Evaluation of Neural Style Transfer Results of Various Pattern Images Using Feature Similarity Index

Main Article Content

Vanessa Metayani
Swat Lie Liliawati
Andreas Widjaja

Abstract

This research was conducted by applying the Neural Style Transfer method to various sets of content and style images, and then calculating the FSIM value for each pair of result and original images. Analysis was done on factors such as art style complexity, resolution and other special characteristics such as colour and texture that can affect the FSIM value. The purpose of this research is to identify whether there are factors that affect FSIM performance such as art style complexity, resolution, or other special characteristics such as colour and texture. This research is expected to be able to help artists who want to change the art style with Neural Style Transfer but still maintain the originality of the image and still be recognised by evaluating the results using FSIM and help artists to develop and produce artistic digital artworks with good quality. The results show that varying FSIM values can depend on the complexity of the art style and image resolution. Simple art styles and high-resolution images tend to produce higher FSIM values, indicating that the image structure is easily preserved. As long as the resolution and colours or textures do not change the main structure, the FSIM results will not decrease significantly. To support the research analysis, the Analysis of Variance (ANOVA) statistical test was used to measure the significance of the effect of complexity and resolution on FSIM and the Cronbach’s Alpha test to test the reliability of the general public and expert surveys. Based on the ANOVA statistical test results, there was insufficient evidence to reject the null hypothesis, so complexity and resolution did not have a significant influence on FSIM. From the Cronbach’s Alpha test results, the public assessment survey received a result of 0.94 and 0.91 for the expert assessment survey. These results indicate that the results from the surveys are reliable as subjective data for the research.

Downloads

Download data is not yet available.

Article Details

How to Cite
[1]
V. Metayani, S. L. Liliawati, and A. . Widjaja, “Evaluation of Neural Style Transfer Results of Various Pattern Images Using Feature Similarity Index”, JuTISI, vol. 10, no. 2, pp. 379 –, Aug. 2024.
Section
Articles