Volume Number: 21 (2005)
Issue Number: 2
Column Tag: New Technology

Imaging In Forensics

by George Reis

Imaging in Forensics - An Overview

Been to a crime scene lately? Chances are that the Crime Scene Investigator was using a digital camera. Within minutes, those images may be on a shared network and available for a detective to view, photographs of the fingerprints will be enhanced and searched for matches in the state's database, and images from the security video made available to the press.

Although many agencies throughout the US have not yet moved to digital imaging, many have, and have made substantial use of some of the very cool technology available. Let's take a look at some of the technology and applications of digital cameras, computers and image processing in the forensics world. But first, let's see how this technology evolved from the early days of digital imaging forensics in the 1990's.

A Little History

I was fortunate to be among the early adopters of digital imaging in forensics. This was due as much to luck as anything else, one of those issues of being in the right place at the right time. My background was in photojournalism, and I made an early attempt at freelancing. After a few years, I found that working for someone else would make my mortgage company (and my wife) much happier, so I found work in a commercial photo lab. After a couple of years of smelling like photo chemicals, I answered a classified ad for a "Police Photographer" at the Newport Beach Police Department. That was in 1989, well before the CSI television shows.

In the early 1990's when I was first exploring how digital imaging technology could be used in forensics, information was hard to find, tools were few and far between, and Macs were still the most popular computer for graphics. The Newport Beach Police Department was a Mac lab at the time, but also ran an HP 3000 for our statistics. There wasn't a Mac at every desk, and the secretaries still used typewriters, but we did have a couple of dozen Macs distributed throughout the agency. I wanted one, and I wanted Photoshop, but it was hard to sell management on the idea. I learned to write Quiz reports using a dumb terminal on the HP 3000, and eventually was able to upgrade to an Apple Quadra 950 with a copy of Photoshop 2.5.

Digital cameras were pretty rare in those days. In 1992, Kodak introduced the DCS 200, which was a 2 megapixel digital SLR with an 80 MB hard drive built-in, that went for $10,000. Sony had the ProMavica, which was a 3 chip camera with a resolution of 640 X 480 that cost about the same. Quite a difference from the camera options and pricing of today.

Imaging and Fingerprints - First Steps

There were a handful of people in forensics experimenting with digital imaging at that time. We were mostly looking at ways to enhance fingerprints. We struggled with some very basic concepts, like how to size a digital photograph of a fingerprint to 1:1, or how to trace a fingerprint to enable it to be searched within the Automated Fingerprint Identification System (AFIS).

This need to trace fingerprints was important, as the technology for entering a latent fingerprint for a database search was relatively arcane at the time. NEC was a primary AFIS vendor. In California, most counties had a database of fingerprints of those arrested within their county, and the state maintained a database of fingerprints for the state. The computers to search these databases required an 8.5 inch floppy to boot-up, and had a low-resolution scanner built into the workstation to scan a physical tracing of a fingerprint. Searches often took two to three days to get results (results are usually within 15 minutes today). The method for creating this tracing was almost comical - a latent fingerprint would be photographed with a Polaroid MP4 copy camera to a 5X enlargement. Then a sheet of tracing paper would be taped to the Polaroid and the ridges were traced by hand with a .7mm pencil. When the tracing was complete, it would be photographed at 1:5, to bring the 5X tracing back to the print's original size. Then, this Polaroid photograph of the tracing of the fingerprint was scanned into the AFIS, the tracing would be cleaned up, additional information entered for the search, and three days later there would be roughly a ten percent chance of a match. That is, 90 percent of the fingerprints enlarged, traced, reduced and entered into the system would not match a print in the database - it may have been the victim's print, the print of a customer at the business that was robbed, not enough detail in the print to generate a match, etc.

At the time, this technology seemed wonderful - it sure beat the cartooning method (an examiner at one agency would describe, by phone, what the fingerprint looked like to another examiner to see if they may have seen one like it!). But, it is obvious to anyone now that this method needed improvement - virtually no enhancement of the fingerprint could be done. It was possible to make minor global contrast or brightness adjustments through exposure and development times (this was the peel-apart Polaroid material), or by choosing the high contrast Type 55 film. The precision of the sizing of the enlargement and reduction was far from accurate. It is no wonder that finding a method to do this digitally would lead to a higher number of fingerprint matches by having better fingerprint detail and more accurate image sizes. But, simple things like figuring out how to make a digital tracing of the fingerprint before Photoshop had layers wasn't obvious to us back then. We weren't computer experts, we weren't Photoshop experts, we just had jobs that involved photography and fingerprints and we knew that computers could help us do a better job - and we managed to make it work.

Image Analysis

Over the years, image processing in forensics expanded to more areas, including questioned document examination; footwear, tire impression and tool mark examination; blood spatter evidence; bullet striation and primer mark examination, etc. In addition, video analysis moved to the digital realm, bringing with it a new set of experts and tools.

