Mathematical methods to estimate image blur and recognize faces in the system of automatic conference participant registration

To enhance the efficiency of the developed system of automatic participant registration in the conference room, the up-to-date mathematical methods were presented for estimation of digital image blur and sifting out the low-quality images prior to the face recognition stage. To increase the rate of processing and reliability of the methods for determination of image quality at experimental verification, several variants of segmentation of frame part were realized: entire image, face area, and $200\times200$ pixels face area. The proposed procedure of preliminary search of a face area within the frame and estimation of its blur on the basis of statistical analysis of the pixel brightness coefficients enabled one to determine 94 % of undistorted frames obtained at automatic registration of the conference participants in the intelligent room.

Presented by the member of Editorial Board: B. T. Polyak