*3.1.1. SIFT algorithm*

(1)

**Figure 4.** Pipeline of a fraud detection algorithm.

as morphing.

the number of overlapping blocks is given by:

**Figure 3.** Image processing operations associated with image forgery.

32 Evolving BCI Therapy - Engaging Brain State Dynamics

reflection, lighting adjustment, or color adjustment. In serious cases, intermediator processes can be combined. Postprocesses, such as noise addition, joint photographic expert group (JPEG) compression, or blurring, can be applied to delete all retraces that can be detected in the copy process, such as sharp edges [2]. A broad range of easily available algorithms has

To detect image forgery, an image is first selected (e.g., converted to gray scale). The image is divided into an auction block of nested pixels. The size of the image *m\_n*, size of block *B*, and

The vector is an extractable characteristic in each block. The vector-matching function is highly similar to pairing functions. Known pairing methods include the arrangement of miracle dictionaries on the element vectors and the identification of the nearest neighbor in the tree *Kd*. The similarity between two attributes can be determined on the basis of similarities between different parameters, such as Euclidean length. In the verification step, extreme values are suppressed and holes are filled up through a basic filtration step, such

been proposed to detect replicated images and functions, as shown in **Figure 4**.

He proposed the SIFT algorithm, which could be used to detect and evaluate the geometrical shifts applied to forged displacement copy-and-paste images. The detection procedure involves three steps: In the first step, SIFT functions are extracted and main points are associated. The second step is committed to keypoint compilation and fraud detection. The third step estimates the engineering shifts, if any, that occurred. SIFT can be executed under the conditions of eminent real rate (TPR) and abject fake positive degree ratio (FRE), JPEG compression, and additional noise. In addition, SIFT can accurately estimate different arguments for affine transmutation. **Figure 5** shows different arguments for affine transmutation.

The first attempts to take advantage of SIFT have been reported in [16]. In SIFT, the correspondence of the key indicator is achieved by first identifying the neighbor closest to the best bin [17]. SIFT has been adopted to identify a single copy in the counterfeit image. SIFT descriptors are usually applied to identify keypoints of copied areas instead of blocks, whereas other algorithms cope with object indicators. Although SIFT exhibits excellent detection performance, its false–positive rate remains unknown. In [18], the main SIFT points were extracted from the image and were then associated to obtain the corresponding keypoints. A vote scheme based on vector direction was applied to distinguish between origin and direction. Then, an

complicated automatic reinstallation of duplicate areas hinders the practical applications of these algorithms. We propose a novel algorithm for the detection and description of scale and constant rotation in images. The algorithm is based on SURF and thus has powerful acceleration functions. SURF approximates or even exceeds the proposed thresholds for redundancy, excellence, and sustainability and rapidly performs calculation and comparison. This operation is performed by relying on image confluence. The exit detection and prescriptive prescriptions are based on their strengths (if a Hessian scale is used to detect and describe the established distribution), and kernel methods are simplified to allow the combination of new detection, description, and correspondence. Correspondence between two images of the same view and the objective is partly achieved by using many computers. In this study, photography, three-dimensional reconstruction, image recording, and objective recoding were conducted. The search for a separate image match—the purpose of our research—can be separated into three principal steps. First, points of interest are specified in the characteristic locations of the image, such as angles, points, and plus T-intersections. The most important property of a detection method is its repeatability, that is, its reliability in finding similar indicators of interest under different conditions. Then, each point of interest is represented by a transmitter characteristic. This description must be distinct and must have similar time strengths under noise conditions, mistake detection, and geometrical and photometrical distortions. Finally, vector descriptors are adapted in different images. Correspondence is based on vector distance. Descriptor size directly affects computational time. Thus, fewer dimensions are desired. We aimed to develop an algorithm for the detection and the identification of fraud. We compared the performance of our proposed algorithm with that of a state-of-the-art detection algorithm. Our algorithm exhibits computational time and robust performance. Downsizing after description and complexity must be balanced while providing sufficient distinction. Various detection and description algorithms have been proposed in the literature (e.g., [1–3, 6, 7, 23]). Furthermore, detailed datasets for comparison and standard assessment have been established [8–10]. We build upon the knowledge gained from previous work to better understand the aspects that contribute to algorithm performance. When used in experiments on standard image sets, as well in the application of actual objective recognition, the algorithm exhibited rapid detection and description, as well as distinctive and reproducible performance. While working with local features, stability is the first issue that requires resolution and depends on the expectation of geometrical and photometrical distortions. This turn of events is identified by the possibility changing in conditioning. We concentrate on the detectors and constant descriptions of the balance and rotation of the image. These detectors offer better compromises among the complexity of the functionality and the durability of the distortions that usually occur. The discrepancy and gradient of anomalies and the effects of perspective are secondary to the effect covered by the overall durability of the description [2]. The additional complexity of affine invariance negatively affects sustainability, unless significant changes are anticipated. In some cases, even analog rotation can be abandoned with solutions in a fixed static version of our description. We refer to this ability as "erect SURF" (U-SURF). In fact, in some applications, such as cell robotic navigation or visual guidance, the camera often only revolves around the vertex. Taking advantage of avoidance of the exaggerated stability of rotation in similar events not only increases speed but also increases discriminatory force. As for the photometric, we assumed a simple linear accelerator example with a scaled factor and displacement. Note that our

