Face recognition using life-centroid distance between featured persistence diagrams

Face recognition plays an important role in identification of indivi-duals for security purposes. The state-of-the-art in face recognition includes techniques that involve machine learning in some form. In the recent years, persistent homology has emerged as a pivotal tool for topological data analysis with extensive application in diverse areas. In the current paper, a novel face recognition technique is proposed that is based on persistent homology for the first time in literature, to the best of our knowledge. A given human face image is preprocessed with local binary pattern (LBP) operator to extract a point cloud. The point cloud is subjected to persistent homology techniques to generate a simplicial complex which is used to construct a featured persistence diagram, a theoretical concept newly proposed in this work. The extent of similarity between two human face images is measured by computing the life-centroid distance between their featured persistence diagrams, again another theoretical concept newly proposed here. Experimentation on several datasets have produced promising results identifying human faces with an accuracy varying around 85%.