Multilayer Perceptron Network to Detect Fraud in Digital Images

Abstract

 The major challenge of data authenticity is how to check for image fraud, which creates a
huge problem for the credibility of visual media. In this paper, we propose a method to investigate the
performance of a Multilayer Perceptron (MLP) to extract the fraud images, this network is a class of
supervised Artificial Neural Network (ANN. The proposal model applies MLP model to allocate extracted
image features in order to distinguish them between real and modified contents. The examined features are
included within statistical matrices, analysis of histogram space, and possible inequality that may arise
during modifications. The proposed MLP was trained with dataset that contains both real and fraudulent
images, thus allowing the model to extract knowledge from the original patterns that differentiate between
those two classes. The model's performance was validated with several metrics, including accuracy,
precision, and computational cost. Furthermore, this paper presents comparisons against traditional
methods that were examined in the procedure. The finding of this work enhances the model with improved
image fraud detection by showcasing the capabilities of MLPs within 162.59 seconds to 86% detection,
while the base algorithm in 205.92 seconds succeeded in recognizing 82%.