Human head detection algorithm on images using graph embedding in the vector space

The work solves the problem of human head detection on the images based on the structural approach. Development of the graphs comparison algorithm by their embedding in the vector subspace is considered. Special points are obtained on the images. These points are clusterized to select characteristic regions of the human's face. Centers of clusters mass are further used as graph points. The graph is built using the Delaunay triangulation and the obtained points. Graphs are projected into the vector subspace using the Young-Householder decomposition. Graphs are characterized by values of the curvature that are calculated using differential geometry. By comparing these values with Hausdorff metric, we obtain a distance matrix that reports whether the candidate image belongs to any particular group. Characteristics of the compared images are prepared at the stage of initialization. The results of the experiments are presented.