Cancelable face recognition using phase retrieval and complex principal component analysis network

Abstract

Considering the necessity of sensitive information protection in face image, a cancelable template generation model for multimodal face images is proposed in this paper. Firstly, the visual meaningful face images are transformed into phase-only functions through phase retrieval in gyrator domain. Then random projection and chaotic-based mask are constituted into modulation for achieving revocability and distinguishability. The interim results are mapped to a higher-dimensional space using random Fourier features. Followed by two-stage complex-valued principal component analysis, the convolutional filters are learned efficiently. Together with binary hashing and decimal coding, local histogram features are obtained and forwarded to SVM for training and recognition. Experiments performed on three publicly multimodal datasets demonstrate that the proposed algorithm can obtain higher accuracy, precision, recall and F-score in comparison with some existing algorithms while the templates are non-invertible and easy to revoke.