Faces selection in images using the spectral graph theory and constraints
Alexei Zakharov, A. E. Barinov, Arkady L. Zhiznyakov · 2017
Selection of human faces in images is a relevant task in the sphere of computer vision. The design of the clusterization of the features for faces identification in the images is considered. To allocate faces, we used the approach based on the spectral theory of graphs. The characteristics of graphs are formed using a random walk of the graph. To get some information about the possible cluster structure, an eigenvector is used corresponding to the minimum value of the commute time matrix. The prior information about the proportions of the head is used for imposing constraints on the clusterization results. The network of proportions is used as the prior information. In this paper the algorithm is described, the research is fulfilled; the results of the experiments are shown. The advantage of the developed algorithm is shown comparing to the algorithm of the normalized cuts on the graphs.