TY - JOUR
AU1 - Vinolin,, V
AU2 - Sucharitha,, M
AB - Abstract Information in the form of the image conveys more details than any other form of information. Several software packages are available to manipulate the images so that the authenticity of the images is being questioned. Several image processing approaches are available to create fake images without leaving any visual clue about the forging operation. So, proper image forgery detection tools are required to detect such forgery images. Over the past few years, several research papers were published in the digital image forensics domain for detecting fake images, thus escalating the legitimacy of the images. This survey paper attempts to review the recent approaches proposed for detecting image forgery. Accordingly, several research papers related to image forgery detection are reviewed and analyzed. The taxonomy of image forgery detection techniques is presented, and the algorithms related to each technique are discussed. The comprehensive analysis is carried out based on the dataset used, software used for the implementation and the performance achievement. Besides, the research issues associated with every approach were scrutinized together with the recommendation for future work. 1. INTRODUCTION With the growing trend in science and technologies, a lot of information in the form of text, graphics, audios, videos and images is spreading in newspapers, radio, television, magazines and websites. Among several information formats, digital information in the form of images has a vital role in conveying information in a better manner, and it serves as a potent communication tool. Digital cameras and the internet are widely used in daily life, and so capturing and sharing of images have become easy for even common persons [1]. Regarding digital images, with the easy availability of image editing software, even a common person can easily generate fake images, which seems to be visually realistic. Fake images are generated to change or modify the information present in the original image. Some forged images are used for advertisements and entertainment purposes, which are harmless to society. However, some fake images are created with the intention to threaten a person or a society [2]. Such fake images are, in fact, more dangerous than any other form of fake information. Therefore, effective differentiation of fake images and real images is a major problem. Digital image forensics is relatively a new research field with the objective to validate the authenticity of the digital images. The authenticities of the images are verified by addressing two major problems, such as identifying the image source and identifying the traces of image forgery [3]. The most common ways to forge images are copy–move, resampling and image splicing. Forgery images are created for various reasons, and several forgery images that created a serious impact on society are reported over several years. Figure 1 shows some forgery images that created issues recently. Figure 1a was captured on July 2008, which shows the launching of Iranian missiles. The left one is the original image, which consists of three missiles, while in the right image, some parts were copied from the image and pasted onto some other region of the original image itself, to depict that four missiles were launched [4]. Figure 1b shows the retouched photo of O.J. Simpson, printed on the front page of ‘Time’ magazine (right side) on June 1994, who was arrested in a murder case. The original photo was published in ‘Newsweek’ magazine (left side). Later, the ‘Time’ magazine was accused of manipulating Simpson’s image with the racist purpose [3]. The right side of Fig. 1c is the original image captured while Queen Elizabeth II awarded Ross Brawn, and the left side was the spliced photo which shows Jeffrey Wong Su En, a Malaysian Politician, being awarded by the Queen. Jeffrey Wong Su En was accused of superimposing his head onto the image of Ross Brawn [3]. Images illustrating image forging techniques. (a) Iranian montage of missiles. (b) O.J. Simpson photo on magazines cover. (c) Spliced image of Jeffrey Wong Su En receiving an award from Queen Elizabeth II (left side is the original image, and the right side is the forgery image). FIGURE 1. Open in new tabDownload slide FIGURE 1. Open in new tabDownload slide Due to several susceptible issues related to detecting forgery images, numerous techniques for detecting forgery images were presented by researchers to restore the faith in digital images. Recently, owing to the emerging techniques in digital image forensics, manipulating images has become a challenging task, and it somewhat increased the trust in digital images. In the literature, several techniques for the identification of fake images are presented. In this paper, a detailed survey is carried out to identify the challenges related to each technique. Accordingly, several research papers were taken for the review and taxonomy of image forgery detection techniques is presented, which illustrated the fundamental difference between various techniques. Then, the analysis is carried out based on various criteria, such as Dataset, Software used for implementation and the performance achieved by each technique. Finally, the research issues related to each technique are discussed, which will be useful for the researchers to accomplish further research on image forensics. The organization of this paper is as follows: Section 2 deliberates the classification of image forgery detection techniques with suitable hierarchy. An analytic study is presented in Section 3 on the reviewed techniques based on various parameters. Section 4 deliberates the research gaps related to each technique, along with the recommendation for future work. The conclusion is given in Section 5. 2. REVIEW AND CLASSIFICATION OF IMAGE FORGERY DETECTION TECHNIQUES This section presents the literature survey of existing techniques developed so far for forgery detection. Moreover, the taxonomy of the forgery detection techniques is elaborated in this section. 2.1. Taxonomy of forgery detection techniques Based on the survey conducted over several research papers, the taxonomy of the image forgery detection techniques is demonstrated in this section. The taxonomy is prepared depending on the techniques employed for image forgery detection. Two major classifications under image forgery detection techniques are active approaches and passive (blind) approaches. Active approaches are further categorized into techniques based on digital signature and digital watermarking. The passive approaches are again classified into pixel-based approaches, format-based approaches, camera-based approaches, geometry-based approaches and physically based approaches. The taxonomy of image forgery detection techniques is given in Fig. 2. Hierarchical classification of image forgery detection techniques. FIGURE 2. Open in new tabDownload slide FIGURE 2. Open in new tabDownload slide 2.2. Descriptive study on literary works In this subsection, all the research works that are categorized on the basis of the taxonomy are briefly explained as follows. 2.2.1. Active image forgery detection techniques In active image forgery detection approaches, pre-processing steps like generating digital signature generation or embedding watermark into the image are done at the time of generating the image. If the image is forged, then the secret message embedded in the image cannot be recovered. In this way, the authenticity of the images can be determined. 2.2.1.1. Digital signature The digital signature process extracts the distinctive features of the image while capturing the image. At the time of checking the authentication of the image, the signature is generated again by applying the same technique. Then, by comparison, the originality of the image can be verified. In [5], a forensic signature approach for image authenticity was proposed. Here, wavelet transform and adaptive Harris corner detection algorithm were utilized to identify the important feature components. Based on the statistical feature components dependent on the neighborhood data, the forensic signature was constructed. Then, the forensic signature and the Fisher criterion were used to decide the authenticity of the image. 2.2.1.2. Digital watermark In digital watermarking, the source makes the watermark image, which is then embedded into an image to generate the watermarked image. The watermarked image is then utilized to identify the watermark embedded to validate the authenticity of the image. In [6], a computer generated hologram (CGH) coding system was developed for watermark embedding, and it was tested. A blind additive embedding technique was used to embed the hologram. 