Image efficient image forgery detection and localization system

Image Tampering DetectionPrashali Shah, Devanshi Sheth, Priya Jain, Prof. Jyoti WadmareDepartment of Computer EngineeringK. J. Somaiya Institute of Engineering and Information Technology, Sion 400022.University of [email protected], [email protected], [email protected], [email protected] Abstract — Over the past few years, various efforts were taken by image forensics to detect the manipulated images and to trace the tampered region. Images are manipulated using various different techniques and algorithms due to easy to use photo editing softwares and it has been very difficult to develop a single algorithm that deals with detecting all types of forgery. This proposed work is an attempt to design an efficient image forgery detection and localization system that can be used by the Image forensics department to detect whether the image is real or manipulated and which section of the image is forged. As lots of research is carried out in this domain, the focus of this paper will be to highlight the effect of fusion of various algorithm so that a sound application is developed that deals with all sort of image forgery. The proposed system design takes into consideration the effects of Copy-Move Technique and Image splicing technique.Index Terms — Image forensics, Image forgery detection, Image forgery localization, Copy-move technique , Splicing technique.INTRODUCTIONThe availability of various softwares both in smartphones and computers allow almost everyone to modify the image and publish them publicly. Image forensics mainly deals with verifying if the image is fake or pristine. With the increasing use of sophisticated easy to use photo editing softwares it has become difficult to identify whether the image considered is tampered or not. Maliciously tampered images would lead to some potentially serious consequences in our daily life1. Digital images are a popular source of information and its reliability is very important. Therefore, image forensics have gained considerable attention during the past decade2. Images forgery classification can be done using two approaches: Active approach and Passive Approach. The maintenance of integrity and authenticity of digital images is a major problem.Figure 1: Techniques of forgery detection.Generally there are two main problems in image forensics, one is forgery detection and the other one is forgery localization3. Forgery detection  mainly deals with determining if image is edited by using editing tools and performing operations such as image splicing and copy move method whereas forgery localization deals with pointing out the area in a fake image which is manipulated.Active Approach: Active approach is also known as non-blind or intrusive method. It has a) Digital Watermarking and b) Digital Signature. Watermarking is a mark that is embedded into an image. This is like signing an image stating that it is a real image. Digital signature is used for validating the authenticity of an image. It uses public and private key algorithm for signature generation. It has certain drawbacks and this is the only reason we use passive approach in order to verify whether the images are manipulated and if they are manipulated then by what means.Passive Approach: Passive approach is also known as blind approach or non intrusive method. In this approach, one does not require prior information about the original image. Detection is done directly on the image and based on the pixels and the sharpness of the image detection is performed and manipulated images are operated. In this paper, we are focusing mainly on the passive approaches of image forgery detection as they are more efficient and output the forgery done with effective results.LITERATURE REVIEWThere are two categories of image tampering detection that is active/intrusive methods and passive/non intrusive method. Passive detection is  tricky because we don’t have any information of the original image such that it can be compared with the tampered image. Using passive detection we can check for tampering like copy move, splicing and image resampling.Copy Move Forgery Detection: Copy and move forgery is one of the common type of forgeries in which, a part of any image which can have any dimension and shape is copied and pasted over that same image at different location in image, pasting can be single or multiple, essence of that copy and paste is that it performed over same image to hide or manipulate some important feature/information of image. Here source and destination image of forgery is single image4. This tampering is mainly about copying some region from image it has different approach to tamper any image, some use that region directly, some apply some transform like scaling, rotation, skewing, stretching and flipping over copied portion so they fool the detector and make a similar copy near to original so it makes detection difficult. There are several techniques to detect the number of forgeries in images.Copy and Move Forgery Detection Methods:Copy and move forgery has strong correlation between the copied and pasted parts.Preprocessing is applied over that, it improves the image that eliminates undesirable distortions and enhances the image features, preprocessing includes color conversions, image