A Face Hashing Algorithm using Mutual Information and Feature Fusion

In this paper, we implement a face-hashing algorithm based on feature fusion and Gabor feature extraction, using conditional mutual information. The proposed method comprises three components: feature extraction, feature discretization and key generation. During the feature extraction stage, global features (PCA-transformed), local features, and a set of informative and non-redundant Gabor features selected by conditional mutual information (CMI) from face images, are used to produce new fused feature sets as input feature vectors of kernel generalized discriminant analysis (GDA) in the unitary space, for further feature enhancement. Then, in the feature discretization stage, a discretization process is introduced to generate a stable binary string from the fused feature vectors. Finally, the stable binary string can be renewed and protected by a helper data schema (HDS), thus, a user's privacy can be preserved.