Image Manipulation Detection and Classification Using SIFT and SURF with Support Vector Machine Classifier

Enhance digital trust and integrity by mastering image forgery detection in machine learning. Uncover the hidden truth behind visuals, ensuring authenticity in a world inundated with digital content. Various image manipulation detection machines have been innovated in the past, but these models face limitations like demanding domain expertise, hyperparameter tuning, time-consuming processes, and overlapping issues. Initially employ the histogram equalization to enhance dataset images, reducing unwanted noise and generating synthetic data for class balance. The pre-processed images are then fed into the model for ongoing analysis. In this paper, propose a Scale-Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF)-based SVM for image manipulation detection and classification, achieving outstanding performance metrics on the CASIA 1.0 dataset: 0.96 accuracy, 0.8120 IOU score, 0.7710 MCC, 0.8850 F1 score, 0.98 precision, and 0.92 sensitivity. This model surpasses the capabilities of existing models, including the Two-Stream Boundary-aware Network, Discrete Cosine Transformation (DCT)-based multitask learning network, and Multilevel Local Binary Pattern Texture Descriptor.