Investigation of image forgery via adaptive CNN and enhanced error level analysis
TL;DR
This study proposes a new method that integrates Enhanced Error Level Analysis (EELA) with an adaptive Convolutional Neural Network model to improve the detection of forged regions in images to improve the detection of forged regions in images.
Abstract
Detecting image forgery is a significant challenge in digital image forensics, especially given the increasing use of advanced image editing software. This study proposes a new method that integrates Enhanced Error Level Analysis (EELA) with an adaptive Convolutional Neural Network (CNN) model to improve the detection of forged regions in images. Unlike traditional methods such as standard ELA or Residual Pixel Analysis, our approach uses EELA to examine compression artifacts in both original and forged images. Enhanced ELA calculates and displays error-level variations within the image at the pixel level. These features are then used to extract spatial features through an adaptive CNN training model to discriminate tampered regions. The proposed method was thoroughly evaluated on image benchmarks, including CASIA v1.0, CASIA v2.0, the Columbia Image Splicing Detection Dataset, and DEFACTO/IMD. The technique shows strong potential for identifying splicing forgeries. The proposed framework demonstrated superior results and greater resilience to image forgery attacks.