resizing, dimensions reductions, low pass filtering etc.Using SIFT(Shift Invariant feature transform):In this method the forged image is taken as input and the features of the image are extracted using SIFT. The key points obtained can be clustered using k++ means algorithm. The features are matched and similar regions are highlighted5.Block based detection:The input image is divided into overlapping blocks, the features of the block are extracted using RDM and save in a feature matrix. Find similar blocs using lexicographical sorting and correlation in row vectors of the matrix4.Image Splicing Detection:Image splicing is a method of combining two or more images to make it composite. When images are spliced resulting image shows edges, regions and blur at the point where they are spliced but editing tools have made it easy to remove those traces, thus it has become very difficult to detect image splicing.There are several techniques to detect splicingUsing Edge color inconsistency:The method is based on consistency check of color distribution in the neighborhood of edge pixels.Hue histogram entropy is then computed to capture abnormality of color distribution at these boundaries. Inconsistency of color distribution among different edge pixel neighborhood is used to localize the splicing boundary6.Using Illuminant Color Inconsistency:Image is divided into many overlapping blocks, then a classifier is used to adaptively select illuminant estimation algorithm based on block content. Illuminant color is estimated on each block, and the difference between the estimation and reference illuminant color is measured. If the difference is larger than a threshold, the corresponding block is labeled as spliced block7.Using block DCT coefficient:DCT coefficients contain useful information which can be used for detection of image splicing.A new discriminative feature representation has been proposed based on the analysis. From experimental results, it is observed that the new feature representation can achieve better detection performance (91.06%) compared with the traditional methods as mentioned in 8. Image Retouching Detection:Image retouching is another type of image forgery and it is mostly used in commercial and aesthetic applications. This approach is followed in order to upgrade or degrade the quality of any image. Retouching can also be done in order to create a fusion of multiple images which may require various effects such as resizing the image, stretching or compressing an image or rotating the image in any dimension. Copy move detection and splicing method are used to determine the Local image retouching. While Global image retouching includes changing contrast and illumination9. These global retouching can be detected using algorithm in 10 it also suggests methods for histogram equalization.PROPOSED SYSTEMThe method involves blind detection of forgery in images. The two main types of forgery i.e. copy move and splicing will be detected.The proposed method includes detecting copy move and splicing forgeries as well as analysing and comparing them to obtain better results.Figure 2: Proposed method for forgery detectionIn this method the first step is to collect datasets which includes gathering images which are original, images on which copy move forgery is performed, images on which splicing is performed. The input image then undergoes various preprocessing operations that removes noise from the image and enhances its quality. The image will then be tested first for copy move detection.The image is analysed to detect copy move forgery using overlapping block based approach. The images in the dataset will be tested for copy move forgery (CMF) which is a common image forgery technique in which part of image is copied and pasted to another location in the same image. The next step will be to detect if splicing i.e. image formed by combining two or more images. This will be done using local illumination estimation. The image will be divided into horizontal and vertical bands. Thereafter, the estimation of each illuminant will be done by generalized grey world algorithm. The fake patches for each illuminant are produced by calculating the intersection between the pristine horizontal and the vertical bands and can be detected i a forgery detection map. Incase of any tampering detection, the forged area will be localized and highlighted.CONCLUSIONWith the increasing use of image editing tools, forgery detection has become a mandatory need. In this paper, we have proposed a system that does forgery detection as well as forgery localization. This paper focuses on a fusion method of Copy-Move and Image Splicing Technique. The images are subject to manipulation by copying a part of that particular image or by cutting a part of the other image in order to make a misleading image. Therefore, there is utmost need of forgery detection technique that can detect any type of forgery.FUTURE SCOPEIn the next step we will further improve the performance of results obtained by combining the results with other techniques such as image segmentation ,computer vision and semi supervised learning.There is a need to design methods for the case where the suspicious tampered region is not known.The robustness of the algorithm against various techniques will also be studied in the future.