Unsupervised Face Recognition Algorithm based on Fast Density Clustering Algorithm

Abstract

Most classic face recognition classification algorithms need to extract enough face images with class label information as training samples. However in most practical applications, face recognition based on supervised methods are incapable to deal with images without any label information. A novel unsupervised face recognition algorithm based on fast density clustering algorithm is proposed in this paper, which doesn't need sample images with class label information. Without any labelled images as examples, the designed method still get higher recognition rate compared with the same classifiers with labelled training sample. The main contributions of this paper include three aspects. Firstly, aiming at most current clustering algorithm has challenges as low clustering purity, parameter sensibility and cluster center manual determination, a fast density clustering algorithm (FDCA) with automatic cluster center determination (ACC) is proposed. Secondly, based on ACC-FDCA, an unsupervised face image recognition algorithm is designed. SSIM, CW-SSIM and PSNR are adopted to calculate face image similarity matrix. Finally, an online unsupervised face video recognition platform is developed based on brought up ACC-FDCA face recognition algorithm. Real life videos are recorded and recognized to testify the high performance of brought up method. We can conclude that classifiers using FDCA to get image samples label information for training could achieve higher recognition rate compared with the same classifiers trained with labelled image samples.