Online Program

Use of Error-Level Analysis and Convolutional Neural Networks to Identify Altered Images (305317)

Keywords: Image forensics, Deep Learning, Error Level Analysis, Convolutional Neural Network (CNN), Neural network

With the emergence of SARS-CoV-2, traditional verification of identity has shifted from hard copies of documents to soft copies such as photographs or scans. In this new digitized world, these images are open to a wide variety of potential image manipulation using commonly available image processing software. Digital forensic techniques used to detect these image modifications for illegal purposes are more in need than ever before. The objective of this paper is to develop an image manipulation detection approach that when applied to documentation can correctly identify the modified image. There are two major factors to this technique, the first uses Error Level Analysis for feature reduction of an image, and the second using Convolutional Neural Networks for feature extraction and classification. Error level analysis is the analysis of compression artifacts in lossy compressed images. This technique is used to detect foreign content by analyzing the compression ratio throughout an image. We benchmark our proposed architectures against previous work and show improvements performance via expanding the image features used and changing the architecture