To reduce the dimensionality of the Gabor feature, this paper presents a novel approach called Gabor Phase-based Texture Representation(GPTR) for face recognition. GPTR is characterized by using the Generalized Gaussian Density(GGD) to model the Gabor phase distribution. The estimated model parameters serve as texture representation, which is then analyzed by Locality Preserving Projections(LPP) to make them more discriminative. Experimental results on FERET and Yale databases show that GPTR is superior to traditional Gabor features in terms of recognition accuracy.%为降低Gabor特征的维数,提出一种基于Gabor相位的纹理表征(GPTR)方法,将其应用于人脸识别.GPTR采用广义高斯分布(GGD)拟合Gabor相位的分布,将拟合的GGD参数作为纹理特征.采用保局投影方法对纹理特征向量进行子空间分析,进一步降低其维数并增强鉴别力.在FERET及Yale人脸库上的实验结果表明,相比传统的Gabor幅值特征,GPTR具有更高的人脸识别准确率.
展开▼
