Skoči na glavni sadržaj

Pregledni rad

https://doi.org/10.19279/TVZ.PD.2019-7-3-05

AUTOMATICALLY COLORIZE BLACK AND WHITE PHOTOS USING ARTIFICIAL INTELLIGENCE ALGORITHMS

Ivan Šlogar ; Tehničko veleučilište u Zagrebu, Zagreb, Hrvatska, student
Miroslav Slamić ; Tehničko veleučilište u Zagrebu, Zagreb, Hrvatska


Puni tekst: hrvatski pdf 2.990 Kb

str. 188-194

preuzimanja: 437

citiraj


Sažetak

This paper has used practical examples to demonstrate the method for automatic colorization of black and white photographs based on deep learning algorithms. With the advancements accomplished in the field of automatic colorization of black and white photographs, the highlight was on the latest method demonstrated in the literature and implemented as a Web service. The paper aims, among other things, to highlight the success of the method itself against existing methods and the innovations it brings. The purpose of developing new methods is to develop improved algorithms that can color black and white photos so well that people cannot distinguish between the original and the „fake“, increasing the fooled percentage to 50%. It is important to point out that thse algorithms do aim to color the photo to look same as the original, instead they aim to color realistically enough to fool the viewer.In this paper, the performance tests of the forementioned algorithm were made on personal photographs. Along with the conducted research, histograms of original and colored photographs were used in order to obtain more information about th method's success. Displayed results of the conducted research were made on two sets of photographs, one originating from Image-net database and the other consisting of personal photographs.

Ključne riječi

convolutional neural networks; RGB; Lab; automatically colorizing photos

Hrčak ID:

230594

URI

https://hrcak.srce.hr/230594

Datum izdavanja:

2.12.2019.

Podaci na drugim jezicima: hrvatski

Posjeta: 1.546 *