Geometry analysis based on the position of control points of the parametric model

U ovom radu prikazan je postupak analize geometrije pomoću kontrolnih točaka parametarskog modela u programskom paketu Matlab R2021a. Referentni kontrolni poligon oblikovan je uz pomoć poznate, u idealnim uvjetima skenirane, geometrije lica. Proračunom prosječnog odstupanja referentnog kontrolnog poligona od svakog kontrolnog poligona skenirane geometrije definirana je krivulja prosječnog odstupanja na osnovu koje su formirani modeli neuronskih mreža i doneseni zaključci o analizi geometrije. Cilj rada je utvrditi način prepoznavanja geometrije formiranjem stabilnog matematičkog modela te postići automatizaciju procesa korištenjem plitkih neuronskih mreža. Time se postiže značajno smanjuje broja parametara potrebnih za analizu geometrije u odnosu na analizu dubokim neuronskim mrežama.This paper presents the procedure of geometry analysis using control points of the parametric model within the Matlab R2021a software package. As an initial step, the scanned geometry was written in matrix form, after which the BSpline surface was approximated based on previously calculated by a matrix scaled notation of control points. The reference control polygon was shaped using the familiar, ideally scanned, facial geometry. By calculating the average deviation of the reference control polygon from each control polygon of the scanned geometry, the curve of the average deviation was defined, based on which neural network models were formed and conclusions on geometry analysis were made. The aim of this paper was to determine the way of recognizing geometry by forming a stable mathematical model and to achieve process automation using shallow neural networks. This significantly reduces the number of parameters required for geometry analysis compared to deep neural network analysis