A passive blind approach for image splicing detection based on DWT and LBP histograms

Abstract

Splicing is the most generic kind of forgery found in digital images. Blind detection of such operations has become significant in determining the integrity of digital content. The current paper proposes a passive-blind technique for detecting image splicing using Discrete Wavelet Transform and histograms of Local Binary Pattern (LBP). Splicing operation introduces sharp transition in the form of lines, edges and corners, which are represented by high frequency components. Wavelet analysis characterizes these short-time transient by measuring local sharpness or smoothness from wavelet coefficients. After first level wavelet decomposition of the image, texture variation is studied along the detailed and approximation coefficients using local binary pattern (LBP), since tampering operations disrupts the textural microstructure of an image. Feature vector is formed by concatenating the LBP histogram from the four wavelet sub bands. The classification accuracy of the algorithm was determined using svm classifier using 10-fold cross validation. The method gives maximum accuracy for the chrominance channel of YCbCr color space, which is weak at hiding tampering traces. It is tested on four different kinds of standard spliced image dataset and its performance is compared with some of the latest methods. The method offers accuracy up to 97% for JPEG images present in the spliced image dataset.