基于K-L交叉熵的岭回归人脸识别

岭回归人脸识别利用正则单形的顶点对每类人脸进行多元标记,通过投影实现高维人脸特征的降维。该算法首先提取人脸图像的局部二进制（LBP）直方图特征向量,通过主成分分析（PCA）和岭回归对该特征向量进行两次降维。识别阶段利用K-L交叉熵计算标记向量和投影后特征向量的相似性,根据熵值最小原则完成对测试样本的类别判断。实验选取ORL和YALE两个标准人脸库对算法进行测试,结果表明,K-L交叉熵测度比传统的欧氏距离测度获得更高的识别率。

Face recognition using ridge regression based on K-L cross entropy

Ridge regression for face recognition uses the vertices of a regular simplex to encode the multiple labels for each face,and maps the high-dimensional feature into a low-dimension subspace. This algorithm firstly extracts the feature vector of LBP histogram in a face image. Principal component analysis（PCA） and ridge regression are used successively to reduce the dimension twice. In face recognition stage, K-L cross entropy is utilized to calculate the similarity between the label vector and projected feature vector. The principle of minimum entropy can determine the category which the test sample belongs to. ORL and YALE face databases are selected to test the algorithm. Experimental results demonstrate that K-L cross entropy will get higher recognition rate than traditional Euclidean distance.