基于邻域相关度和LBP算子的人脸图像识别

针对传统局部二值模式(local binary pattern,LBP)算子提取的图像纹理特征不完整、不能全面刻画人脸局部特征的问题,提出一种基于邻域相关度的改进LBP算子.该算子首先计算窗口内每个像素点的邻域相关度;其次利用邻域相关度的均值和方差构造新的NC_LBP算子,进而提取图像局部直方图特征,作为人脸识别的依据;最后利用Chi平方统计法计算直方图的不相似度,并用KNN算法进行分类.仿真实验表明,改进NC_LBP算子在ORL,JAFFE和YALE人脸数据库的识别中效果较好,特征区分度明显,识别准确率较高.

Face Recognition Method Based on NeighborhoodCorrelation Modified LBP Operator

For the problem that the grain feature extracted by the traditional LBP operator is not complete and cannot fully represent the local feature of face, an improved LBP operator was proposed based on neighborhood correlation. The operator first calculated the neighborhood correlation of each pixel within the window. And then a new NC_LBP operator was constructed by the neighborhood correlation mean and variance; furthermore, the image local histogram feature was extracted as the basis for face recognition; finally, the Chi square statistic method was used to calculate the histogram dissimilarity, and the classification was performed with KNN algorithm. Simulation results show that the proposed improved NC_LBP operator has achieved good recognition results in the ORL, JAFFE and YALE face databases, and the discrimination of features is obvious. The recognition accuracy is greatly improved.