2.2.2. Passive image forgery detection techniques In a passive image forgery detection approach, the authenticity of the image is verified from the image itself [7]. This approach does not require any additional information, such as signature or watermark as like active approaches. The passive image forgery detection techniques are broadly classified as follows: pixel-based techniques, format-based techniques, camera-based techniques, physically based techniques and geometry-based techniques. 2.2.2.1. Pixel-based techniques Pixel-based approaches detect forgery images by detecting the pixel-level statistical abnormalities. Pixel-based approaches are further classified as copy–move, image splicing and image retouching approaches. (i) Copy–move forgery detection methods Copy–move forgery is a simple, effective and common image forging method. In this method, certain regions of the image are copied, moved and pasted on some other region in the same image. Such kind of image forgery is done to conceal a particular region in the image or to create a duplicated region in the image [8]. (a) Block-based methods The block-based methods segment the input image into regular, overlapping or non-overlapping blocks. After that, by matching the extracted features from the blocks, the possibility of image forgery was identified. In [9], two algorithms were presented as the first attempt to detect copy–move image forgery. One of the algorithms utilized the precise match, and the other used the approximate match to detect copy–move image forgery. In this manner, the forged regions were identified. A similar approach for detecting the duplicate region was presented in [10]. Initially, principal component analysis was employed to small image blocks, which resulted in compressed dimensional depiction. Subsequently, the image blocks were lexicographically sorted, and thus, the duplicate regions were identified. By this approach, forgery detection is possible even with the presence of noise. The authors in [11] used a block-based approach which divided the image into small overlapping blocks. For each block, seven characteristic feature vectors were extracted. Then, the similarities between these blocks were compared using their characteristic feature vector. The correct matching pair was identified, and finally, the possible duplicated regions were identified. In [12], another block-based forgery detection method was presented for the images under high geometrical constraints. High computational complexity problem of the block-based methods was solved in [13]. The image was initially divided into a texture part and a smooth part. In the texture part, key points were extracted and matched. Non-overlapping blocks were used as a candidate in the smooth region. Thus, the computational complexity was reduced, and the matching accuracy was improved. Another blocking-based method was proposed in [14]. Here, a convolutional neural network (CNN) was used as the processing block in each unit. (b) Feature-key based methods In feature-key based methods, the image key points were extracted and matched over the entire image to detect the duplicate region. The features were fed as input to the matching block, and the features undergo a similarity test. In [8], forgery images were detected by adaptive over-segmentation and feature point matching. In [11], a contrast context histogram (CCH) was used for the effective detection of copy–move forgery, and the key points were isolated by the k-means clustering algorithm. (ii) Image resampling-based forgery detection methods In image resampling, the correlations among the resampled image aid in forgery detection [7]. An effective approach for detecting image resampling was proposed in [12], which was based on the fact that interpolation operation used in image resampling indicates some discrepancy among the original and the forged images, particularly in the high-frequency domain. Two methods for detecting and localizing image forgery, depending on combining resampling features and deep learning, were proposed in [17]. Radon transform was used in the first method to estimate features. Forged regions were identified using the random walker segmentation algorithm, which determined the segmentation of an image from a set of markers labeling several phases. In random walker segmentation, less number of pixels with pre-defined labels was considered. Then, the pixels analytically and quickly determined the probability. After that, a random walker, initiating at every unlabeled pixel, moves to any one of the user-defined pixels. By labeling every pixel to the label which has the greatest probability, high-quality image segmentation was done. In the second method, the extracted features were sent via a long short-term memory (LSTM)-based network for classification and localization [17]. (iii) Image splicing In the splicing approach, a particular area from the image was copied and then pasted onto some other image for creating the forgery image. In spliced images, it is difficult to visually detect the borders between the spliced regions [14]. An image splicing detection method using textural features based on gray level co-occurrence matrices (TF-GLCM) was developed in [15]. Level co-occurrence matrices were computed based on difference block discrete cosine transform (DBDCT) arrays to capture the texture information and spatial relationship between image pixels. Moreover, the mean and the standard deviation of textural features of all GLCM in four directions were estimated to combine the feature vector, and it was fed into the support vector machine (SVM) classifier which differentiates the original and the spliced images. 2.2.2.2.Format-based techniques Format-based approaches are based on the statistical correlation caused by the lossy compression method. (i) Double JPEG More commonly, the JPEG format is used to store images. For manipulating the image, it has to be uploaded into some image manipulating software and has to be saved again in JPEG format. Hence, the image is compressed twice, and this induces certain patterns in the manipulated image. The induced pattern then serves as an indication of image manipulation. In [16], an approach was proposed for detecting the forged JPEG images and also to detect the forged regions by investigating the double quantization effect in the discrete cosine transform (DCT) coefficients. In [17], an approach was proposed to detect non-aligned double JPEG compression. This approach was based on the integer periodicity of the DCT coefficients, which allows applying a threshold detector. (ii) JPEG blocking JPEG images use image blocks and the blocking pattern of the image changes when it is manipulated. In [18], an approach for detecting image manipulation in JPEG images was proposed. The lossy JPEG compression introduced inherent blocking artifacts in the images, and such artifacts serve as a clue for forgery detection. Image forgery detection by measuring the inconsistency in the blocking artifact was presented in [19]. The blocking artifact due to the JPEG compression was used as evidence of image forgery. (3) JPEG quantization Digital image forgery detection based on JPEG quantization was proposed in [20]. The lossy compression schemes in the digital camera employed a quantization table, which controls the amount of image compression. Different cameras will have different quantization tables, and comparing the quantization table of the image under consideration with the database of the known camera indicates whether the image is authentic or fake. 2.2.2.3. Camera-based techniques In camera-based approaches, the artifacts generated by the sensor or lens of the camera are effectively utilized by various image forensic techniques. These techniques are applied to estimate the camera artifacts based on various modeling theories. (i) Camera response An automatic method to detect image forgery depending on camera response was presented in [21]. In this approach, the image under consideration was initially divided into different arbitrarily shaped segments. Then, camera response function (CRF) was identified from every segment by locally planar irradiance points (LPIPs) geometric invariants. To identify a region as original or forged, CRF-based cross fitting and local image features were calculated and applied to the statistical classifier. Then, the segment level scores were combined to decide the authenticity of the image. (ii) Chromatic aberration Irregularity in the chromatic aberration was used as a clue to detect image forgery in [22]. With the initial assumption that only a small region of the image is forged and it has no effect on the global estimate, the aberration was determined for the entire image. Then, the global estimate was compared with the aberration estimated from the image blocks, thus detecting the forgery image. (iii) Color filter array In [23], the correlations due to the CFA interpolation was enumerated and described how such correlations were automatically detected in the image. This approach was efficient for finding forgery images in compressed color images. (iv) Sensor array In [24], a technique for verifying the authenticity of the camera output based on a set of features was presented. Here, the forgery images were identified by using statistical noise features set such as denoising operations, wavelet analysis and neighborhood prediction. In [25], an approach for identifying the camera and verifying the integrity of the image was discussed. Both tasks were carried out by recognizing pixels’ photo response non-uniformity (PRNU) noise in the image. Another approach for image source identification was presented in [26]. This method had been worked by detecting the existence of camera pattern noise, which is an exclusive feature of imaging sensors. The region of the image without pattern noise was considered as the forged region. 