This study introduces a novel algorithm rooted in spectral graph theory for the reconstruction of occluded facial images, with the aim of enhancing face recognition accuracy. The proposed method, Face Reconstruction using Graph Fourier Transform (FRGF...
This study introduces a novel algorithm rooted in spectral graph theory for the reconstruction of occluded facial images, with the aim of enhancing face recognition accuracy. The proposed method, Face Reconstruction using Graph Fourier Transform (FRGFT) comprises three key stages: fi rstly, construct a graph representation of the image where each pixel is a node, and edges between nodes represent pixel adjacency with weights that capture both intensity and spatial relationships; secondly, identify occluded parts of the image using a graph-based segmentation technique to update the occlusion mask; and thirdly, reconstruct the original image from the occluded version using the Laplacian matrix of the graph and the updated occlusion mask. The whole process lies on the Graph Laplacian (GL) technique, which draws the concepts from graphical models and the Laplace equation, proving to be eff ective in the high-quality restoration of faces that are damaged or occluded. This paper establishes the connection between GL and the traditional Fourier transform methods. To assess the utility of the GLbased restoration, completed facial images processed through the algorithm are subjected to face recognition testing. The eff ectiveness of the proposed approach is validated on two datasets: extended AR and CelebA face dataset, which have been signifi cantly noticeable as our proposed approach achieves 98.5% accuracy as compared with the existing method.