CLPEX Chat Board

My Bureau have recently updated our AFIS and photoshop CS2 was included. Are there any official guidelines as to which tools would be regarded as enhancement tools and which tools would be considered manipulation.

The most "official" guideline I would use is that written by SWIGIT. They have a sample SOP that states the following when dealing with enhancements of images:

Process the working image using enhancement techniques. All processes applied to the working image are recorded using the enhanced image history tool. Approved processing techniques for use on working images are those that have direct counterparts in traditional darkrooms including brightness and contrast adjustment, dodging and burning, and color balancing. The tools include Brightness/Contrast, Levels, Curves, Color Balance, Hue/Saturation, and Invert. Using Mode, Channels, and Fast Fourier Transform filters (FFT) are acceptable. The following tools are prohibited: Rubber Stamp, Airbrush, Paintbrush, Paint Bucket, Eraser, and Blur.

Basically anything that truly alters the image is prohibited. Photoshop or other imaging programs are used to clean or enhance the image without adding any artifacts. Most procedures in digital can be equated to silver-based photography and darkroom techniques. It is a good idea to document any enhancements and save the original to be retrieved, if necessary for future needs.

Hopefully that helps...

I suggest that we follow best practices in a digital imaging workflow, which I suggest include:

1) Archive the original file
2) Only work on duplicates of the original
3) Only use valid forensic procedures
4) All processes should be repeatable and verifiable

The first two are self-defining, although one could go into detail about checksums, media, level of redundancy, etc.

As to valid forensic procedures, I suggest that this generally means that we can adjust the image quality but not change the image content. There are exceptions to note. In a photo line-up, in which the suspect has a different background, it would be proper to change all backgrounds to match, which is a change to content. And, it also should be noted that over-processing can lead to artifacts that could be misinterpreted as features.

As to repeatability and verifiability - what one does to an image should be able to be repeated with another copy of the original and get the same result; and another examiner should be able to do the same and validate that proper forensic procedures were used. The method for getting this level of repeatability and verifiability can be as simple as always processing images with the same tools in the same manner, or can be done with software that records what was done (such as the History Log in Photoshop) or with notes, or by using adjustment layers, etc. Oh, there are so many ways and they really are pretty easy to incorporate.

I'm not sure that I'd agree that we should be restricted to, or necessarily embrace traditional darkroom adjustments. For example, it has been traditional to make composite images in the darkroom (using one negative to print clouds in a blank sky, for example). And, the processes such as pattern removal with an FFT or using deconvolution to deblur an image are not possible in a traditional darkroom but are valid forensic procedures. Blurring an image could also be done in a traditional darkroom and is a natural result of pretty much any noise reduction method (except for frame averaging), so I'm not sure why that is prohibited by SWGIT.

George

George,
Please excuse my ignorance; I do not have an extensive knowledge of photography (especially digital.) You mentioned that we should never change the “content.” Would you please elaborate on that? If I remove the background from a photograph depicting a latent print on a dollar bill (remove dollar bill design) is that not changing the content? I’m thinking along the lines of a defense attorney asking “You removed the background, so did you not alter the content?”

Wayne:

Your question is quite insightful.

I consider those four parts of best practices to be guidelines, not absolutes. I also listed an example of an exception to changing the content.

The question you pose revolves on whether dropping out a background is another example of a change of content that is acceptable, or if it is an example of a qualitative change (making everything green lighter, for example). I think an argument can be made for either, but I would suggest that it is an enhancement or clarification brought about by a qualitative change, and that the content of the subject (fingerprint) has not been changed. One could argue the same with the photo line-up example in that one is not changing the content of the subject (the person).

Also, the guidelines are intended to be general, in that they are intended for digital imaging in various disciplines including forensic photography, latent print examination, forensic video analysis, etc.

Thanks for bringing this up. It's worth discussing, and it's an example I'll be sure to use in my classes in the future!

George

If someone really wants to pick nits, they could take it to the pixel level that you've changed the value of an individual pixel.

But regarding the use of digital enhancement to remove a specific color range, I don't see that as making a change any more than using a photographic filter to soften or eliminate one color while darkening another color (such as the magenta hue of a ninhydrin latent on a green background w/a green filter and B&W film). Or the use of a combination of the ALS and specific filters to be able to visualize a DFO print on a paper item that can't be seen in the visual range.

