基于PCA和支持向量机的人脸识别

Face Recognition Based on PCA and Support Vector Machine

随着大数据和人工智能时代的来临,人脸识别已经成为一个非常重要的研究领域。在这项工作中,输入对象是一张张人脸的照片,我们可以将每张照片视为一个多维向量,该向量的分量代表图片某处的像素值,那么我们要处理的将是维数高达数万乃至数十万维的向量,这意味着非常巨大的运算量,即使如今计算机性能十分强大,然而我们要处理的一般是规模比较大的人脸数据库,如果我们对采集的人脸数据不加处理,必将导致时间和资源的浪费。本文基于主成分分析(Principal Component Analysis, PCA)和支持向量机(Support Vector Machine, SVM)实现人脸识别。PCA算法能够将高维数据简化成低维问题,简单、快速,且主成分之间相互正交,可消除原始数据成分间的影响,基于PCA算法的人脸识别技术能够在一定程度上去除光照、姿态、遮挡产生的噪音。使用核函数的SVM方法对于非线性问题具备良好的分类效果。该算法结合PCA和SVM方法,基于ORL人脸数据库先对训练数据进行PCA降维及特征提取,然后将降维后的数据输入到使用高斯核函数的SVM中进行训练,最后将测试数据输入训练好的模型中测试模型的准确率。实验证明该算法具有较高的识别率。

With the advent of the era of big data and artificial intelligence, face recognition has become a very important research field. In this work, the input object is a picture of a face. We can treat each picture as a multi-dimensional vector. The components of this vector represent the pixel value somewhere in the picture, so what we have to deal with will be the dimensionality, up to tens of thousands or even hundreds of thousands of dimensional vectors, which means a large number of calculations. Even though the computer performance is very powerful now, what we have to deal with is generally a relatively large-scale face database. If we analyze the face dataset untreated, it will inevitably lead to a waste of time and resources. This paper implements face recognition based on Principal Component Analysis (PCA) and Support Vector Machine (SVM). The PCA algorithm can simplify high-dimensional data into low-dimensional problems. It is simple, fast, and the principal components are orthogonal with each other, which eliminates the influence of the original data among variable components. The face recognition technology based on the PCA algorithm can remove noises caused by lighting, posture, and cover. The SVM method using the kernel function has a good classification effect for nonlinear problems. The algorithm combines PCA and SVM methods, based on the ORL face database, first performs PCA dimensionality reduction and feature extraction on the training data, and then inputs the reduction data into the SVM using the Gaussian kernel function for training, and finally inputs the test data into the trained model to test the accuracy of the model. Experiments indicate that the algorithm has a high recognition rate.