ADAPTIVE DATA FUSION PREDICTION FOR RELIABLE AND UNCERTAINTY-AWARE IMAGE FORGERY PREDICTION.
Keywords:
Adaptive data fusion, Confidence estimation, Image forgery Prediction, Image tampering, Multi-source prediction, Uncertainty modeling, weighted decision-makingAbstract
The prediction of image tampering has become an important problem because of increased manipulation and forgery of images in terms of copy-move forgery, splicing, deep fakes, etc. The existing methods mainly use one source for feature extraction, decision-making, or prediction without any adaptation to uncertainty and complexity in image scenario. In order to solve this problem, an Adaptive Data Fusion Prediction algorithm has been designed to provide reliable prediction about image tampering. This ADFP algorithm uses the selection of proper data source among various data sources, allocation of proper weight according to the relevancy and reliability of the data sources and finally the prediction according to confidence level. The main techniques used in this research include data fusion from various sources, adaptive weight allocation, and decision-making by taking into account the confidence level and modeling of uncertainties. ADFP has been tested for image tampering prediction and authentication using the CASIA 2.0 database. It was found that ADFP achieves an accuracy rate of 98.90%.
