基于空间-光谱信息融合的Gabor PCA高光谱人脸识别算法研究

RESEARCH ON GABOR PCA HYPERSPECTRAL FACE RECOGNITION ALGORITHM BASED ON SPATIAL-SPECTRAL INFORMATION FUSION

高光谱成像可以通过捕获更多的生物特征量来增加面部特征,为人脸识别提供了新的机会,然而其数据维度高,低信噪比等特点也为人脸识别带来了新的挑战。研究了空间-光谱信息融合结合Gabor PCA算法提取人脸特征,并将所得特征放到KNN分类器进行分类。空间-光谱信息融合在传统的波段融合上加入局部空间信息,有效地去除了随机传感器噪声,并降低了数据维度;Gabor PCA通过Gabor小波对融合后的人脸图像进行滤波;运用主成分分析法(PCA)对Gabor提取的特征进行数据处理,除去冗余数据。实验在CMU高光谱人脸数据库上进行,实验结果表明该方法能很好地处理高维度数据,对姿态鲁棒性较高,识别率明显优于基于特征波段融合的2D PCA方法,识别率提高了20.9%。

Hyperspectral imaging can increase facial features by capturing more biometric features,providing new opportunities for face recognition. However, its high data dimension and low signal-to-noise ratio also pose new challenges for face recognition. This paper studied spatial-spectral information fusion and Gabor PCA algorithm to extract facial features,and put the resulting features into the KNN classifier for classification. Spatial-spectral information fusion added local spatial information to the traditional band fusion,effectively removing the random sensor noise and reducing the data dimension. Gabor PCA first filtered the face image after fusion by Gabor wavelet,and then used Principal Component Analysis（ PCA） to process the features of Gabor extraction to remove redundant data. Experiments were performed on a CMU hyperspectral face database. The experimental results showed that this method can handle highdimensional data very well,and had higher robustness to gestures. The recognition rate was obviously better than the 2 D PCA method based on feature band fusion. The recognition rate was improved by 20. 9%.