基于最值平均的人脸识别LBP算法

FACE RECOGNITION LBP ALGORITHM BASED ON THE MOST VALUE AVERAGE

为解决局部二值模式(LBP)在提取人脸特征时容易缺失部分细节特征问题,提出一种基于最值平均的改进LBP算法。该方法针对3×3模板,计算其九个像素方差。若方差在限定范围内,取中心像素周围八个像素的最大值与最小值的平均值作为阈值进行比较,避免因中心像素值偏大或偏小以致湮没细节的现象,从而保留更多的局部细节;否则取九个像素的中值作为阈值进行比较,减少噪声。通过主成分分析法(PCA)降低高维特征维数,利用K近邻算法(KNN)完成人脸特征分类。实验结果表明,该方法有很好的识别效果。

In order to solve the problem that local binary pattern (LBP) was easy to lose some details when extracting facial features, we proposed an improved LBP algorithm based on the most value average. We calculated the variance of nine pixels for a 3×3 template. If the variance was within a limited range, the average value of the maximum and minimum values of eight pixels around the central pixel was taken as the threshold value to compare, so as to avoid the phenomenon of annihilating details due to the large or small value of the central pixel, thus retaining more local details. Otherwise, the median values of nine pixels were used as thresholds to reduce noise. Principal component analysis (PCA) was used to reduce the dimension of high-dimensional features, and K-nearest neighbor algorithm (KNN) was used to complete face feature classification. The experimental results show that the method has a good recognition effect.