Abstract

The number of images on the internet is astronomical and only continue to grow. It is necessary to devise some way to categorize and determine information on each image. Some images are not even real to make matters more complicated. The purpose of this study was to use machine learning to categorize images as being real or fantastical. Screenshots from video games and professionally taken photographs will be compared to determine the if the program can correctly categorize each. Neural networks built and trained on various sizes of images allow us to determine the minimum size of image which can be accurately classified. Four tests were run which were split into two categories. The primary test and the secondary tests. The primary test was to determine if the AI is capable of identifying real and fake images. This was confirmed and thus the secondary tests were run. The first secondary test was to find the optimum image resolution which resulted in smaller images being the best. The second was to identify the optimum training time. Training time is measured in epochs which represents one propagation of networks. The optimum time was 20 to 25 epochs. The third was to measure the impact image size had on training time. The processing time followed an exponential growth when image size increased.

Advisor

Montelione, Thomas

Department

Computer Science

Publication Date

2023

Degree Granted

Bachelor of Arts

Document Type

Senior Independent Study Thesis

Download

Share

COinS
 

© Copyright 2023 Ethan O. Samangy