SWT Based heterogeneous features to detect Spliced images

Today, unauthorized users intercepts images and modify those using splicing techniques which cannot be perceived by human eyes. This has become a common problem in human lives. It necessitates to design a robust expert system to validate the image authenticity considering the advantage that the splicing introduces some amount of distortion that can be used to detect forged tampered images. The proposed method considers both the statistical features (SF) and the textural features (TF) from upper level bands of stationary wavelet transform (SWT) using HAAR mother wavelet decomposed to first level. The frequency domain features are extracted in multi scale form sub bands of SWT. Markov model (MM) based SF’s are extracted from bands other than the low frequency band to obtain the transition probability matrices which are used then to extract the Haralic’s TF resulting from gray level co-occurrence matrices (GLCM). All the features are then concatenated to form an input vector for the classifier. The combination of TFs and SFs for classification produces significant results when used with Support Vector machine (SVM). The system provides a robust performance for different attacks. The expert systems can be tested for various components including Cb, Cr and mean of Cb-Cr with varying threshold with T=2, 3 and 4. Also, the performance can be evaluated by considering various mother wavelets. Finally, the proposed expert system differentiates good and bad images with SVM. The metrics are evaluated on CASIA v1.0 dataset using hybrid feature set obtained from the SFs and TFs. We obtained an accuracy of 99.30% which outperformed other existing approaches.