Rotation Invariant on Harris Interest Points for Exposing Image Region Duplication Forgery

http://dx.doi.org/10.5772/intechopen.76332

35

detection and description do not apply color.

**Figure 5.** Different arguments for affine transmutation.

efficient two-fold sub-window search algorithm (EES) was used to locate duplicated areas within the border box. Finally, a pixelwise partition was identified. The experiment solutions demonstrated that the proposed algorithm remains robust even with background noise and engineering manipulation [19]. He suggested a SIFT algorithm that could detect and then estimate the geometrical transformation applied to forge displacement copy-and-paste images. The detection process involves three steps. In the first step, the SIFT function is extracted and corresponding keypoints are identified. The second step involves the consolidation and detection of fraud. The third step identifies changes that occurred. SIFT has high positive identification rate and low false positive rate even under JPEG image compression and added noise conditions. In addition, it accurately estimates several affine transformation parameters. Refs. [20, 21] suggested a SIFT-established detecting algorithm that can be used to estimate the geometrical transformation applied to the copy. The algorithm begins by converting the suspected image into grayscale. SIFT is then applied to collect image characteristics for the detection of keypoint sources. In SIFT, the keypoint sources are initially adapted in accordance with the characteristics of the vector sum used in the better bin-first algorithms. The potential geomagnetic distortion of the refined areas is estimated on the basis of the assumed paired keypoints by applying RANSACK. SIFT is more robust than intermediary processes even when JPEG compression or noise are added to the processed image. Furthermore, affine transformation is exactly estimated, particularly in larger duplicated areas. A different scenario is to integrate SIFT into copy detection systems [22]. Instead of applying SIFT to detect keypoints, the Harris quicker from SIFT is applied. After all keypoints are revealed, SIFT is applied to generate the descriptive characteristics of extracted features. Then, the *Kd* trees algorithms are applied to match the keypoints to identify duplicated areas. The algorithms can effectively detect copied areas, such as unrotated scanlines or Gaussian noise conditions, that have undergone transformation [5, 22]. Harris detection, which is quicker than SIFT, has been used to detect keypoints. After keypoint detection, SIFT is applied to identify a unique characteristic from extracted keypoints. The *Kd* tree algorithm is then applied to match keypoints to determine duplicate areas. This algorithm can efficiently detect areas, such as scanlines, that have undergone transformation.

