一种缩减图像维数的方法及其在人脸图像上的应用
doi: 10.3724/SP.J.1146.2006.00935
An Approach to Image Dimension Reduction and Its Application to Face Images
-
摘要: 2DPCA是一种快速且有效的图像特征抽取方法。不同于传统的主分量分析(PCA)方法,该方法以全新的思路应用主分量分析技术,它直接计算图像矩阵到矢量的投影,并将其看作图像特征。实际上,2DPCA是此种思路下的最优压缩技术。对2DPCA而言,存在两种抽取图像矩阵特征的技术路线,这两种路线将图像变换到不同的空间,且分别突出人脸图像横向和纵向的特质。由于这两种技术路线抽取的特征具有互补性,该文分别设计两种方案对这两类特征加以融合。基于特征融合的识别实验取得了较优的识别正确率。Abstract: As a technique of feature extraction, 2DPCA is effective and efficient. Different from traditional PCA, it directly computes projection of one image matrix onto vector, to obtain feature for the image. In fact, 2DPCA is optimal for dimension compression under this consideration. There are two approaches to implement 2DPCA. The two approaches transform images into different spaces, and emphasize horizontal feature and vertical feature of face images respectively. Because the features extracted by the two approaches may complement each other, two schemes are designed to perform feature fusion. Experiments based on the fused features achieve high classification right rates.
计量
- 文章访问数: 3358
- HTML全文浏览量: 85
- PDF下载量: 875
- 被引次数: 0