Image analysis and enhancement today is quite different than what we were attempting in the early 90's. Now, we can do quite a bit of non-destructive processes using adjustment layers, we can localize enhancements with layer masks, we can utilize Photoshop's Channel Mixer to extract a fingerprint from a background or to find detail that wasn't visible to the eye. Photoshop plug-ins like FoveaPro and Optipix from Reindeer Graphics give us the capability to apply a Fast Fourier Transform to eliminate patterned backgrounds, apply deconvolution to correct for motion blur or poor focus, and utilize frame averaging to eliminate noise from a sequence of images. Plug-ins like Human Software's PhotoFixLens enable us to correct lens distortion.

With the introduction of the History Log in Photoshop CS, the ability to have an audit trail stored in the file's metadata has been a huge benefit in forensics. There is no court requirement that one have an audit trail, but it is a key feature in validating that good techniques were used and makes it possible to easily repeat most enhancements. The History Log can be invoked through the General Preferences pane in Photoshop CS. The metadata recorded can be viewed by choosing History from the File Info dialog box or in the Metadata tab in Photoshops File Browser.

The History Log is off by default. It can be activated in Photoshop's General Preferences pane, with options for saving in the file's metadata, as a separate test file, or both. The History Log can record sessions only, a concise record, or detailed listings of the tools used and parameters set.

If the History Log is recorded as file metadata, it can be viewed in File Info from the File menu.

An objective in image analysis is often to extract some image data so that it can be seen with the best possible detail. Processes for extracting this information my be as simple as increasing contrast globally or to local areas of the image, increasing the sharpness, controlling the color channels to build contrast between the print and the background, removing a patterned background, etc.

The introduction of Adjustment Layers in Photoshop a few versions ago enabled many adjustments to be made non-destructively. Prior to this, making a Levels Adjustment or a Curves Adjustment would frequently result in the loss of some image data. With The problem with this is that if additional image processing is desired, or if modifications to the original adjustments are needed, it is too late. Adjustment layers resolved this. In addition, Adjustment Layers also retain the adjustment settings, so even if the History Log is not active, or if one is using an older version of Photoshop, the audit trail exists.

As they say on the knife commercial, "but wait, there's more." And there is. Masks can be used with Adjustment Layers to isolate the adjustment to localized areas of the image. This provides substantially more power to this feature, allowing different areas of the image to be treated separately, while maintaining a visual mask to illustrate precisely what areas of the image were affected by an adjustment.

The top image is uncorrected, the bottom image has been given a Levels Adjustment Layer with a Layer Mask, so that the background could be lightened, while keeping the foreground from becoming too bright.

One of the most powerful tools used for image enhancement tasks is the Channel Mixer. This tool allows one to combine the channels of a color space to be mixed in different percentages to affect the tonal contrast between various colors. In forensics, this can extract a fingerprint from a busy background, isolate an endorsement signature from the bank or store imprinting, or even recover "invisible" writing from stolen checks that have been bleached and rewritten.

In this image, the signature is overwritten by the endorsement. Using the Channel Mixer, the color channels can be adjusted separately to eliminate the color in the endorsement, and isolate the color in the signature, as shown in the bottom image.

Video analysis has several unique features unto itself. Analog video is an interlaced signal - two fields make up a single frame. The odd field contains every other horizontal line of the image, and the even field contains the remaining lines. Combining the two fields makes one frame. In security systems, several camera views are often recorded onto the same tape using a multiplexer, which allows the recording to sequence from camera to camera in a number of different methods. Some systems will even record different cameras on each field of a single frame! Deinterlacing is a readily available tool, an can be done directly in Photoshop and most video applications.

The image on the left is an example of two cameras each being recorded onto separate fields of the same video frame. Choosing Photoshop's Deinterlace filter enables one to separate the two fields, which are separated on the right of this image.

Separating each camera view into separate movie files is called de-multiplexing, or deplexing. Multiplex systems use hardware to do this task, but it compresses the files, which can ruin the value of the images. Several Windows systems offer the ability to de-multiplex virtually automatically within their software. On the Mac, this can be done manually by exporting a movie file to a still image sequence and manually sorting the files - neither an elegant nor an expedient method.

One aspect of video that is advantageous is that 29.97 frames of video are recorded per second. The advantage of this is that information from several images can be combined to improve the image data. Low light imaging produces a very low signal to noise ratio. The image noise can make an otherwise sharp image virtually useless. Through frame averaging - taking the average value of each pixel in a set of images - the noise can be greatly reduced, thereby, providing information that had been impossible to see. Reindeer Graphics makes a nice Photoshop plug-in for frame averaging. A key to getting the best value from frame averaging is to have each image precisely aligned. Unfortunately, there are not any good tools for image alignment available on the Mac platform at this time. Combining multiple images as separate layers in Photoshop and using the Difference mode is helpful, but limited, especially if there is rotation in the image, or if sub-pixel movements are necessary.

