검색 상세

Forensics Analysis on Digital Images: Forgery Detection

Forensics Analysis on Digital Images: Forgery Detection

초록/요약

Forensic analysis is the utilization of limited information to investigate fraudulent activity or suspicious object or steps. The term forensic analysis used in different contexts with slightly different meanings. For digital images, it refers to all the procedures and processes used to prove whether a digital image is not an authentic representation of a reality. Putting it simply, it is an investigation to check if an image is corrupted. Multimedia Forgery is the action of developing an adapted copy of a document with the intent of deceiving or altering the public perception about the media. When the forgery activity is on digital images, we refer to it as digital image forgery. Digital image forgery is the process of manipulating original photographic images to create tampered or forged copies. Various digital image forgery techniques exist; however, copy-move, image splicing, photo retouching and cropping forgery are the common forgery types. Copy-move forgery (CMF) is the most widely used technique for image tampering. CMF is created by copying a portion of an image and pasting it within the image either to create duplicates of some objects of the image or to hide some objects from the image. Researchers have proposed many CMF detection methods, yet each has its own limitations for use in the real-time application. For instance, some are very fast but their performance is poor, while others perform very effectively, but they are very slow and require considerable resources. Two broad groups of forgery detection methods exist based on the method of their descriptor designs. These are distribution-based descriptors and binary descriptors-based CMF detection methods. The distribution-based algorithms are also called histogram-based (HB) methods. The distribution-based approaches represent visual information using distributions of image measurements such as gradients and gradient orientations. The binary methods encode image patch appearance using a compact binary string. The binary methods encode image patches using a binary string computed by sampling or comparing pixel-pairs in the patch. The distribution-based descriptors are proven to be effective in a wide range of applications. However, their size, issues associated with efficient searching operations, and time complexity are among the challenges of distribution-based methods. Though the binary descriptors may not be as descriptive as their histogram counterparts may. However, the binary descriptors have a compact size, low memory footprint and low computational complexity. This work proposes an optimized copy-move forgery detection method based on the learned arrangement of three patch Codes. In the proposed method image patches are used to compute binary descriptors for a detected keypoint instead of using pixel-pair comparisons as in the conventional binary descriptors. The proposed method benefits from the advantages of using both binary descriptors and the distribution (histogram methods) descriptors to extract features used for CMF detection algorithms. The proposed method minimizes the performance gap between the fast but low performing binary descriptors-based methods(BDs), and the slow but efficient distribution-based CMF detection methods. Experimental results prove that the proposed method can perform competitively to many of the existing distribution-based CMF detection methods including SIFT and SURF while it demands a significantly smaller processing time than the histogram-based CMF detection methods. In this work, we also conducted extended experimental tests on various detector-extractor pairs to suggest improved copy-move forgery detection algorithms.To our knowledge, it is the first supervised learning approach where a sparse binary-descriptor coding method is used for CMF detection

more

목차

1. Introduction 1
1.1 Digital Image Forgery 1
1.2 Types of Digital Image Forgery 3
1.2.1 Copy-move Forgery (cloning) 3
1.2.2 Image Splicing / photomontage 5
1.2.3 Photography Retouching 7
1.2.4 Cropping Forgery 7
1.2.5 Image Resampling 8
2. Related Work 10
2.1 Digital Image Forgery Detection 10
2.2 Copy-move Forgery Detection 13
2.3 Limitations of Existing CMF Algorithms 25
3. Proposed Method 27
3.1 Descriptor Computation 27
3.2 Evaluation Metrics 31
3.2.1 Image-Level Evaluation Metrics 32
3.2.2 Pixel-Level Evaluation Metrics 32
3.3 Dataset and Reference Methods 34
3.4 Implementation and Parameter Settings 35
4. Experimental Results 37
4.1 Evaluation Results 37
4.1.1Results at Image Level 37
4.1.2 Evaluation at Pixel Level 41
5. Conclusions 48
5.1 Summary of contributions 48
5.2 Future Works 51
5.3 List of Publications 52

more
반출 Meta View 목록