposed method has the same fea-ture matrix as (2D)2PCA. Experiments on four databases show that the proposed method achieves better recognition results in terms of recognition accuracy compared to other more advanced face recognition algorithms.