A simple and fast method based on DWT and image texture for detection of splicing forgery in images

This paper proposes a simple and fast method for splicing detection in images based on discrete wavelet transform (DWT) and statistical texture features of the image. Splicing is a common image forgery operation involving merging of two different images to create a new image to conceal or change the information conveyed by the original images. The fact that splicing forgery introduces new texture into the original image in addition to sharp transitions and abrupt changes is exploited in the proposed method. Images are firstly subjected to 3 level DWT decomposition followed by image reconstruction using the detail sub bands only. Using the original image and the reconstructed images five Gray level difference method (GLDM) texture features are obtained to form feature vectors. To lower the feature dimensionality, instead of the features their statistical mean and standard deviation is used to form the feature vector. Finally support vector machine classifier is trained to classify the test images as authentic or spliced. The proposed approach can be used to detect if a given test image is authentic or spliced with an accuracy of 92 % on CASIA image dataset and 82% on Columbia image dataset.