2.2.2.4. Geometry-based techniques In the geometry-based technique, the variation in the principal point in the image acts as proof to check the image authenticity. The geometric techniques for image forgery detection are classified into three categories [4] based on (i) the camera’s intrinsic principle, (ii) metric measurements and (iii) multiple-view geometry. (i) Based on the camera’s intrinsic parameters The camera’s intrinsic parameters, such as principal point and skew, can be employed to detect the forgery images. (a) Principal point In [27], an approach based on the estimation of the principal point of the camera from the person’s eye in the image was presented to detect the composite images. The difference in the principal point of the person’s eye in the image was considered as the clue to detect the forged image. Authentic images have a principal point near the center of the image, but in forged images, the principal point was moved proportional to the alteration of objects or persons in the image. (b) Skew In [28], an approach was presented for automatic detection of recorded video from the theater screen. The re-projection introduced skew, which was based on the perception between the horizontal and vertical pixel axes, to the camera’s intrinsic principal. This same concept was used to detect forgery image if the image was re-photographed. (ii) Based on metric measurements In [29], the authors reviewed some techniques for transformation estimation, considering metric measurements on planar surfaces from an image. By these techniques, the planar surface transformation was improved. (iii) Based on multiple-view geometry In [30], a composite image detection technique was proposed by implementing two-view geometrical constraints such as planar homography or fundamental matrix. This approach was capable of detecting forged regions effectively on images of a similar scene, but it required two correlated images. 2.2.2.5. Physically based techniques Even though the forger carefully adjusts the lighting conditions in the scene, there might be some traces of forgery such as inconsistency in reflection, inconsistency in lighting or inconsistency in the shadow [31] which may present in the image, which aids in detecting forgery image. Physically based forgery detection technique is classified into three major categories: (i) based on lighting inconsistency, (ii) based on shadows and (iii) based on reflection inconsistency. (i) Based on lighting inconsistency Usually, different photographs are taken under varying lighting environments. While merging images captured under varying lighting environments, it is hard to equalize the lighting conditions of the images. As a result, detection of lighting inconsistency detection in an image can be used for identifying the tampered image. (a) Based on lighting direction If a certain part of the image is very much illuminated when compared to other parts, then we can conclude that the light source is in the position towards the specific part of the image. (1) Infinite light source (2D) If a light source is illuminating an object from a certain direction, one side of the object facing the direction of the light source will be more illuminated than the opposite side. This observation is dignified by assuming that the amount of light hitting a surface is proportional to the surface normal and the direction of the light [7]. An algorithm that automatically estimates the direction of the light source from an image was presented in [32]. Light source direction estimation and the lighting inconsistency measurement, used to detect the spliced image, were presented in [33]. Since it is not possible to obtain 3D surface normals from a single image, 2D surface normals at the occluding boundaries were alone taken into consideration. (2) Infinite light source (3D) The drawbacks of 2D lighting-based techniques are overcome by the 3D lighting-based techniques. The 3D lighting-based technique holds the reality that with minimum training, the 3D structure of the scene is estimated from which the 3D lighting can be estimated [31]. Since the entire 3D lighting properties are estimated, the forensic technique becomes more powerful, and while forging the image, the forger has to correct the entire 3D lighting properties of the image. This is a tedious task and consumes more time. In [31], an approach was proposed for estimating the 3D lighting properties from a single image, which depends on human interference to obtain the local 3D structure of the scene. In [34], the estimation of the 3D direction from the reflection of light in the human eye was presented. Then, by comparing the estimated light direction among various people in the image, the forgery image is detected. (b) Based on the lighting environment In the basic lighting model, an assumption was made that there was only a dominant light source (e.g. Sun) positioned infinitely far away from the object. This assumption is suitable only for outdoor images. By comparing the lighting consistencies of objects or persons present in that image, the spliced image can be detected. (1) 2D lighting environment In [35], the complex lighting environment was approximated with a low-dimensional model. In this approach, the lighting environment and the reflectance functions were expressed in terms of spherical harmonics. Then, the inconsistency in the lighting model serves as the clue for forgery. In [36], a robust technique which uses the information along the 2D occluding contour was presented, and spherical harmonic frames were used to estimate the lighting feature. (2) 3D lighting environment In [37], 3D lighting environment approximation and model parameter estimation from a single image were discussed. In this approach, the 3D model of the person’s head in the image was estimated, and then it was registered to the face under analysis. Then, 3D surface normals and their intensities were utilized for estimating the lighting environment. In [38], the 3D shape was learned from the image itself, which generated the 3D lighting based-forensic tool, which works on objects of arbitrary shape. An optimized 3D lighting estimation method was proposed in [39], which considers the occlusion geometry and texture information. Better accuracy in lighting estimation was achieved by this approach, and it effectively differentiates authentic and fake images. (c) Based on illumination color Illumination color-based approaches are based on the discrepancy in the interface between the color of the object and the color of the light. In [40], an approach for accessing illumination color consistency over the scene was presented. In this approach, the illuminant color was estimated on the illuminated region, and then local illuminant color estimation was performed over the entire image. Another approach was proposed in [41] to detect the composite image using inconsistency in the illuminated color. In [42], the lighting color was estimated first, and then the texture and edge-based features were extracted and are applied to a machine learning approach to automatically decide about the authenticity of the image. In [43], the transformed spaces, represented by image illuminate maps, were explored for imaged forgery detection. Image forgery detection based on illuminant color disparity between various persons in the image was proposed in [48]. The dichromatic reflection model was used to estimate the illumination color. (ii) Based on inconsistency in reflection In a forged image, inconsistency in reflection occurs with the insertion of forged reflection into the image. In [44], an approach was proposed using the fundamental concepts of reflective geometry and linear perspective projection. This