论文部分内容阅读
目的本文针对基于最小均方差准则的主成分分析算法(如2DPCA-L_2(two-dimensional PCA with L_2-norm)算法和2DPCA-L_1(two-dimensional PCA with L_1-norm)算法)对外点敏感、识别率低的问题,结合信息论中的最大相关熵准则,提出了一种基于最大相关熵准则的2DPCA(2DPCA-MCC)。方法 2DPCA-MCC算法采用最大相关熵表示目标函数通过半二次优化技术解决相关熵问题,降低了外点在目标函数评价中的贡献,从而提高了算法的鲁棒性和识别精度。结果通过对比2DPCA4VlCC算法和2DPCA-L_2、2DPCA-L_1在ORL人脸数据库上的识别效果,表明了2DPCA-MCC算法的识别率比2维主成分分析算法的识别率最低提高了近10%,最高提高了近30%。结论提出了一种基于最大相关熵的2DPCA算法,通过半二次优化技术解决非线性优化问题,实验结果表明,本算法能够较好地解决外点问题显著提高识别精度,适用于解决人脸识别中的外点问题。
Objective In this paper, the principal component analysis algorithm based on the least mean square error criterion, such as 2DPCA-L_2 (two-dimensional PCA with L_2-norm) algorithm and 2DPCA-L_1 Based on the maximum entropy criterion in information theory, a 2DPCA (2DPCA-MCC) based on the maximum entropy criterion is proposed. Methods The 2DPCA-MCC algorithm uses the maximum entropy to represent that the objective function solves the entropy problem through the semi-quadratic optimization technique and reduces the contribution of the outer point in the evaluation of the objective function, thus improving the robustness and recognition accuracy of the algorithm. Results By comparing 2DPCA4VlCC algorithm and 2DPCA-L_2,2DPCA-L_1 recognition results on ORL face database, it shows that the recognition rate of 2DPCA-MCC algorithm is the lowest than that of 2-D principal component analysis algorithm. Increased by nearly 30%. Conclusion A 2DPCA algorithm based on the maximum entropy is proposed. The semi-quadratic optimization technique is used to solve the nonlinear optimization problem. The experimental results show that this algorithm can solve the out-point problem and improve the recognition accuracy significantly, and is suitable for the face recognition In the outer point of the problem.