There is no reason that we should be limiting ourselves to the visible light spectrum and only what we can see. If that were the case, black-powedering a latent would be just as attackable - you added black powder to something that couldn't be seen, thus making a change.

I think it's important to have an understanding of what it is that you're doing, regardless of the technique, if it is a regular part of your digital processing workflow.

I'll give you an example of a technique and I'm curious to hear opinions on it:

While creating a court charted enlargement, I noticed while overlaying the latent on top of the known, with the opacity reduced, that the ridges don't line up. If I count out 20 ridges on the latent, I've only reached the 15th ridge on the known print. So I line up the cores and make a mark on the latent and the known of a corresponding ridge detail that's at the 20th ridge. Then I use the Free Transform option to pull the latent so that, when overlaid, this specific detail lines up and the ridges are in a more consistent alignment throughout the areas of the known and latent that are represented in both images. Then I move the latent so that it's side-by-side with the known and begin charting the details.

Any problem doing this?

Thanks for all the input. Your suggestions are along the lines I've implemented. Is that your book I've seen on Amazon, George?
Steve, You mention using image transfrom to align latent and print for court presentation.
I would have reservations about using that tool. One of the tenets of fingerprint evidence is that while the latent and print may not be identical in appearence, they were made by the same finger. The differences are explained by elasticity of the skin, pressure and movement.
Using free transform to make the latent "more identical" is putting more emphasis on a "spot the difference" presentation rather than explaining why these differences occur.

Ger.

Steve,

I also would have reservations about using Free Transform to fiddle with the images. People already expect the latent and known prints to overlay each other because of the computer wizardry of CSI. I would feel much more comfortable talking about how skin is flexible and how we would expect that the images NOT to overlay each other.

I'm not much of one to talk of slippery slopes, but I can see a fairly simple use of the Free Transform tool inch to a more complex use of Free Transform. Later, if a bifurcation in the inked print looks like a ridge ending in the latent, I could see someone using Rubber Stamp to turn the ridge ending into a bifurcation to make a more consistent alignment. Now we're really in dangerous waters.

I think Photoshop is a great tool for latent print examiners, but I would personally draw the line at using the Free Transform tool to create a latent to overlay the known.

Ger - Yes, "Photoshop CS3 for Forensics Professionals" is my book. It's not really limited to CS3, as most processes in it can be done with earlier versions.

Steve - You like to get us all thinking, don't you?

Suppose that instead of distorting a latent to fit the exemplar you wanted to use the Free Transform to correct perspective distortion? Or, how about for a court display of two vehicles instead of two fingerprints? Suppose one uses the lock button to constrain proportions, so one is only scaling and not distorting?

One of the difficulties with guidelines is that it's easy to find exceptions to any rule - which is why Steve's comment about understanding what you are doing is key to any image processing technique. Because I provide training, I am frequently hesitant to say that there is no substitute for good training (it is a bit self-serving), but I think it is true (which is why I do teach).

George

Ger208k wrote:My Bureau have recently updated our AFIS and photoshop CS2 was included. Are there any official guidelines as to which tools would be regarded as enhancement tools and which tools would be considered manipulation.

My advise would be to #1: Get Georges book and start learning from that. and #2: Try to go to a digital class somewhere.

George wont give too much praise to his book on here, but I will. You can read that book and see nearly all of the accepted forensic techniques you can use in PS. He also teaches you how to perform those techniques in a manner that is verifiable and repeatable. Its pretty much a must addition to your library if you are going to be using PS for forensic work.

As far as my personal belief on what techniques are ok: You can use techniques that dont change the content of the detail of interest. The content of the detail of interest is different in each image. In one image the detail my be a fingerprint, in the next it may be luminescent blood stains in a crime scene photo. You have to make an image by image determination of what detail is of value to you and process images according to that. During image processing you will change the values of pixels, thats OK. If the ridge flow of your fingerprint, the sequence of unique details, etc are preserved....You are not changing the detail you are interested in.

For example: Color may be very important in a crime scene photo, so performing a channel mixer process to change the color may not be appropriate. However (in most cases) preserving color isn't a big deal in processing fingerprint images, so channel mixer is a very valid technique.

Some processes are valid to use in moderation, but if you take them too far you can change the detail of interest within the image. For example: A levels adjustment is perfectly valid in most cases, however if you take it too far you can change (or lose) detail of interest via clipping. You have to know when you get to the point that you are clipping pixels in your detail of interest and stop before you get there.