approach was based on the fact that geometric inconsistencies occur while inserting forged reflections in the image. (iii) Based on shadows Forensic approaches based on the analysis of lighting and shadows are considerably more effective since it is difficult to modify the 3D lighting effects using easily available photo editing tools. Inconsistencies in the shadow location serve as evidence for forgery detection in [45]. In this approach, both the geometric and photometric constraints were extracted from a single scene, and it was then used for image forgery detection. Photometric inconsistencies of illumination in the shadow were used to detect image composites in [46]. In [47], an approach which detects physical inconsistency from the shadows and shadings in the image was proposed. In this method, lighting was estimated from the shading under the linear perspective model. The approach proposed in [48] was a combination of multiple constraints from shadows to confine the position of point light source. The shadow inconsistencies were estimated by merging several effective constraints. In [49], the forged image was detected depending on the inconsistency in the shadow texture, which was based on the reality that the surface texture of the background of the shadow is not changed. The approach proposed in [50] was based on the reality that the relative amount between the hue, saturation and value (HSV) components on neighboring surfaces with shadow and without shadow may be dissimilar for forgery images and the same for authentic images. 3. ANALYTICAL SURVEY ON IMAGE FORGERY DETECTION TECHNIQUES In this section, the analysis of the image forgery techniques based on several criteria, such as dataset, software used and performance achievement, is presented. The intention of this analytic study is to identify the influence of the techniques, the outcome until achieved and the commonly used datasets for experimentation. 3.1. Analysis based on the datasets employed In this section, the datasets used by the research works in the literature are discussed. Figure 3 shows an analysis based on the dataset used. Depending on the type of forgery to be detected, the various datasets have been used. Twenty percent of research works have used the CASIA (The Institute of Automation, Chinese Academy of Sciences) Image Tampering Detection Evaluation Database dataset [56], which was more commonly used for composite image detection. Similarly, 12% of the research articles are utilized the monochromatic shadow and MICC-F220 (Media Integration and Communication Center) datasets [57]. On the other hand, 11% of the research articles utilized DSO-1 (real-world dataset) and DSI-1 (dataset collected from internet) datasets [46]. The datasets, such as UCID (Uncompressed Color Image Database) [58], RAISE (Raw Images Dataset) [59], USC-SIPI (Signal and Image Processing Institute, University of Southern California) [60], CoMoFoD (Image Database for Copy-Move Forgery Detection) [61], BOSSRAW (Break Our Steganographic System—Raw dataset) [62], Columbia [63] and GRIP (Image Processing Research Group of the University Federico II of Naples) [64] are utilized in 6% of research articles taken for the literature review. Based on this analysis, it is clear that several datasets are utilized by various researchers, but the CASIA dataset [14, 15, 39, 41] is more frequently utilized by researchers. Analysis based on the datasets employed. FIGURE 3. Open in new tabDownload slide FIGURE 3. Open in new tabDownload slide 3.2. Analysis based on software utilized This subsection analyzes the various softwares utilized by the research articles to perform the experimentation of the techniques. Table 1 shows an analysis based on the software used. Accordingly, various implementation platforms such as MATLAB, Pbrt (physically based rendering), C++ and OpenCV (Open Source Computer Vision Library) are utilized for implementing the techniques. Based on the analysis, most of the research works considered in this survey have been implemented using MATLAB. Nearly, 10 research works have been implemented using MATLAB. For face rendering in the physically based approach, Pbrt has been used. On the other hand, Pbrt and MATLAB are combined to implement the image forgery in only one of the techniques considered for the analytical study. The important scenario noted here is that most of the techniques are not mentioned about the name of the tool in their paper. Analysis based on the software used. TABLE 1. Analysis based on the software used. Softwares References MATLAB [5, 13, 14, 19, 29, 30, 31, 43, 53, 54] Pbrt [39, 42] Pbrt+ MATLAB [55] Not mentioned Others Softwares References MATLAB [5, 13, 14, 19, 29, 30, 31, 43, 53, 54] Pbrt [39, 42] Pbrt+ MATLAB [55] Not mentioned Others Open in new tab TABLE 1. Analysis based on the software used. Softwares References MATLAB [5, 13, 14, 19, 29, 30, 31, 43, 53, 54] Pbrt [39, 42] Pbrt+ MATLAB [55] Not mentioned Others Softwares References MATLAB [5, 13, 14, 19, 29, 30, 31, 43, 53, 54] Pbrt [39, 42] Pbrt+ MATLAB [55] Not mentioned Others Open in new tab Analysis based on performance achievement. TABLE 2. Analysis based on performance achievement. Metrics Measurement Articles Metrics Measurement Articles Accuracy 85–90% [28, 46] Bhattacharya coefficient 76.8 [45] 90–95% [24, 41, 43] Detection rate 86% [42] 95–98% [18, 51, 15] Area under the curve (AUC) 0.9138 [13] 0.5795 [17] Precision 70–80% [21] True positive rate 78.55% [11] 90–98% [52, 39] False positive rate 0.3% [28] 35% [11] 80–90% [53] Average angular error 10.5 [34] Recall 70–80% [21] Probability of correct detection 0.95 [5] 80–90% [53, 39] 90–95% [52] Error 0.03% [35] F1 score 87.55% [53] 6.67% [12] 89.97% [39] Metrics Measurement Articles Metrics Measurement Articles Accuracy 85–90% [28, 46] Bhattacharya coefficient 76.8 [45] 90–95% [24, 41, 43] Detection rate 86% [42] 95–98% [18, 51, 15] Area under the curve (AUC) 0.9138 [13] 0.5795 [17] Precision 70–80% [21] True positive rate 78.55% [11] 90–98% [52, 39] False positive rate 0.3% [28] 35% [11] 80–90% [53] Average angular error 10.5 [34] Recall 70–80% [21] Probability of correct detection 0.95 [5] 80–90% [53, 39] 90–95% [52] Error 0.03% [35] F1 score 87.55% [53] 6.67% [12] 89.97% [39] Open in new tab TABLE 2. Analysis based on performance achievement. Metrics Measurement Articles Metrics Measurement Articles Accuracy 85–90% [28, 46] Bhattacharya coefficient 76.8 [45] 90–95% [24, 41, 43] Detection rate 86% [42] 95–98% [18, 51, 15] Area under the curve (AUC) 0.9138 [13] 0.5795 [17] Precision 70–80% [21] True positive rate 78.55% [11] 90–98% [52, 39] False positive rate 0.3% [28] 35% [11] 80–90% [53] Average angular error 10.5 [34] Recall 70–80% [21] Probability of correct detection 0.95 [5] 80–90% [53, 39] 90–95% [52] Error 0.03% [35] F1 score 87.55% [53] 6.67% [12] 89.97% [39] Metrics Measurement Articles Metrics Measurement Articles Accuracy 85–90% [28, 46] Bhattacharya coefficient 76.8 [45] 90–95% [24, 41, 43] Detection rate 86% [42] 95–98% [18, 51, 15] Area under the curve (AUC) 0.9138 [13] 0.5795 [17] Precision 70–80% [21] True positive rate 78.55% [11] 90–98% [52, 39] False positive rate 0.3% [28] 35% [11] 80–90% [53] Average angular error 10.5 [34] Recall 70–80% [21] Probability of correct detection 0.95 [5] 80–90% [53, 39] 90–95% [52] Error 0.03% [35] F1 score 87.55% [53] 6.67% [12] 89.97% [39] Open in new tab 3.3. Analysis based on performance achievement This section describes the performance achieved by the existing forgery detection techniques. Table 2 shows the analysis based on performance achievement. In forgery detection, accuracy of 85–90% is achieved by the research works in [28, 46]. Ninety to 95% of accuracy is achieved by the research works of [24, 41, 43]. The highest accuracy of 95–98% is achieved by the research works [18, 51, 15]. In terms of precision, the research works given in [12, 43] obtained the result in the range of 90–98%. Similarly, the recall value of 90–95% is achieved in the works given in [13]. An F1 score of 89.97% is achieved by the technique proposed in [43]. In terms of error, 6.67% is obtained as performance in the research work given in [16]. The finding is that future work can be possible by developing the techniques by showing the performance improvement over the results given in Table 2. 