DCT系数重组在特征脸中的应用研究

Research on eigenface based on reordered DCT coefficients

针对现在图像压缩主要采用DCT变换方法,提出了一种对DCT系数进行重组的方法,获取三级多分辨率子带。通过对各单独子带实施特征脸方法可得到不同的识别效果。提出在子带S0和子带S1,S2和S3上进行特征脸方法识别,如两者结果不一致,则在满足一定门限条件下输入多张人脸,分析了所提出方法与原方法所需识别时间的关系。在ORL库和YALE库上的实验表明,提出方法的识别率要远高于直接采用特征脸方法。

Because current image compression schemes are mainly focused on DCT transformation,three level multiresolution subbands from reordered DCT coefficients for face image are acquired ahead of the using of eigenface method.Different results are achieved by applying the eigenface method on each single method.A method is proposed which recognizes the eigenface on subband S0 and subband S1 ,S2 and S3 .If this paper can get different results,more face images are input under certain threshold condition.The relationship of recognition time between the proposed method and the original method is analyzed.The experimental results on ORL and YALE database show that the recognition rate of the proposed method is higher than those of directly using the eigenface method.