One extra zero in a bank check, and the account risks losing $1,000 instead of $100. Fortunately, there are methods that let forensic experts from police departments find out whether the document was falsified, even if it looks perfectly natural. One of them is hyperspectral image analysis.
What is hyperspectral image analysis in general?
Hyperspectral image analysis is a subset of spectral analysis. The technique allows you to examine the properties of objects by obtaining their spectral characteristics. Those characteristics are a spectrum of the reflected signal, the strength of which depends on the chemical and physical properties of the object. A fresh apple has different spectral characteristics than a rotten apple. So do inks in documents, even though they may look identical to the naked eye.
Hyperspectral analysis involves using specialized hyperspectral imaging (HSI) systems that are able to collect spectral information. Such systems capture an object, process information about the wavelength of light within an image, and output the spectral characteristics—unique spectral signatures—for each pixel.
An ordinary photo camera operates with three bands: red, green, and blue within the range of visible light (400-700 nm). With a hyperspectral camera, it’s possible to capture an object in the whole electromagnetic spectrum: e.g., in ultraviolet (UV), visible (VIS), near-infrared (NIR), mid-infrared (MIR), and thermal infrared ranges.
The wavelength range of most solutions used for hyperspectral imaging extends from the visible light spectrum into the NIR spectrum (700-1,000 nm). However, the range of the electromagnetic spectrum a particular system or device can handle depends on its purpose. For example, hyperspectral devices used in pharmaceuticals can go deeper into NIR (1,000 - 2,400 nm), devices used in forensics work in the range from UV to NIR (254-1,000 nm), and medical diagnosis techniques may employ the MIR spectrum (2,500-25,000 nm).
When the data is captured, the system then processes it to generate a color-coded representation of the spectral data, which provides insights into the chemical and physical properties of the object within the image.
Generally, hyperspectral image analysis supports a wide range of applications, including diagnostics in medicine, evaluating food quality, and sensing the environment. Narrowing down to forensics, there are two areas of application for hyperspectral analysis:
Crime scene investigation (e.g., examination of blood stains, fingerprints, paints, etc.)
Forgery detection (e.g., comparison of different inks used for printing and writing)
Below, we’ll illustrate the method in relation to document examination and forgery detection.
How hyperspectral image analysis works for forgery detection
Document fraud often occurs through the misuse of blank documents or alterations of official documents. Hyperspectral analysis allows an expert to chunk the colors and inks used in the document down to the pixel and examine their physical optical properties. Such an examination lets them discover if there were any alterations to the original content, or compare the attributes of a questioned document with a genuine one.
The idea behind the process is that different inks behave differently in various light sources.
Say there’s a personal check written with a blue pen, and then someone adds an extra zero, also with a blue pen. At first glance, this check may look absolutely normal. The difference between the real number and the added one is too small to notice in white light or even in the UV and IR lights traditionally used for this type of verification. However, using hyperspectral analysis, an expert will detect different strengths of the reflected signal because of the different properties of the inks used.
The result of the hyperspectral analysis is visually presented: there are actual photographs of an object taken in different light sources, so any signs of forgery or any other irregularities will be evident.
To apply hyperspectral analysis, police departments use video spectral comparators like the Regula 4308. Such devices are equipped with high-resolution cameras, zoom lenses, numerous light sources (from UV to IR), and a range of viewing filters. They also come with hyperspectral image analysis software.
Here’s the drill:
A questioned object is placed in the comparator. It can be any document: an ID, a bank statement, a receipt, or a handwritten IOU.
An expert defines the necessary wavelength range for examination and sets the step at which it will move. If they set the step at 1 nm, then the camera will take a picture every 1 nm: at 395 nm, 396 nm, etc.
The object is illuminated by different narrow-band illuminators, so the expert can see in real time how the inks behave under certain wavelengths.
The camera documents the reflected signal. When the device has taken all the images, it collects all the data into a file with a hyperspectral data analysis extension.
The expert examines the images and creates graphs to visually present the results.
In the fifth step, the actual analysis and interpretation happen. The expert selects the areas of the document they are interested in—handwritten records, printed text, images, stamps, etc.—and creates graphs for them. These graphs show the percentage of the signal reflected at a particular wavelength. For example, you can create a graph for each digit of a handwritten “1,000” and compare them. In the example above, the last zero differs in its characteristics from the others, so you can conclude that it was added: one hundred turned into one thousand.
Use cases for hyperspectral analysis
Hyperspectral analysis is invaluable for a wide range of forensic science applications. Here are just a few examples of how it can be used for forgery detection:
Identifying a single source of counterfeit currencies for field officers to decide if they should look for one counterfeit cash factory or multiple.
Examination of handwritten and printed records for alterations to find out whether the details, e.g., the date of a will, date of birth on an ID, the expiration date of a document, etc., have been changed.
Identification of the initial content in the altered elements (original date, amount of money, etc.)
Retrieving contents of blotted, crossed-out, or blurred-over records.
Determining a document’s age by examining various details, such as stamp imprints, inks in the signature, signs left by the printer, etc., and comparing them with documents of the same time frame from archives.
Researching and elaboration of standards for internal use. Since there are standards for document protection, the inks used in documents and banknotes must always behave the same.
Advantages of hyperspectral analysis
The beauty of hyperspectral analysis is that the process of imaging is fast and doesn’t require preparing the object to be examined in any way. All you need to do is to place the object, click “Start,” wait a couple of minutes (depending on the chosen range of examination and step size), and the examination file is ready.
With the hyperspectral analysis file, you no longer need to have the object at hand, so you can forward it for further examination or return it to the owner. This is invaluable in cases when an object is submitted for a series of tests: for example, one for alterations and one for handwriting examination.
Another benefit of the technique is that it’s a non-destructive way to obtain valuable insights. Hyperspectral analysis minimizes the risk of destruction of crime traces or the object itself, like in the case of artworks that are sensitive to any manipulations involving chemicals or physical alterations.
Disadvantages of hyperspectral analysis
There are hardly any disadvantages, technically speaking. However, like any science-based tool, hyperspectral analysis has some limitations and considerations to keep in mind. In the case of hyperspectral analysis, it’s the interpretation of results.
While an HSI module works like a charm, allowing you to quickly get an extensive data package for research, the interpretation of the results still depends on the expert. The device draws graphs, the shapes of which allow the expert to conclude whether there were identical inks used or not. However, there are some subtleties.
For example, there may be differences in the percentage of reflected signals. In the case of printed text, this can happen due to the unevenness of the inks in the strokes. For handwriting, this can be even more significant. Ballpoint pens (unlike gel and ink pens, which draw solid strokes) often distribute ink unevenly. There may also be some uninked areas, depending on how the hand moves. All of this affects the strength of the reflected signal, so, at the end of the day, it takes a seasoned forensic analyst to make a competent conclusion.
The bottom line
Hyperspectral image analysis is a powerful expert-level technique for document forgery detection. As forgeries are becoming more sophisticated, it’s crucial for forensic analysts to have as many tools as possible in their toolkits. Hyperspectral analysis can help you back up assumptions with facts, thus supporting the final conclusion.
Here at Regula, we have solid experience in applying hyperspectral analysis, as well as helping national security agencies effectively use this technique for forgery detection. Recently, our hyperspectral analysis module got positive feedback from the Office of the Attorney General of Colombia, the most authoritative institution in forensic examination in the country. Feel free to reach out to us if you have any questions about the method.