4. CHALLENGES AND RESEARCH OPPORTUNITIES To renovate the reliance on digital images, many forensic methods have been proposed by the image forensic community to distinguish original and fake images. However, because of the complex nature of the image manipulation operations and huge quantities of images, there is still no commonly effective forgery detection tool. Various types of methods aim at several forgery circumstances, and there exist a few general-purpose methods that fail on images of low quality. In literature, a number of forgery detection algorithms, based on active and passive approaches, are presented. Table 3 shows a summary of the comparison of active methods based on their advantages and disadvantages. Comparison of active methods. TABLE 3. Comparison of active methods. No. Techniques Advantages Disadvantages Ref. no. 1. Digital signature Suitable for all image types Require specially equipped camera or software [5] 2. Digital watermarking Suitable for all image types Require specially equipped camera or software [6] No. Techniques Advantages Disadvantages Ref. no. 1. Digital signature Suitable for all image types Require specially equipped camera or software [5] 2. Digital watermarking Suitable for all image types Require specially equipped camera or software [6] Open in new tab TABLE 3. Comparison of active methods. No. Techniques Advantages Disadvantages Ref. no. 1. Digital signature Suitable for all image types Require specially equipped camera or software [5] 2. Digital watermarking Suitable for all image types Require specially equipped camera or software [6] No. Techniques Advantages Disadvantages Ref. no. 1. Digital signature Suitable for all image types Require specially equipped camera or software [5] 2. Digital watermarking Suitable for all image types Require specially equipped camera or software [6] Open in new tab The main negative aspect of the active approach is that specially equipped cameras or software are required to generate a signature or embed a watermark in the image, which limits its practical application. Also, with this method, it is not possible to detect the manipulations that were done before embedding the watermark [7]. The passive approaches overcome the active approaches since the passive approaches do not entail special software or camera to capture the image. Table 4 shows a summary of the comparison of the different pixel, format and camera-based passive methods based on their advantages and disadvantages. Some issues have to be solved in passive approaches also. While considering the copy–move image forgery detection methods, the block-based approaches have the following drawbacks: for images that are divided into an overlapping block, the computational cost increases for high-dimensional images. Moreover, the number of required computations increases exponentially as a large number of blocks are used, and also feature extraction is difficult [51]. The major drawback of the feature key-based method is that it is less accurate when compared to other approaches. Since the key points are related with high entropy regions, copy–move manipulation in smooth regions is usually not detected [11]. Regarding the resampling approach for image forgery, the performance is affected by the JPEG compression effects, i.e. performance decreases as the compression ratio increases [12]. The overall drawbacks of 2D lighting direction-based techniques are that they focus only on 2D lighting properties estimation since 3D lighting estimation entails prior information about the 3D structure of the scene that is not always obtainable [31]. Comparison of pixel, format and camera-based passive methods. TABLE 4. Comparison of pixel, format and camera-based passive methods. No. Techniques Advantages Disadvantages Ref. no. 1. Block-based approach Can detect even simple details Failed on images with geometrical distortions [9]–[14] 2. Feature-based approach Feature point matching instead of region matching reduces the computational complexity Feature extraction is difficult [8, 15] 3. Resampling Easily separate the forged parts using statistical behaviors in the high-frequency domain Performance decays due to JPEG compression artifacts [7, 16, 17] 4. Splicing Correlations caused by splicing is used as a clue for forgery detection Works only with the non-existence of digital watermarking or signature [18, 19] 5. Double JPEG Able to operate on compressed JPEG images Requires multiple examinations of recompressed images [20, 21] 6. JPEG blocking Inconsistency in JPEG blocking artifacts is measured to detect forgery Detecting forgery without the aid of signature or watermarking tends to be complex [22, 23] 7. JPEG quantization Can detect forgeries whereby image source can be definite or denied Typically usage of different JPEG quantization table is difficult for balancing the compression ratio and image quality [24] 8. Camera response Fully automatic and passive approach, no manual input is required Deeper physical insight and knowledge is required [25] 9. Chromatic aberration Detect tampering by simply comparing the estimates of chromatic aberration and the global aberration The estimation of the model parameters requires an effective image registration algorithm [26] 10. Color filter array Based on a simple fact that for authentic images, there will be a high correlation between the pixels and their neighbors Failed in situations where watermarking technologies are not suitable [27] 11. Sensor array Applicable to all cameras Sensor imperfections degrade the performance [28]–[30] No. Techniques Advantages Disadvantages Ref. no. 1. Block-based approach Can detect even simple details Failed on images with geometrical distortions [9]–[14] 2. Feature-based approach Feature point matching instead of region matching reduces the computational complexity Feature extraction is difficult [8, 15] 3. Resampling Easily separate the forged parts using statistical behaviors in the high-frequency domain Performance decays due to JPEG compression artifacts [7, 16, 17] 4. Splicing Correlations caused by splicing is used as a clue for forgery detection Works only with the non-existence of digital watermarking or signature [18, 19] 5. Double JPEG Able to operate on compressed JPEG images Requires multiple examinations of recompressed images [20, 21] 6. JPEG blocking Inconsistency in JPEG blocking artifacts is measured to detect forgery Detecting forgery without the aid of signature or watermarking tends to be complex [22, 23] 7. JPEG quantization Can detect forgeries whereby image source can be definite or denied Typically usage of different JPEG quantization table is difficult for balancing the compression ratio and image quality [24] 8. Camera response Fully automatic and passive approach, no manual input is required Deeper physical insight and knowledge is required [25] 9. Chromatic aberration Detect tampering by simply comparing the estimates of chromatic aberration and the global aberration The estimation of the model parameters requires an effective image registration algorithm [26] 10. Color filter array Based on a simple fact that for authentic images, there will be a high correlation between the pixels and their neighbors Failed in situations where watermarking technologies are not suitable [27] 11. Sensor array Applicable to all cameras Sensor imperfections degrade the performance [28]–[30] Open in new tab TABLE 4. Comparison of pixel, format and camera-based passive methods. No. Techniques Advantages Disadvantages Ref. no. 1. Block-based approach Can detect even simple details Failed on images with geometrical distortions [9]–[14] 2. Feature-based approach Feature point matching instead of region matching reduces the computational complexity Feature extraction is difficult [8, 15] 3. Resampling Easily separate the forged parts using statistical behaviors in the high-frequency domain Performance decays due to JPEG compression artifacts [7, 16, 17] 4. Splicing Correlations caused by splicing is used as a clue for forgery detection Works only with the non-existence of digital watermarking or signature [18, 19] 5. Double JPEG Able to operate on compressed JPEG images Requires multiple examinations of recompressed images [20, 21] 6. JPEG blocking Inconsistency in JPEG blocking artifacts is measured to detect forgery Detecting forgery without the aid of signature or watermarking tends to be complex [22, 23] 7. JPEG quantization Can detect forgeries whereby image source can be definite or denied Typically usage of different JPEG quantization table is difficult for balancing the compression ratio and image quality [24] 8. Camera response Fully automatic and passive approach, no manual input is required Deeper physical insight and knowledge is required [25] 9. Chromatic aberration Detect tampering by simply comparing the estimates of chromatic aberration and the global aberration The estimation of the model parameters requires an effective image registration algorithm [26] 10. Color filter array Based on a simple fact that for authentic images, there will be a high correlation between the pixels and their neighbors Failed in situations where watermarking technologies are not suitable [27] 11. Sensor array Applicable to all cameras Sensor imperfections degrade the