A Security-Preserving Ensemble Convolutional Neural Network Framework for Automated Forensic Image Evidence Classification

Abstract

Image-based evidence that is gathered on a variety of diverse and often unsecured sources in digital forensic investigations is being used more and more, necessitating the need to analyze it accurately, automatically and securely. This paper will present a security-saving ensemble convolutional neural network (CNN) model in automated classification of forensic image evidence. The proposed system will work on the images obtained in reality in digital forensic context, such as at the scene of a crime, on a confiscated device, and in a surveillance system where the lighting, noise, and complexity of the background, and the quality of a captured image may vary. The framework makes use of a collection of transfer-learning-trained CNN models to derive discriminative forensic features that are associated with texture patterns, color distributions, structural anomalies and object characteristics found in digital evidence. In an attempt to overcome the issue of data sensitivity and integrity that is demonstrated by forensic investigations, a security-preserving learning mechanism is added to reduce the exposure of data and reduce evidence reliability at the same time. Data augmentation methods are used to increase robustness, reduce overfitting as well as address the problem of class imbalance in forensic data. The suggested system has multi-class classification, which allows recognizing the different classes of forensic image evidences that have a similar visual look. The high accuracy of classification and high generalization results are experimentally proven on heterogeneous forensic databases. The findings show that the automated forensic image analysis using the CNN ensemble framework is a reliable, scalable and secure method of automated forensic analysis. The paper is a step toward a smart and safe digital forensic infrastructure, which will help to make informed and timely decision-making in the working with crimes