基于正交余弦变换域概率主成分分析的嵌入隐马尔可夫人脸识别模型

提出并证明了概率主成分分析作用于正交余弦变换(DCT)域与作用于空域所获得的结果相同.利用DCT变换的快速压缩性能和概率主成分分析的软降维性能来稳定地获取和表示人脸的局部特征,并把得到的特征向量作为嵌入隐马尔可夫模型的观察向量,建立了基于DCT域概率主成分分析的嵌入隐马尔可夫人脸识别模型.该模型应用于不同表情和光照下的人脸识别,取得了较好的实验结果.

PPCA in DCT Domain for Face Recognition Using Embedded HMM

This paper presented that probabilistic principal component analysis(PPCA) can be directly implemented in discrete cosine transform(DCT) domain and the results are exactly same as the one obtained from the spatial domain.PPCA in DCT domain(D-PPCA) can take the advantage of both rapid compression of DCT and soft dimension reduction of PPCA to extract and describe the local facial features robustly.Then D-PPCA is(adopted) to extract facial feature vectors as the observation vectors fed into embedded hidden Markov model(EHMM) and build the D-PPCA EHMM model for face recognition.The proposed model is performed extensively on pose-invariant and illumination-invariant face recognition,and achieves better recognition performance than other methods.