performance [28]–[30] No. Techniques Advantages Disadvantages Ref. no. 1. Block-based approach Can detect even simple details Failed on images with geometrical distortions [9]–[14] 2. Feature-based approach Feature point matching instead of region matching reduces the computational complexity Feature extraction is difficult [8, 15] 3. Resampling Easily separate the forged parts using statistical behaviors in the high-frequency domain Performance decays due to JPEG compression artifacts [7, 16, 17] 4. Splicing Correlations caused by splicing is used as a clue for forgery detection Works only with the non-existence of digital watermarking or signature [18, 19] 5. Double JPEG Able to operate on compressed JPEG images Requires multiple examinations of recompressed images [20, 21] 6. JPEG blocking Inconsistency in JPEG blocking artifacts is measured to detect forgery Detecting forgery without the aid of signature or watermarking tends to be complex [22, 23] 7. JPEG quantization Can detect forgeries whereby image source can be definite or denied Typically usage of different JPEG quantization table is difficult for balancing the compression ratio and image quality [24] 8. Camera response Fully automatic and passive approach, no manual input is required Deeper physical insight and knowledge is required [25] 9. Chromatic aberration Detect tampering by simply comparing the estimates of chromatic aberration and the global aberration The estimation of the model parameters requires an effective image registration algorithm [26] 10. Color filter array Based on a simple fact that for authentic images, there will be a high correlation between the pixels and their neighbors Failed in situations where watermarking technologies are not suitable [27] 11. Sensor array Applicable to all cameras Sensor imperfections degrade the performance [28]–[30] Open in new tab Comparison of geometry and physically-based passive methods. TABLE 5. Comparison of geometry and physically-based passive methods. No. Techniques Advantages Disadvantages Ref. no. 12. Principal point A simple and effective approach Not suitable for low-resolution images [31] 13. Skew Known geometry of points is not required Suitable only for frames containing planar surface [32] 14. Metric measurements Improves planar surface transformation Manual interference is required [33] 15. Multiple view geometry Re-projection caused by non-zero skew guides in forgery detection Requires two correlated images [34] 16. Reflection Statistical dependencies enable easy detection of the forgery Only suitable for images with reflection [49] 17. Lighting color Suitable for images with illuminant color Human interference is required [44]–[48] 18. 2D lighting direction Simple and effective technique by estimating the direction of light Estimates 2D lighting properties only [7, 36, 37] 19. 3D lighting direction It preserves the real-time direction oriented parameters to detect the tampering Not suitable for complex lighting environments [35, 38] 20. 2D lighting environment Capable of modeling light model lead to effective estimation of tampering Cannot detect small differences in lighting conditions [39, 40] 21. 3D lighting environment It can able to keep the real environment parameters so it would be very effective Increased complexity [41]–[43] 22. Shadow Minimal assumptions about the underlying scene geometry Only suitable for images with shadow [50]–[54] No. Techniques Advantages Disadvantages Ref. no. 12. Principal point A simple and effective approach Not suitable for low-resolution images [31] 13. Skew Known geometry of points is not required Suitable only for frames containing planar surface [32] 14. Metric measurements Improves planar surface transformation Manual interference is required [33] 15. Multiple view geometry Re-projection caused by non-zero skew guides in forgery detection Requires two correlated images [34] 16. Reflection Statistical dependencies enable easy detection of the forgery Only suitable for images with reflection [49] 17. Lighting color Suitable for images with illuminant color Human interference is required [44]–[48] 18. 2D lighting direction Simple and effective technique by estimating the direction of light Estimates 2D lighting properties only [7, 36, 37] 19. 3D lighting direction It preserves the real-time direction oriented parameters to detect the tampering Not suitable for complex lighting environments [35, 38] 20. 2D lighting environment Capable of modeling light model lead to effective estimation of tampering Cannot detect small differences in lighting conditions [39, 40] 21. 3D lighting environment It can able to keep the real environment parameters so it would be very effective Increased complexity [41]–[43] 22. Shadow Minimal assumptions about the underlying scene geometry Only suitable for images with shadow [50]–[54] Open in new tab TABLE 5. Comparison of geometry and physically-based passive methods. No. Techniques Advantages Disadvantages Ref. no. 12. Principal point A simple and effective approach Not suitable for low-resolution images [31] 13. Skew Known geometry of points is not required Suitable only for frames containing planar surface [32] 14. Metric measurements Improves planar surface transformation Manual interference is required [33] 15. Multiple view geometry Re-projection caused by non-zero skew guides in forgery detection Requires two correlated images [34] 16. Reflection Statistical dependencies enable easy detection of the forgery Only suitable for images with reflection [49] 17. Lighting color Suitable for images with illuminant color Human interference is required [44]–[48] 18. 2D lighting direction Simple and effective technique by estimating the direction of light Estimates 2D lighting properties only [7, 36, 37] 19. 3D lighting direction It preserves the real-time direction oriented parameters to detect the tampering Not suitable for complex lighting environments [35, 38] 20. 2D lighting environment Capable of modeling light model lead to effective estimation of tampering Cannot detect small differences in lighting conditions [39, 40] 21. 3D lighting environment It can able to keep the real environment parameters so it would be very effective Increased complexity [41]–[43] 22. Shadow Minimal assumptions about the underlying scene geometry Only suitable for images with shadow [50]–[54] No. Techniques Advantages Disadvantages Ref. no. 12. Principal point A simple and effective approach Not suitable for low-resolution images [31] 13. Skew Known geometry of points is not required Suitable only for frames containing planar surface [32] 14. Metric measurements Improves planar surface transformation Manual interference is required [33] 15. Multiple view geometry Re-projection caused by non-zero skew guides in forgery detection Requires two correlated images [34] 16. Reflection Statistical dependencies enable easy detection of the forgery Only suitable for images with reflection [49] 17. Lighting color Suitable for images with illuminant color Human interference is required [44]–[48] 18. 2D lighting direction Simple and effective technique by estimating the direction of light Estimates 2D lighting properties only [7, 36, 37] 19. 3D lighting direction It preserves the real-time direction oriented parameters to detect the tampering Not suitable for complex lighting environments [35, 38] 20. 2D lighting environment Capable of modeling light model lead to effective estimation of tampering Cannot detect small differences in lighting conditions [39, 40] 21. 