Transform Domain Techniques combined with LBP for Tampered Image Detection using Machine Learning

Detection of tampered images is very important for ensuring the authenticity of visual information in the domain of digital image forensics. In the context of machine learning, this research proposes methods for detecting tampered images using transform domain techniques such as discrete cosine transform (DCT), discrete wavelet transform (DWT). These techniques are combined with local binary pattern (LBP), which is effective at capturing local information. Dimensionality reduction is achieved with the help of principal component analysis (PCA) to reduce the number of features and analyzed the performance in terms of accuracy and time. Experiments have been conducted on CASIA 2.0 image tampering detection dataset which consists copy-move and splicing forgery images. Various classifiers, including Support Vector Machine (SVM), Random Forest (RF), XG Boost, and Light GBM, are used to train machine learning models on the retrieved features. Based on the accuracy, precision, recall, fl-score, and execution time of these classifiers, the effectiveness of tamper detection is evaluated.