An efficient supervised learning approach for splicing forgery detection with low classification error rates is proposed in this work. Existing Literature is analysed to produce the research gap and and PCA is used for feature extraction to make the detection process fast and intelligent. As PCA is the process of dimension reduction without eliminating the significant information from the image. Canny edge detection is used to detect strong edges in the image. . Back propagation neural networks Model for classification is trained by feeding dataset images. A benchmark dataset CASIA V2 is used for evaluating performance of proposed algorithm. The images are then tested for authenticity, whether the image is forged or authentic. Then the performance is evaluated by using parameters like precision, Recall and Mean Square Error. Proposed approach is able to increase the accuracy with low classification error rate while the existing work takes the optimal value to get their required result. Simulation results for the proposed algorithm are presented.
Key words: Neural network, image forgery detection, dataset, edge detection, feature extraction, authenticity
|
