数据增广下的人脸识别研究

随着计算机技术的发展和应用,人脸识别技术以其具有的非强制性、非接触性、并发性等优势得到了越来越广泛的应用。大规模数据是提高基于深度学习人脸识别准确率的关键因素,但往往数据不易获得,并且存在训练数据缺乏测试数据样本的情况,如模糊、亮度失真和腐蚀感画质样本等。针对训练数据缺乏问题,提出了利用滤波、亮度调节和腐蚀操作3种传统图像处理方法10种增广方式增加数据量和数据的多样性,进而提高识别算法的性能。将原始数据和增广数据作为训练数据训练模型,选择从不同地方拍摄的视频上截取的人脸图像组成了四个测试集,实验结果表明,增广数据与测试集样本存在一致性时,增广方式对提升识别性能都有一定的效果,其中最好的效果是对图像整体调亮时在一个测试集上的识别率提高了4.02%。

Research on Face Recognition Based on Data Augmentation

With the development and application of computer technology,face recognition technology has been more and more widely used because of its non-mandatory,non-contact,concurrency and other advantages.Large-scale data is the key factor to improve the accuracy of face recognition based on deep learning.However,it is often difficult to obtain the data,and there is a lack of test data samples for training data,such as blur,brightness distortion and corrosion quality samples.In order to solve the problem of lack of training data,three traditional image processing methods,filtering,brightness adjustment and eroding operation,are proposed to increase the amount of data and the diversity of data in 10 ways,so as to improve the performance of the recognition algorithm.The original data and augmented data are used as training data training models,and the face images intercepted from different places are selected to form four test sets.The experiment shows that when the augmented data are consistent with the test set samples,the augmented method has a certain effect on improving the recognition performance,among which the best effect is that the recognition rate on a test set is increased by 4.02%when the image is brightened as a whole.