|
| Reference | Framework | Limitations |
|
| Bilal et al. [1] | SURF descriptor along with mDBSCAN clustering technique was employed to locate the forged area in a given image. This method exhibits better CMFD performance. | The approach is unable to detect the manipulation from the flat regions of the image. |
| Roy et al. [43] | The SURF descriptor together with the RLBP approach was utilized for keypoint computation, while the g2NN method was used for similarity measurement. Finally, hierarchical clustering [44] approach was utilized to cluster the manipulated part of the input image. The method works well under the presence of postprocessing operations. | The approach exhibits poor detection accuracy over samples of low quality. |
| Alkawaz et al. [9] | In this method, correspondence between the DCT coefficients was computed by employing the Euclidian distance formula to identify the forgeries from the input images. This framework exhibits better CMFD performance. | The false choice of block size can lead to a serious reduction in detection accuracy. |
| Bilal et al. [10] | In this approach, two methods, namely, SURF and BRISK, were used to compute the image features. The hamming distance was computed to measure the similarity between the keypoints. And the DBSCAN clustering approach was employed to localize the altered content. This method is robust to image transformation operations. | The intense changes in scaling, brightness, and color reduction may degrade the detection performance. |
| Bi et al. [45] | This work employed SIFT descriptor for features computation along with an adaptive patch matching algorithm for similarity measurement. The work performs well for CMFD. | This technique is computationally complex. |
| Chen et al. [46] | A block-based CMF detection approach, namely, the BSMRG algorithm, was applied to identify the altered image patches. The method is computationally efficient. | The performance of this method is highly dependent on the block size. |
| Muzaffer and Ulutas [47] | This method used a SIFT descriptor-based approach for CMFD where binarized descriptors were employed to identify the manipulated regions. This work is computationally less expensive. | Performance needs further improvements. |
| Tian et al. [30] | After dividing the image into small blocks, the ORB algorithm was employed over each block to compute the features. Then, the cosine and Jaccard distance metrics were used to measure the similarity to locate the CMF. The approach exhibits better CMFD accuracy. | Performance degrades for the samples with huge scale variations. |
| Abdel-Basset et al. [32] | In this work, the SIFT approach together with the density-based spatial clustering technique was used to identify the forensic changes. The method is robust CMDF and exhibits better detection accuracy. | Unable to locate the changes made within flat image regions. |
| Islam et al. [33] | A deep learning-based framework DOA-GAN was introduced to locate the image forgeries. The approach shows better manipulation detection accuracy even under the occurrence of postprocessing attacks. | This technique is economically inefficient. |
| Niyishaka and Bhagvati [34] | In this technique, the LoG was applied to compute the blobs of the input sample. Then, BRISK features were computed from each blob, and Euclidian distance was computed among them to locate the matching areas. The approach shows better performance to CMFD. | The performance of this method degrades for samples with a large background area. |
| Soni et al. [35] | In the presented framework, the SURF algorithm along with the MSER technique was applied to detect the digital alterations from the input samples. The method performs well for CMFD under the occurrence of noise and light alterations. | Not robust to detect the multi-CMF attacks. |
|
