Abstract: In order to solve the problem that POEM （Pattern of Oriented Edge Magnitudes） ignores the pixel among blocks, a new face recognition method based on scaled-block LBP of oriented edge magnitude is proposed. Firstly, the orientation and magnitude of a face are extracted by gradient operator. The magnitude is accumulated on the basis of orientation with the same quantitative value. Secondly, divided histograms are extracted from each accumulated magnitude encoded by the scaled-block LBP operator. All the histograms are cascaded as the final recognition feature. Finally, the dimensionality reduction method of WPCA is used to improve the effectiveness of the proposed algorithm. Experimental result shows that the proposed algorithm is more robust than other comparing algorithms and can achieve good recognition performance with short dimension after feature dimensionality reduction method used.

Key words: face recognition, gradient, scaled-block Local Binary Pattern（LBP）, feature extraction, histogram, dimensionality reduction

摘要： 针对方向边缘幅值模式（POEM）忽略了块与块之间的像素问题，提出一种基于方向边缘幅值的尺度块LBP人脸识别方法。该方法首先用梯度算子提取出人脸的方向图和幅值图，将具有相同量化方向上的幅值累加，然后利用尺度块LBP算子提取每幅累加幅值图的分块直方图特征，并将所有直方图特征串联起来作为最终的识别特征，最后采用WPCA降维方法提高算法的有效性。实验结果表明，该算法的鲁棒性高于其他对比算法，运用降维处理后能以较低的特征维数达到良好的识别性能。

关键词: 人脸识别, 梯度, 尺度块局部二值模式（LBP）, 特征提取, 直方图, 降维