3D lighting environment It can able to keep the real environment parameters so it would be very effective Increased complexity [41]–[43] 22. Shadow Minimal assumptions about the underlying scene geometry Only suitable for images with shadow [50]–[54] Open in new tab Table 5 shows a summary of the comparison of different geometry and physically based passive methods. Also, light source direction estimation in 2D was restricted because it is not possible to determine the 3D surface normals from a single image. In the case of 3D lighting direction estimation [34], there was a common assumption that there was only a single dominant light source. However, in reality, multiple lights can be located at various positions, which lead to a complex lighting environment. This approach was suitable only when the illumination is from a single light source and is not suitable for complex lighting environments, which consists of multiple light sources. The shortcoming of the 2D lighting environment estimation approach is that when several regions of the image are captured in fairly similar lighting conditions, then such small differences in the lighting conditions cannot be detected [35]. The drawback of the 3D lighting environment estimation approach is, since knowledge about the 3D geometry of the suspicious person should be known, an estimation process is required for fitting the 3D model, which increases the complexity and the possible error sources [37]. The approaches presented in [35, 37, 38] are not practically applicable because the lighting environment can be determined only from Lambertian material. Moreover, all the regions under consideration should have similar material. Moreover, in a 3D complex lighting environment estimation approach [39], the accuracy is improved with higher computational complexity for rendering additional images. Also, this approach highly depends on exact 3D face models, and the performance decays with the decrease in the shading prominence. Moreover, for the real-world dataset, this method is not suitable for a complex scene that fails to satisfy the basic assumption. The problems with the illumination color-based approaches [41] are human interference required for selecting the reference block, and this approach is unsuitable for indoor images. Also, illumination-based approaches are naturally more susceptible to estimation errors [42]. Also, illumination color-based approaches fail to detect the forgery image when the chromaticity of the face area was carefully adjusted. The shadow-based approach presented in [46] failed to consider the physical clues provided by the shadows in the real-world images for forgery identification. Another important fact in shadow-based approaches is that these approaches can be used only for regions with shadows [49]. The limitation of reflection-based approaches is that these approaches are applicable only to images that contain reflections [44]. 5. CONCLUSION In this paper, a critical study of the forgery detection techniques in digital images was presented together with its research gaps and recommendation for the future works. The hierarchical classification diagram is provided for the image forgery detection techniques. From the survey, it is apparent that, for real-world images, passive approaches are more beneficial than active approaches, because passive approaches do not require a priori information about the image. Passive approaches are mainly used to detect copy–move, image splicing, image re-sampling and image compression types of image forgeries. Among the several research works proposed for image forgery detection, only a few of them can locate the forged region. Also, for the image forgery analyst to detect forgery, the technique used to tamper the image should be known. So, further research can focus on devising a tool to detect the category of forgery used in the image automatically. From the literature, it is clear that face composite image forgery is a serious issue, and physically based techniques are more suitable to detect such kind of forgeries. The drawbacks associated with physically based approaches are also discussed in this survey, which leads to a new path for researchers. In the future, the performance of the optimized 3D lighting environment estimation method can be further improved by using a recent optimization algorithm and neural network. Along with the optimized light estimation, the geometry information and the texture information will also be included in the general light reflection model. Moreover, by applying clustering and segmentation, the forged region in the spliced image will also be detected. Also, the forgery detection approach will be developed to detect the forgeries in videos, audios and text document. References [1] Schetinger , V. , Iuliani , M. , Piva , A. , & Oliveira , M. M. ( 2016 ) Digital Image Forensics vs . Image Composition : An Indirect Arms Race . 1 – 13 . Retrieved from http://arxiv.org/abs/1601.03239 Google Preview WorldCat COPAC [2] Devi Mahalakshmi , S. , Vijayalakshmi , K. and Priyadharsini , S. ( 2012 ) Digital image forgery detection and estimation by exploring basic image manipulations . Digit. Invest. , 8 , 215 – 225 . Google Scholar Crossref Search ADS WorldCat [3] Redi , J.A. , Taktak , W. and Dugelay , J.L. ( 2011 ) Digital image forensics: A booklet for beginners . Multimed. Tools Appl. , 51 , 133 – 162 . Google Scholar Crossref Search ADS WorldCat [4] Wu , L. and Wang , Y. ( 2011 ) Detecting image forgeries using geometric cues . Computer Vision for Multimedia Applications , 1 , 197 – 217 . Google Scholar Crossref Search ADS WorldCat [5] Wang , X. , Xue , J. , Zheng , Z. , Liu , Z. and Li , N. ( 2012 ) Image forensic signature for content authenticity analysis . J. Vis. Commun. Image R. , 23 , 782 – 797 . Google Scholar Crossref Search ADS WorldCat [6] Spagnolo , G.S. and De Santis , M. ( 2011 ) Holographic watermarking for authentication of cut images . Opt. Laser. Eng. , 49 , 1447 – 1455 . Google Scholar Crossref Search ADS WorldCat [7] Farid , H. ( 2009 ) Image forgery detection: A survey . IEEE Signal Process. Mag. , 26 , 16 – 25 . Google Scholar Crossref Search ADS WorldCat [8] Pun , C.-M. , Yuan , X.-C. and Bi , X.-L. ( 2015 ) Image forgery detection using adaptive oversegmentation and feature point matching . IEEE T. Inf. Foren. Sec. , 10 , 1705 – 1716 . Google Scholar Crossref Search ADS WorldCat [9] Fridrich , J. , Soukal , D. and Lukáš , J. ( 2003 ) Detection of Copy-Move Forgery in Digital Images. In Proceedings of Digital Forensic Research Workshop , pp. 652 – 663 . IEEE Computer Society, Washington, DC , Cleveland , August 6–8 . Google Preview WorldCat COPAC [10] TR2004-515 ( 2004 ) Exposing Digital Forgeries by Detecting Duplicated Image Regions . Department of Computer Science , Dartmouth College, Hanover . Google Preview WorldCat COPAC [11] Weiqi , L. , Jiwu , H. and Guoping , Q. ( 2006 ) Robust Detection of Region-Duplication Forgery in Digital Image. In Proceedings of International Conference on Pattern Recognition , pp. 746 – 749 . IEEE, New Jersey, USA , Hong Kong, China , August 20–24 . Google Preview WorldCat COPAC [12] Zhong , J. , Gan , Y. , Young , J. , Huang , L. and Lin , P. ( 2017 ) A new block-based method for copy-move forgery detection under image geometric transforms . Multimed. Tools Appl. , 76 , 14887 – 14903 . Google Scholar Crossref Search ADS WorldCat [13] Sun , Y. , Ni , R. and Zhao , Y. ( 2018 ) Nonoverlapping blocks based copy-move forgery detection . Secur. Commun. Netw. , 2018 , 1 – 11 . WorldCat [14] Zhou , J. , Ni , J. and Rao , Y. ( 2017 ) Block-Based Convolutional Neural Network for Image Forgery Detection. In International Workshop on Digital Watermarking , pp. 65 – 76 . Springer, Verlag, Berlin , Magdeburg, Germany , August 23–25 . Google Preview WorldCat COPAC [15] Vaishnavi , D. and Subashini , T.S. ( 2015 ) A passive technique for image forgery detection using contrast context histogram features . International Journal of Electronic Security and Digital Forensics , 7 , 278 – 289 . Google Scholar Crossref Search ADS WorldCat [16] Peng , A. , Wu , Y. and Kang , X. ( 2017 ) Revealing traces of image resampling and resampling antiforensics . Adv. Multimed. , 2017 , 1 – 13 . Google Scholar Crossref Search ADS WorldCat [17] Bunk , J. et al. ( 2017 ) Detection and Localization of Image Forgeries Using Resampling Features and Deep Learning. In Proceedings of Computer Society Conference on Computer Vision and Pattern Recognition Workshops , pp. 1881 – 1889 . IEEE, USA , Honolulu, USA , July 21–26 . Google Preview WorldCat COPAC [18] Xu , B. , Liu , G. and Dai , Y. ( 2014 ) Detecting image splicing using merged features in chroma space . Sci. World J. , 2014 , 1 – 8 . WorldCat [19] Shen , X. , Shi , Z. and Chen , H. ( 2017 ) Splicing image forgery detection using textural features based on the grey level co-occurrence matrices . IET Image Process. , 11 , 44 – 53 . Google Scholar Crossref Search ADS WorldCat [20] He , J. , Lin , Z. , Wang , L. and Tang , X. ( 2006 ) Detecting Doctored JPEG Images Via DCT Coefficient Analysis. In Proceedings of the 9th European Conference on Computer Vision , pp. 423 – 435 . Springer-Verlag Berlin Heidelberg , Graz, Austria , May 7–13 . Google Preview WorldCat COPAC [21] Bianchi , T. and Piva , A. ( 2011 ) Detection of Non-aligned Double JPEG Compression with Estimation of Primary Compression Parameters. In Proceedings of International Conference On Image Processing , pp. 1929 – 1932 . IEEE, USA , Brussels, Belgium , Sep 11–14 . Google Preview WorldCat COPAC [22] Luo , W. , Qu , Z. , Huang , J. and Qiu , G. ( 2007 ) A Novel Method For Detecting Cropped And Recompressed Image Block. In Proceedings of International Conference on Acoustics, Speech and Signal Processing , pp. 217 – 220 . IEEE, USA , Honolulu, USA , April 15–20 . Google Preview WorldCat COPAC [23] Ye , S. , Sun , Q. and Chang , E.C. ( 2007 ) Detecting Digital Image Forgeries by Measuring Inconsistencies of Blocking Artifact. In Proceedings of International Conference on Multimedia and Expo , pp. 