增强LGBP融合广义平均神经网络的鲁棒人脸识别研究

针对人脸识别中由于人脸表情、姿态、尺度等变化而影响识别效果的问题,提出一种融合增强局部Gabor二值模式和广义平均神经网络方法。首先,对归一化的人脸图像进行多方向、多尺度Gabor小波滤波,将生成的Gabor幅值模式分成多个子区域;然后,利用增强局部二值模式对每个子区域进行操作形成特征向量;最后,利用广义神经网络进行识别。在ORL、FERET及LFW上的实验利用错误接受率、错误拒绝率等错误率和时间复杂度评估方法的有效性。实验结果表明,相比其它几种较新的方法,取得了更好的识别效果。

The Research of Face Recognition Based on Fusion of Enhanced LGBP and GNNN

For the issue that the recognition performance of face recognition is affected by the facial expression, pose and scale, a method that use Gabor wavelet and generalized neural network for face recognition is proposed. The normalized face image in multi direction, multi-scale Gabor wavelet filter, the amplitude of Gabor pattern generation is divided into a plurality of sub region; then the enhanced local two value model for each sub region operation to form a feature vector; finally, by using the generalized neural network recognition. Experiments on ORL, FERET and LFW has comparied proposed method. Experimental results show that proposed method has better recognition efficiency than several other advanced methods.