Spliced Image Forgery Detection By Texture And Statistical Features

With the rapid advancement of technology,all fields across the world are getting smarter and digital.Nowadays all sorts of information are directly or indirectly attached to computers and smart technical systems.Digital information includes video&audio streaming as well as images.Social media-based reporting has become popular.Therefore,it has become vital to verify the visual information passing through social media.In past decades,lots of forged digital information passed through the authentic organization.As the world is beeoming smarter,it also makes criminals to find more smarter ways.They use technology and powerful software to forge original court proofs.Image forensics i5 a field of study to eope with such suspicious activities going on over the digital world.As there are many sub-fields of image forensics,we are dealing with 4Cspliced image detection,.When an image is forged or manipulated,its underlying properties are manipulated as well.This manipulation changes the consistency of underlying properties of an image.A robust model and technique is developed to detect splieed images.This technique is based on statistical and texture features analysis.This method uses both spliced and authentic image datasets.Our proposed method works by calculating the texture and statistical features.To evaluate statistical features,some popular texture descriptors i.e.,Grey-level Run Length Matrix(GLRLM)and Gray-Level Co-occurrence matrix(GLCM)are examined.For the evaluation of texture features Histogram of Gradients(HOG),Discrete Wavelet Transform(DWT),GWT(Gabor Wavelet Transform)and local Phase Quantization(LPQ)descriptors are examined.Afterwards,features are extracted with different texture descriptors to compute different texture information in the forged and authentic images separately.Thus,in the first step texture and statistical features are calculated from authentic images.In second step,texture and statistical features are calculated from forged images.Two different classifiers are used to find accuracy of the proposed technique.Two different datasets are examined to find robustness and accuracy.Colombia dataset contains spliced and authentic images.We used 286 authentic images and 287 spliced images.CoMoFoD is a copy-move image dataset.Images in this database are forged with Copy-Move technique.We examined 200 copy-move forged images and 200 authentic images from this dataset.Proposed technique worked well on both datasets.Although,a comprehensive research has been carried out on image splicing but still there is not any benchmark model or algorithm which works for all kinds of forged images.We aimed on specific features of the image.Different kinds of forged images require different schemes and techniques to acquire adequate results for the detection.