12 – 15 . IEEE, USA , Beijing, China , July 2–5 . Google Preview WorldCat COPAC [24] TR2006-583 ( 2006 ) Digital Image Ballistics from JPEG Quantization . Department of Computer Science , Dartmouth College, Hanover . Google Preview WorldCat COPAC [25] Hsu , Y.F. and Chang , S.F. ( 2010 ) Camera response functions for image forensics: An automatic algorithm for splicing detection . IEEE T. Inf. Foren. Sec. , 5 , 816 – 825 . Google Scholar Crossref Search ADS WorldCat [26] Johnson , M.K. and Farid , H. ( 2006 ) Exposing Digital Forgeries through Chromatic Aberration, ACM Multimedia and Security Workshop, pp. 48 – 55 . Geneva, Switzerland , September 26–27 , ACM, New York . COPAC [27] Popescu , A.C. and Farid , H. ( 2005 ) Exposing digital forgeries in color filter array interpolated images . IEEE Trans. Signal Process. , 53 , 3948 – 3959 . Google Scholar Crossref Search ADS WorldCat [28] Hongmei , G. , Ashwin , S. and Min , W. ( 2007 , Sept. 16–Oct. 19 ) Noise Features for Image Tampering Detection And Steganalysis. In Proceedings of International Conference on Image Processing , pp. 97 – 100 . IEEE, USA , San Antonio, USA . Google Preview WorldCat COPAC [29] Chen , M. , Fridrich , J. , Goljan , M. and Lukas , J. ( 2008 ) Determining image origin and integrity using sensor noise . IEEE T. Inf. Foren. Sec. , 3 , 74 – 90 . Google Scholar Crossref Search ADS WorldCat [30] Lukas , J. , Fridrich , J. and Goljan , M. ( 2006 ) Detecting Digital Image Forgeries Using Sensor Pattern Noise. In Electronic Imaging, Security, Steganography, and Watermarking of Multimedia Contents VIII , pp. 362 – 372 . SPIE, Bellingham, Washington USA , San Jose, CA , January 16–19 . Google Preview WorldCat COPAC [31] Johnson , M.K. and Farid , H. ( 2007 ) Detecting Photographic Composites of People. 6th International Workshop on Digital Watermarking (IWDW), pp. 19 – 33 . Springer, Verlag, Berlin , Guangzhou, China , December 3–5 . COPAC [32] Wang , W. and Farid , H. ( 2008 ) Detecting Re-Projected Video, International Workshop on Information Hiding, pp. 1 – 15 . Springer, Verlag, Berlin , Santa Barbara, USA , May 19–21 . COPAC [33] TR2006-579 ( 2006 ) Metric Measurements on A Plane from A Single Image . Computer Science Metric , Dartmouth College, Hanover . Google Preview WorldCat COPAC [34] Wei , Z. , Xiaochun , C. , Zhiyong , F. , Jiawan , Z. and Ping , W. ( 2009 ) Detecting Photographic Composites Using Two-View Geometrical Constraints. In Proceedings of International Conference on Multimedia and Expo (ed) June 28–July 3 , pp. 1078 – 1081 . IEEE, USA , New York, USA . COPAC [35] Carvalho , T. , Farid , H. and Kee , E. ( 2015 ) Exposing Photo Manipulation from User-Guided 3D Lighting Analysis. Proceedings of SPIE - the International Society For Optical Engineering. In Media Watermarking, Security, and Forensics (Vol. 940902 ). SPIE, Bellingham, Washington USA , San Francisco, CA, USA , February 9–11 . Google Preview WorldCat COPAC [36] Nillius , P. and Eklundh , J.O. ( 2005 ) Automatic Estimation of the Projected Light Source Direction. In Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR) (ed) June 20–26 , pp. 1076 – 1083 . IEEE, USA , San Diego, USA . COPAC [37] Johnson , M.K. and Farid , H. ( 2005 , August 1–2 ) Exposing Digital Forgeries by Detecting Inconsistencies in Lighting, Proceedings of the 7th workshop on Multimedia and security, pp. 1 – 10 . ACM, New York , New York, USA . COPAC [38] Johnson , M.K. and Farid , H. ( 2008 ) Exposing Digital Forgeries through Specular Highlights on the Eye. In 9th International Workshop on Information Hiding , pp. 4567 – 311–325 . Saint-Malo, France , June 11–13 , Springer, Verlag, Berlin . Google Preview WorldCat COPAC [39] Johnson , M.K. and Farid , H. ( 2007 ) Exposing digital forgeries in complex lighting environments . IEEE T. Inf. Foren. Sec. , 2 , 450 – 461 . Google Scholar Crossref Search ADS WorldCat [40] Zhao , W. and Liu , H. ( 2015 ) An Image Forensic Technique Based on 2D Lighting Estimation Using Spherical Harmonic Frames. In Proceedings of CCF Chinese Conference on Computer Vision , pp. 546 – 325–326 . Springer, Verlag, Berlin , Xi'an, China , September 18–20 . Google Preview WorldCat COPAC [41] Kee , E. and Farid , H. ( 2010 ) Exposing digital forgeries from 3-D lighting environments. International Workshop on Information Forensics and Security. In WIFS , pp. 1 – 6 . IEEE, USA , Seattle, USA , December 12–15 . Google Preview WorldCat COPAC [42] Fan , W. , Wang , K. , Cayre , F. and Xiong , Z. ( 2012 ) 3D Lighting-Based Image Forgery Detection Using Shape-From-Shading, Proceedings of the 20th European Signal Processing Conference (EUSIPCO), Bucharest, Romania, August 27–31 , pp. 1777 – 1781 . IEEE , USA . Google Preview WorldCat COPAC [43] Peng , B. , Wang , W. , Dong , J. and Tan , T. ( 2017 ) Optimized 3D lighting environment estimation for image forgery detection . IEEE T. Inf. Foren. Sec. , 12 , 479 – 494 . Google Scholar Crossref Search ADS WorldCat [44] Riess , C. and Angelopoulou , E. ( 2010 ) Scene Illumination as an Indicator of Image Manipulation. 12th International Workshop on Information Hiding, pp. 66 – 80 . Springer, Verlag, Berlin , Calgary, Canada , June 28–30 . COPAC [45] Wu , X. and Fang , Z. ( 2011 ) Image Splicing Detection Using Illuminant Color Inconsistency. In Proceedings of 3rd International Conference on Multimedia Information Networking and Security , pp. 600 – 603 . IEEE, USA , Shanghai, China , November 4–6 . Google Preview WorldCat COPAC [46] De Carvalho , T.J. , Riess , C. , Angelopoulou , E. , Pedrini , H. and Rocha , A.D.R. ( 2013 ) Exposing digital image forgeries by illumination color classification . IEEE T. Inf. Foren. Sec. , 8 , 1182 – 1194 . Google Scholar Crossref Search ADS WorldCat [47] Carvalho , T. , Faria , F.A. , Pedrini , H. , Da Torres , R.S. and Rocha , A. ( 2016 ) Illuminant-based transformed spaces for image forensics . IEEE Trans. Inf. Foren. Sec. , 11 , 720 – 733 . Google Scholar Crossref Search ADS WorldCat [48] O’brien , J.F. and Farid , H. ( 2012 ) Exposing photo manipulation with inconsistent reflections . ACM T. Graphic. , 31 , 1 – 11 . Google Scholar Crossref Search ADS WorldCat [49] Zhang , W. , Cao , X. , Jiawan , Z. , Zhu , J. and Wang , P. ( 2009 ) Detecting Photographic Composites Using Shadows. In International Conference on Multimedia and Expo (ed) , pp. 1042 – 1045 . IEEE, USA , Cancun, Mexico , June 28–July 3 . COPAC [50] Liu , Q. , Cao , X. , Deng , C. and Guo , X. ( 2011 ) Identifying image composites through shadow matte consistency . IEEE T. Inf. Foren. Sec. , 6 , 1111 – 1122 . Google Scholar Crossref Search ADS WorldCat [51] Kee , E. , O’brien , J.F. and Farid , H. ( 2014 ) Exposing photo manipulation from shading and shadows . ACM T. Graph. , 33 , 1 – 21 . Google Scholar Crossref Search ADS WorldCat [52] Kee , E. , O’brien , J. and Farid , H. ( 2013 ) Exposing photo manipulation with inconsistent shadows . ACM T. Graph. , 32 , 1 – 12 . Google Scholar Crossref Search ADS WorldCat [53] Tuba , I. , Tuba , E. and Beko , M. ( 2016 ) Digital Image Forgery Detection Based on Shadow Texture Features. In 24th Telecommunications Forum (ed) , pp. 679 – 682 . IEEE, USA , Belgrade, Serbia , November 22 and 23 . COPAC [54] Tuba , V. , Jovanovic , R. and Tuba , M. ( 2017 ) Digital image forgery detection based on shadow HSV inconsistency. In 5th International Symposium on Digital Forensic and Security , pp. 1 – 4 . IEEE, USA , Tirgu Mures, Romania , April 26–28 . Google Preview WorldCat COPAC [55] Peng , B. , Wang , W. , Dong , J. and Tan , T. ( 2015 ) Improved 3D Lighting Environment Estimation for Image Forgery Detection. In International Workshop on Information Forensics and Security (WIFS) , pp. 1 – 6 . IEEE, USA , Rome, Italy , November 16–19 . Google Preview WorldCat COPAC [56] CASIA Image Tempering Detection Evaluation Database (CAISA TIDE) V1.0. http://forensics.idealtest.org. [57] MICC-F220 dataset: http://www.micc.unifi.it/downloads/MICC-F220.zip [58] Ucid database: http://jasoncantarella.com/downloads/ucid.v2.tar.gz [59] RAISE dataset: http://loki.disi.unitn.it/RAISE/ [60] USC-SIPI dataset: http://sipi.usc.edu/database/ [61] CoMoFoD dataset: http://www.vcl.fer.hr/comofod [62] http://agents.fel.cvut.cz/boss/index.php?mode=VIEW&tmpl=materials [63] Columbia dataset: http://www.ee.columbia.edu/ln/dvmm/downloads/AuthSplicedDataSet/dlform.html [64] http://www.grip.unina.it/ © The British Computer Society 2019. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Hierarchical Categorization and Review of Recent Techniques on Image Forgery Detection
JF - The Computer Journal
DO - 10.1093/comjnl/bxz148
DA - 2003-10-01
UR - https://www.deepdyve.com/lp/oxford-university-press/hierarchical-categorization-and-review-of-recent-techniques-on-image-ykISE3NWKZ
SP - 1
VL - Advance Article
IS - 
DP - DeepDyve
ER -