Image splicing forgery detection using noise level estimation

Digital image forgery has become one of the serious issues in today's era. Digital images can be forged in several ways. Image splicing is a simple and most commonly used forgery technique. In image splicing, two or more images are used to create a single composite image. However, the detection of image splicing forgery is not easy. Motivated by the fact that the images captured from different devices show different noise levels, this paper proposes a new method to detect and localize the image splicing forgery based on noise level estimation. In the proposed method, initially, the input image is divided into irregular-shaped superpixel blocks using the Simple Linear Iterative Clustering technique. Secondly, the PCA-based image estimator is used to estimate the noise level from the superpixel blocks. Finally, the k-means clustering technique is used to cluster the blocks into authentic and spliced blocks based on the noise levels. The experimental results performed on the CUISDE dataset demonstrate that the proposed method can localize the image splicing forgery with better accuracy as compared to existing state-of-the-art methods.