Machine Learning for Face Recognition in Shallow Data

Face recognition has been widely studied in the artificial intelligence field. Especially, with the increasing availability of machine learning algorithms such as deep learning, most existing research uses massive datasets to facilitate better recognition performance. However, it is time-consuming and labor-intensive to collect and annotate a large face dataset. On the contrary, with a small dataset, some of the machine learning algorithms may fail to perform the recognition task. In order to study the performances of different machine learning algorithms, this work compares the recognition performances of seven widely used machine learning algorithms on three different face datasets. With shallow data, the experimental results reveal that the best recognition accuracy on different datasets can be achieved by different algorithms. Specifically, the best performance of 66.41% on the Extended Yale B dataset is achieved by the logistic regression algorithm, the LDA algorithm achieves the best performance of 97.5% on the Olivetti face dataset, while the random forest algorithm obtains the best accuracy of 10.31% on the LFW face dataset, with five shallow training samples. The obtained results can be used to select appropriate ML algorithms to recognize faces with shallow data.