Image and video forgery identification and localization from multimedia forensics perspective

Abstract

Abstract newlineIn the computing world, multimedia forensics is an exciting and challenging newlinefield, which is basically a branch of digital forensics. With the newlinerapid increase in the use of digital technology, crimes today are committed newlineusing contemporary techniques that do not involve physical newlinecontact. As a result, forensic specialists are unable to examine and newlineanalyze the data at the crime scene. A change in the investigation newlinetechniques is necessary to achieve effective investigation of crimes involving newlineadvanced technology. This thesis focuses on image and video newlineforgery analysis from a multimedia forensics perspective. newlineForgeries of digital images compromise the authenticity and integrity newlineof the images. We focus on two frequently used image forgery attacks newlinei.e., copy-move and image splicing. The use of deep learning-based newlineapproaches in image recognition tasks inspired us to develop a model newlinethat can detect and locate manipulated regions in an image. We propose newlinean LSTM-CNN based hybrid model for the generation of binary newlinemasks and detect the forged region with an improved SIFT algorithm. newlineThe SIFT algorithm helps the model invariant to detect and localize newlineforged objects. Then generate the bounding box around the forged newlineregion to classify the image tampering as copy-move or image splicing. newlineVideo forgery has become an easy and on-going task for the users of newlinesmart devices since easily available software tools made the task of a newlinenaive user effortless for video forgery. The impact of video forgery is newlinecritical when it is used to defame a personality and hide (or forge) newlineimportant information to prove innocent in a crime scene and escape newlinefrom legal action. The traces left behind after the forgery can be newlineused to distinguish between genuine and manipulated videos. Using newlinepassive approaches, we can detect any unauthorized manipulation, newlinewhether it s done within a frame (intra-frame level) or between newlinevii newlineframes (inter-frame level). The investigation of inter-frame video forgeries newlineis the main emphasis of our work. We propose, a deep learnin