AAI_2025_Capstone_Chronicles_Combined

5 Pixel value statistics are one of the simplest methodologies used when analyzing image datasets. For this section, I extracted a subset of 10000 images and calculated the distribution of individual red, green and blue channels, contrast and brightness. If there were to be a clear difference between the distributions of the real and fake images, it could mean that the quality of the fake images is not good enough and it would be too easy for a model to categorize them. Figure3 Histogram of the pixel value distribution, Real vs Fake images

As it can be seen, there are some slight differences between the Real and the Fake sets, but it is too small to have any meaningful impact on the final results. For the second stage of my data exploration I will be using Image Quality Assessment (IQA) techniques. These are different algorithms and neural models that try to measure how “good” or “distorted” a image looks. This should be dependent on how good a real image was taken or the quality of a generated fake, and will be a very interesting variable to study in the results of the research. To perform this analysis, I will

362