The vast majority of the work done in Forensics with Photoshop is image enhancement. We want to be able to show the differences that are there in a more compelling or useful manner, or to isolate components of an image so they can be analyzed.

It is important that any such work be scientifically valid and especially that it be repeatable… that anyone with the same set of tools and the list of operations can produce the same result.

We offer two different packages that bring useful (and different) functionality to the world of Forensic Imaging: FoveaPro and Optipix.

This ninhydrin shows a problem when trying to discern magenta (the fingerprint) from blue (the ink) on the back of a check.

Enhancement by leveling the image and stretching the contrast can show the contrast better in the image, but this does not completely isolate the fingerprint.

By transforming the image into HLS space (Hue, Luminance, Saturation) and performing a histogram equalization step in just the Hue channel, we can make the colors of the ink and ninhydrin enough different (in this case complimentary colors) so that we can isolate just the portion of the image that originally had been magenta. This step was performed using FoveaPro' s ColorSpace function.

There are many types of contrast enhancement. There are basic tools in Photoshop (Levels and Curves) that can be used to change the perceived differences between similar brightnesses within image.

This is the Curves function in Photoshop. In essence, it can allow you to compress and expand contrast at different brightnesses within your image.

In this, Curves was used to change a histogram that was mostly gray (with a big peak in the middle of the range) into something that more effectively covered the entire brightness range, without clipping data at the ends. This S-shaped curve is a common tool and good at stretching midtones and compressing bright and dark tones in an image.

The process can make it much easier to see midtone detail that is actually present. The goal isn't the same as in Photography – we're not trying to make a pretty picture, but rather a picture that shows critical information that is already present.

However, simple contrast enhancement is often not enough. FoveaPro and Optipix provide better tools for looking at local differences, not just changing global contrast in an image:

In this fingerprint image, the print is on both a bright and dark background. It is not possible to simply adjust the contrast to isolate the print. Using the function in FoveaPro called Adaptive Equalization we can isolate small local contrast (in this case the ridges):

Now, the overall contrast of the image has been suppressed while the local differences are shown.

This image shows the ability to enhance fine detail in an image using Optipix without the bad side-effects of Unsharp Mask. This function will isolate and enhance DETAIL without causing large halos around strong edges.

Original Image (compressed 5:1, click here for uncompressed, 536K )

Compressed Image (15:1, JPEG)

One of the reasons that this kind of compression looks reasonable is because we recognize familiar objects. However, if you look closely at the centers of the flowers, at the stems, and at the colors there are very large differences induced, simply because of the compression.

This compression is even more obvious when we look just at the hues that are within the image:

Hue channel from original image (compressed 5:1, click here for uncompressed, 296K )

As you can see, even at fairly reasonable compression levels like 15:1, there are disastrous changes to the colors present in the image.

JPEG Compression Artifacts Checklist:

This example shows how adding several (still) frames together can improve the S/N (Signal to Noise Ratio) in an image. Notice how the detail in the brick wall has improved to the point that you can tell it is a brick wall, or that the “Public Library” letters are now complete. In multiple frames, the random noise tends to cancel out, thus the amount of signal is more versus the noise in the image. The first image might have 3-4 real bits of information per pixel, but second image might have 5-6 bits of real information per pixel.

Optipix also has a Refocus function that can be used to try to correct slightly out-of-focus images. However, this technique only works when there is very little noise. Frame Averaging is the only safe technique for reducing noise prior to using Refocus.

FoveaPro has a better tool for deconvolving an image, but it requires that you capture the Point Spread Function from the scene. This is also regarded as the “Hubble Trick” which is a technique for fixing out-of-focus images when you can measure how they're out-of-focus.

There are a number of other noise removal tools, both within Photoshop and in other packages.

Photoshop CS2

Gaussian Blur: This is good at removing shot noise from your image. Unfortunately this improvement comes at the cost of your edges and detail.

Surface Blur: This is a smarter version of a Gaussian blur. It will try to preserve edges. Fine detail will still go away.

Median Filter: This can remove noise in uniform areas, but it will round corners off and remove fine lines.

DespeckleAdobe has not documented how this function works. We recommend that you do not use it.

Dust and Scratches: Adobe has not documented how this function works. We recommend that you do not use it.

Optipix

Frame Averaging: This can average a series of images (up to 30,000 frames) to improve your signal to noise ratio (S/N) and allow you to both see finer brightness details, sharpen the image, and potentially use Refocus or Deconvolution

Safe Median: This is a smart version of Adobe's Median Filter. Corners and fine lines are preserved, hence the name “Safe.”

JPEG Cleaner: This will blur the tile boundaries in JPEG images. The result is a more soft image, but there is a large debate whether that information was actually real.

FoveaPro

Conditional Smooth: This is a blur that uses a disc as its basis, but also has a threshold setting for preserving edges. It can also be called an “Edge Preserving Blur.”

Color Median: This is a round median that preserves colors. It is similar to Adobe's median, but does not make large square artifacts on your image.

Hybrid Median: Similar to the Safe Median in Optipix

Hit Or Miss: This lets you control the shape of the neighborhood with a median filter. It is extremely useful in removing interlace artifacts, møiré patterns, etc.

Rolling Ball: A classical method of removing impulse noise in the literature.

There are other 3rd party tools (e.g. Noise Ninja and Fred Miranda), but we can provide implementation details for legal purposes.

“Forensic video analysis is the scientific examination, comparison and/or evaluation of video in legal matters.”

While most of our tools are oriented toward working on a single frame for analysis, it is well worth remembering, especially in the case of MPEG, that there are I-Frames (also called “Key Frames”) that are essentially lower-compression JPEG images and then there are B-Frames and P-Frames that contain either partial or predicted information that can be limited or incorrect. Unless you are presented with very high quality motion video, you will spend most of your time working with still images. Photoshop, Optipix, and FoveaPro are the best tools for enhancing and measuring images in a completely reproducible manner.