Development of an approach for image forgery detection using machine learning algorithms

Abstract

Detection of Image Forgery or Image Tampering have been a research area for several decades and emerged as a research paradigm because of the Internet, Online Platforms, Social Media and abundant use of digital images. There are various techniques or methods associated with the detection of tampering or forgery of images and failure rate is one of them that have an impact on the detection of image tampering. The present work is carried out in four phases. In the first phase, bicubic interpolation is implemented in the pre-processing stage to resize and preserve the quality of the images and k-fold cross validation is implemented to split the dataset into training and testing sets. In second phase, features are extracted using various deep learning models and applied to the most popular machine learning algorithms. In third phase fine-tuning is done on the best deep learning models with various hyper-parameters optimization based on the various performance metrics. In the last phase, decision fusion based approach is implemented on the basis of the best fine-tuned deep learning models. In the proposed work, experiments are performed on the benchmark datasets MICC-F220, Columbia and CoMoFoD with implementation of six popular machine learning algorithms on the extracted features from the Spatial Exploitation deep learning models, Lightweight based deep learning models and Residual based deep learning models. The proposed approach based on machine learning algorithms and deep learning based features improved the efficiency in terms of forgery detection rate and reduced the false positive rate and the results are validated on benchmark image forgery datasets and compared with the state-of-the-art approaches. newline