Quasi Non-Negative Quaternion Matrix Factorization with Application to Color Face Recognition

To address the non-negativity dropout problem of quaternion models, a novel quasi non-negative matrix factorization (QNQMF) model is presented for color image processing. implement QNQMF, projected gradient algorithm and alternating direction method multipliers are proposed via formulating QNQMF as non-convex constraint optimization problems. Some properties algorithms studied. The numerical experiments on reconstruction show that these encoded perform better than red, green blue channels. Furthermore, we apply to face recognition. Numerical results indicate accuracy rate recognition channels well single channel gray level images same data, when large facial expressions shooting angle variations presented.