Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Digital image forensics : theory and implementation
Roy A., Dixit R., Naskar R., Chakraborty R., Springer International Publishing, New York, NY, 2020. 89 pp.  Type: Book
Date Reviewed: Sep 14 2021

The introduction promises a great deal for an 89-page volume: the text “investigate[s] which device (or class of device) captured or formed the image” with “several state-of-the-art techniques ... reviewed in detail”; and seven kinds of image forgery detection are detailed, ranging from copy-move image forgery (where the source and target are in the same image) to copy-paste image forgery (different images) to fully computer-generated images.

Alas, reality is different. There is only one chapter (2) on camera source identification, and it only deals with identifying the brand of camera, not the model or the device. There are indeed four chapters (3 through 6) on image forgery detection, but they all deal with copy-move image forgery. Furthermore, the writing is confusing in places, and occasionally wrong: Figure 3.1 is described as “a forest scene in which lions are moving,” whereas it looks like a savanna scene in which zebras are moving.

Chapter 2’s experiments are conducted with the Dresden image database, and their best classifier achieves 99.1 percent among the ten camera brands here. They then validate this against a different dataset, which has two of the same brands and achieves 97.1 percent accuracy. A significant weakness is that these datasets are from digital cameras; these days the vast majority of images come from mobile phones, and these are not studied or even mentioned.

Chapter 3 states, without explanation, that images are to be converted from color to grayscale first. This would seem like quite a loss of information. Chapter 3 concludes by stating that “a three-way parameterization technique has been proposed,” but I couldn’t really see one. There is advice depending on what one is looking to minimize (false positives, false negatives, and so on), but that’s not really a parameterization technique. The chapter claims to have rigorous experiments, but very few details are given.

Chapter 4 is also on copy-move detection. It’s very short, uses different experiments from chapter 3, and seems to have been written in isolation.

Chapters 3 and 4 both compare blocks in the images. Chapter 5 states, for the first time, that this is only one approach and discusses, in isolation, another approach via keypoint matching, which is claimed, without evidence, to be more efficient. This chapter also discusses the similar but genuine objects (SGO) problem, which previous chapters ignore. The experiments are again all different, and the only metric considered is accuracy, which is not very helpful if one is looking for rare forgeries where one would be more interested in measures like false positives and false negatives.

Chapter 6 takes us back to block-based methods of copy-move detection, this time using dyadic wavelet transforms (DyWT), which were also used in chapters 3 and 4 though neither chapter is mentioned here. Like chapter 3, this one converts to grayscale first. The experiments are different from those in previous chapters.

The conclusions are a disappointment. Despite having four chapters on copy-move detection, there is no real comparison. No source code is given or pointed to, and the descriptions of the data sources, when present, are too vague to enable the reader to reproduce the experiments.

Reviewer:  J. H. Davenport Review #: CR147354
Bookmark and Share
  Featured Reviewer  
Would you recommend this review?
yes
no

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright © 2000-2021 ThinkLoud, Inc.
Terms of Use
| Privacy Policy