Razvoj refleksione optoelektronske metode za određivanje kolorimetrijskih vrednosti boje štampanih uzoraka

Abstract

U okviru ove doktorske disertacije je razvijena refleksiona optoelektronska metoda za određivanje kolorimetrijskih vrednosti boje štampanih uzoraka. Predložena metoda bazirana je na estimaciji reflektovanog spektra elektromagnetnog zračenja u vidljivom opsegu talasnih dužina. Na bazi metode dizajniran je senzorski sistem koji se sastoji od svetlosnih LED izvora na predajnoj i širokopojasnog silicijumskog fotodetektora na prijemnoj strani. Rezolucija A/D konvertora pri merenju intenziteta reflektovanih signala određena je sa 12 bita, dok je pokrivenost spektra LED izvorima ograničena na 6 tačaka, između kojih se vrši interpolacija. Ponovljivost sistema je ±0.27%. Razlike u boji ΔE00 koje predstavljaju grešku merenja, računate su u odnosu na referentnu spektrofotometrijsku metodu i nalaze se unutar granica tolerancije (0-5) za sve testirane uzorke. U tezi je takođe data i korekciona metoda bazirana na mašinskom učenju, koja je znatno redukovala ovu grešku ΔE00 na vrednost 3.

As part of this doctoral thesis a reflection-based optoelectronical method for determening colorimetric values of the color of printed samples was developed. The proposed method is based on the estimation of the reflected spectrum of electromagnetic radiation in the visible spectrum (of vawelenghts). On the basis of this method a sensing sustem was designed, consisting of LED sources on the emitting side and a wide range silicium photodetector on the recevieng side. Resolution of the spectral sensitivity of the sensing system is defined with 11 bits, while the coverage of the spectrum by the LED sources is limited to 6 points, between which interpolation was preformed. The repeatability of the system is +-27%. Diferences in color ΔEoo, which represent the measurement error, were calculated with the respect to the referent spectrophotometric method and are within the boundary of tolerance (0-5) for all tested samples. Furthermore in the thesis a corrective method, based on ANN, was presented, which significantly reduced the error to 3.