#### *3.1.2. SURF algorithm*

SURF has been adopted to detect image editing processes, such as rotation and gradation. SURF is superior to SIFT in detecting image strengths and performs as well as SIFT. The complicated automatic reinstallation of duplicate areas hinders the practical applications of these algorithms. We propose a novel algorithm for the detection and description of scale and constant rotation in images. The algorithm is based on SURF and thus has powerful acceleration functions. SURF approximates or even exceeds the proposed thresholds for redundancy, excellence, and sustainability and rapidly performs calculation and comparison. This operation is performed by relying on image confluence. The exit detection and prescriptive prescriptions are based on their strengths (if a Hessian scale is used to detect and describe the established distribution), and kernel methods are simplified to allow the combination of new detection, description, and correspondence. Correspondence between two images of the same view and the objective is partly achieved by using many computers. In this study, photography, three-dimensional reconstruction, image recording, and objective recoding were conducted. The search for a separate image match—the purpose of our research—can be separated into three principal steps. First, points of interest are specified in the characteristic locations of the image, such as angles, points, and plus T-intersections. The most important property of a detection method is its repeatability, that is, its reliability in finding similar indicators of interest under different conditions. Then, each point of interest is represented by a transmitter characteristic. This description must be distinct and must have similar time strengths under noise conditions, mistake detection, and geometrical and photometrical distortions. Finally, vector descriptors are adapted in different images. Correspondence is based on vector distance. Descriptor size directly affects computational time. Thus, fewer dimensions are desired. We aimed to develop an algorithm for the detection and the identification of fraud. We compared the performance of our proposed algorithm with that of a state-of-the-art detection algorithm. Our algorithm exhibits computational time and robust performance. Downsizing after description and complexity must be balanced while providing sufficient distinction. Various detection and description algorithms have been proposed in the literature (e.g., [1–3, 6, 7, 23]). Furthermore, detailed datasets for comparison and standard assessment have been established [8–10]. We build upon the knowledge gained from previous work to better understand the aspects that contribute to algorithm performance. When used in experiments on standard image sets, as well in the application of actual objective recognition, the algorithm exhibited rapid detection and description, as well as distinctive and reproducible performance. While working with local features, stability is the first issue that requires resolution and depends on the expectation of geometrical and photometrical distortions. This turn of events is identified by the possibility changing in conditioning. We concentrate on the detectors and constant descriptions of the balance and rotation of the image. These detectors offer better compromises among the complexity of the functionality and the durability of the distortions that usually occur. The discrepancy and gradient of anomalies and the effects of perspective are secondary to the effect covered by the overall durability of the description [2]. The additional complexity of affine invariance negatively affects sustainability, unless significant changes are anticipated. In some cases, even analog rotation can be abandoned with solutions in a fixed static version of our description. We refer to this ability as "erect SURF" (U-SURF). In fact, in some applications, such as cell robotic navigation or visual guidance, the camera often only revolves around the vertex. Taking advantage of avoidance of the exaggerated stability of rotation in similar events not only increases speed but also increases discriminatory force. As for the photometric, we assumed a simple linear accelerator example with a scaled factor and displacement. Note that our detection and description do not apply color.

efficient two-fold sub-window search algorithm (EES) was used to locate duplicated areas within the border box. Finally, a pixelwise partition was identified. The experiment solutions demonstrated that the proposed algorithm remains robust even with background noise and engineering manipulation [19]. He suggested a SIFT algorithm that could detect and then estimate the geometrical transformation applied to forge displacement copy-and-paste images. The detection process involves three steps. In the first step, the SIFT function is extracted and corresponding keypoints are identified. The second step involves the consolidation and detection of fraud. The third step identifies changes that occurred. SIFT has high positive identification rate and low false positive rate even under JPEG image compression and added noise conditions. In addition, it accurately estimates several affine transformation parameters. Refs. [20, 21] suggested a SIFT-established detecting algorithm that can be used to estimate the geometrical transformation applied to the copy. The algorithm begins by converting the suspected image into grayscale. SIFT is then applied to collect image characteristics for the detection of keypoint sources. In SIFT, the keypoint sources are initially adapted in accordance with the characteristics of the vector sum used in the better bin-first algorithms. The potential geomagnetic distortion of the refined areas is estimated on the basis of the assumed paired keypoints by applying RANSACK. SIFT is more robust than intermediary processes even when JPEG compression or noise are added to the processed image. Furthermore, affine transformation is exactly estimated, particularly in larger duplicated areas. A different scenario is to integrate SIFT into copy detection systems [22]. Instead of applying SIFT to detect keypoints, the Harris quicker from SIFT is applied. After all keypoints are revealed, SIFT is applied to generate the descriptive characteristics of extracted features. Then, the *Kd* trees algorithms are applied to match the keypoints to identify duplicated areas. The algorithms can effectively detect copied areas, such as unrotated scanlines or Gaussian noise conditions, that have undergone transformation [5, 22]. Harris detection, which is quicker than SIFT, has been used to detect keypoints. After keypoint detection, SIFT is applied to identify a unique characteristic from extracted keypoints. The *Kd* tree algorithm is then applied to match keypoints to determine duplicate areas. This algorithm can efficiently detect areas, such as scan-

SURF has been adopted to detect image editing processes, such as rotation and gradation. SURF is superior to SIFT in detecting image strengths and performs as well as SIFT. The

lines, that have undergone transformation.

**Figure 5.** Different arguments for affine transmutation.

34 Evolving BCI Therapy - Engaging Brain State Dynamics

*3.1.2. SURF algorithm*
