A Comparative Analysis of Local Binary Pattern(LBP) Variants for Image Tamper Detection

Abstract

Abstract The proliferation of image tampering in the digital age poses a significant challenge to the authenticity and integrity of visual content. This study presents an approach for detecting image tampering using the Local Binary Pattern (LBP) techniques in conjunction with Convolutional Neural Network (CNN). LBP is a powerful image texture descriptor. The LBP method is employed to extract robust and discriminative features by capturing local texture and intensity patterns from tampered images. These features are then input into a CNN architecture, which is trained using 5-fold cross-validation to ensure generalization and prevent overfitting. A comprehensive benchmark image dataset CASIA-2.0 comprising of 7,541 authentic and 5,124 tampered images is utilized to evaluate the proposed method, and performance evaluation metrics, including accuracy, and confusion matrix, are employed to assess the effectiveness of the system. Experimental results demonstrate the efficiency of the proposed approach over existing state-of-the-art methods, achieving high accuracy in detecting image tampering. A comparative analysis of four types of LBP variants is presented in this work. With circular LBP, Rotation-Invariant LBP, Default, and Uniform LBP we achieved an accuracy of 68%, 72%, 84%, and 96% respectively. This research has significant implications in various domains, including forensic investigations, journalism, and image integrity verification, as it addresses the challenges posed by image tampering, enhancing trust and confidence in digital visual content by ensuring its authenticity and reliability.