The image on the left is a video frame with too much noise to be able to read the license plate. The image on the right is a Frame Average of five video frames using the Reindeer Graphics Optipix plug-in.

Photography for Documentation

The other main use of digital imaging in forensics application is documenting a crime scene, traffic accident or piece of evidence. In this instance, an analyst isn't comparing one image to another, but attempting to use photography to illustrate to a detective, attorney, judge and/or jury what they saw. Photography has been used for this purpose since at least 1859.

The weaknesses of the early digital cameras were a lack of resolution, high prices and focal length limitations. It wasn't until the mid to late 1990's that digital cameras were exceeding 3 megapixels and moving to an affordable price range. For this reason, evidence photography for documentation continued to be film-based, even for agencies using image processing for image analysis. In many agencies, film was scanned and made available to others for viewing on a network. Prints were made with dye sublimation printers or mini-lab equipment with digital capabilities.

Today, thousands of agencies are using digital cameras to document crime scenes, traffic accidents and photograph evidence.

The key issues involved in evidence photography are to create images that represent the subject matter, maintain a chain-of-custody, and that images can be shared with those who need them (detectives, investigators, attorneys, courts). Because each agency is autonomous and has different staffing and volume, the approaches may vary - but the general concepts are basically the same.

Generally accepted best practice recommendations in forensics are that access to digital images should be limited and that image processing should only be done to copies of the original.

Images stored on a server can restrict access through permissions. Some systems allow one to set security levels to allow specific users to have read and write permissions but not rewrite capability. This helps guarantee that original images cannot be changed.

Asset management applications are not always a part of a forensics workflow. A primary reason for this is that images tend to belong to specific cases, so a simple hierarchical filing system works well. If an agency either also has public relations photographs or wants to search their images by criteria such as crime type or reporting district, an asset manager may be used. Photoshop's File Browser may be adequate to meet some needs, and when something more substantial is needed, an asset manager like Extensis Portfolio may be used. A few companies have created Windows-based asset management applications directed specifically toward the forensics market - some offering image authentication and/or encryption as part of their package. These, however tend to be expensive solutions that don't really offer needed features.

This is Photoshop's FileBrowser - on the left are tabs for navigation, a preview image, metadata and adding and searching by keyword, the right side shows thumbnails of any given folder. All pane sizes are adjustable.

It isn't uncommon for crime scene photographs to need some basic image adjustments to correct for incorrect exposures and color shifts from fluorescent or tungsten light.. Most of this is easily done in Photoshop with simple adjustment layers. Localized adjustments can be accomplished with layer masks. As mentioned in the image analysis section, the advantages of non-destructive editing that adjustment layers bring are substantial in a forensics workflow.

The PhotoFixLens Plug-in from Human Software allows one to correct the barrel distortion of wide-angle lenses. The use of wide-angle lenses is common, and often necessary in a forensics environment. The problem with this, however, is that wide-angle lenses are subject to barrel distortion. Fortunately, this can be corrected, so that straight walls don't appear to have curved surfaces.

In the top image the garage door appears warped because of barrel distortion from using a wide-angle lens. Using PhotoFixLens from Human Software, this can be corrected, as shown in the bottom image.

Multiple photographs are frequently taken to show specific aspects of a piece of evidence. For instance, bodily fluids will fluoresce under UV light, but this lighting makes it difficult to discern the object - only the fluorescence is visible. Another example is photographing a bullet trajectory in the dark, using a laser - the trajectory shows, but the background goes black. To resolve this issue, photographs are taken of the subject with normal lighting, and again under the specialized lighting required in the specific instance. The two images can then be combined as separate layers in Photoshop using the Lighten mode, shich displays the lightest values from each layer.

The top left photograph is of a living room in daylight with flash. The photo beneath it shows a bullet trajectory, made by using a laser pointer with no room illumination. The image on the right combines these two images, and shows the well-lit room with the trajectory by using the Lighten Mode in Photoshop's Layers.

Summary

This article attempted to show some of the uses and tools used imaging by the forensics community. Image processing techniques and digital photography have contributed a lot to the toolbox in this field. Although there isn't yet a "CSI" button on the keyboard, there are many capabilities that were impossible just a few years ago. At the same time, there is a great opportunity for developers to create more tools, and to make those we have more efficient. Hopefully this overview will spark an interest in the Macintosh developer community to build more tools in forensics applications.

And, now the next time you see a Crime Scene Investigator with their digital camera, you'll have a pretty good idea of what will be happening with the images that they take.

George Reis is a Forensic Imaging expert specializing in Digital Imaging Analysis, Enhancement, Photography, Training, and Consulting in the investigative, law enforcement, and legal communities. He can be found at http://www.imagingforensics.com