Steve,

Interesting question that got my gears turning. I was not really clear on if you were transforming your latent while constraining porportions, or if you weren't. I'm assuming you weren't warping the image. I think that it would be easier to explain the elasticity of skin than to go in depth about free transform and interpolation, but thats just me. I dont really see a reason you CANT do it as long as you aren't using the warp feature.

Of course it would depend on the resolution of the image, how much you enlarged it to create a chart, how much free transform was applied, type of interpolation used, etc. I'm interested to hear your rationale behind doing this.

Steve Everist wrote: I'll give you an example of a technique and I'm curious to hear opinions on it:

While creating a court charted enlargement, I noticed while overlaying the latent on top of the known, with the opacity reduced, that the ridges don't line up. If I count out 20 ridges on the latent, I've only reached the 15th ridge on the known print. So I line up the cores and make a mark on the latent and the known of a corresponding ridge detail that's at the 20th ridge. Then I use the Free Transform option to pull the latent so that, when overlaid, this specific detail lines up and the ridges are in a more consistent alignment throughout the areas of the known and latent that are represented in both images. Then I move the latent so that it's side-by-side with the known and begin charting the details.

Any problem doing this?

Ok, this has had a few days for consideration. Actually this isn't something that I'm doing. I thought it may be better to propose it in this manner in hopes that there would be replies from people who were concerned about what I was doing and had no problems pointing it out. And that's exactly what I got. Primarily I was looking for discussion regarding both the change in content of the image vs. changing quality and reference to the differences that could occur as a result of the elasticity of the skin, deposition pressures, rotations, etc... Ger. nailed it with the first response. ER, George and Andrew took it even further than I had thought about.

The reason for mentioning free transform as opposed to just the scale adjustment from the transform menu was that I wanted to indicate that I was going to be pushing and pulling wherever necessary to get them to look the most same with no regard to proportions. I don't really see a huge issue with constraining proportions, although if you were to make it larger it could introduce interpolation. But these images wouldn't be for the purpose of doing the comparison itself. But then there would probably be other size-related issues and it would be a lot less effort to have them not be in exact proportions and just explain the previously mentioned factors.

The reason I brought it up was that it is something that was shown in a class that I took on digital imaging a few years ago. Also I think it's important that everyone who uses digital imaging as part of their workflow understand, to the best of their ability, what it is that's happening to an image under their control. There's nothing wrong with being critical of a procedure that's taught by someone even when they have a greater level of knowledge, skill, and experience.

So thanks for all your feedback - it's exactly what I was hoping for.

Steve,

My first reaction was, he can't do that! Then I started thinking about it and started thinking about all the variables. Looked up a little on free transform and how it interpolates pixels. And decided that:

Doing something like that really depends on a lot of factors. Probably the two biggest ones would be what your resolution is in your image and how much transform you are applying. I'd definately throw out performing any warping type technique, but I'm one that thinks that under the right circumstances interpolation isn't all that evil. Heck those voodoo video guys (George) use interpolation all the time .

Suppose that you have an image that is 2000ppi and you are using free transform to stretch it in one direction but not the other (ie making it a little taller but not wider). If you stretch it by a factor of 10%, you are performing a rather small amount of interpolation. Probably on the order of one pixel interpolated for every 10 in a column. In a 2000ppi image is there really any chance that this will alter the position and sequence of your fingerprint detail? I'm thinking not.

Again, I dont think that I'd do this because I think that its much easier to explain to a jury the elasticity of skin. But then again I dont think that I could fault someone (with the knowledge to back up their processes) for doing so. Like you said, you wouldn't be performing comparisons off the chart. I think that I know what class you went to, I had heard that a particular person was teaching that method of charting.

George, what are your views on interpolation? I know that you know a whole lot more about it than I.

Hello,

I'm new to this forum, and thanks to a forensic collegue of mine, Gerrit Volckeryck, I was able to read this thread.

Having bought Georges' book, and followed a training course at UNIL Lausanne (digital forensic photography), I was taught to use CS3 (always in combination with Camera Raw format) in the "Smart Object" mode:

every change to the image is made on metadata level only (read: non destructive editing). By doing so, not all editing functions in CS3 are available, but we have to keep in mind that the changes we make to forensic images should always stay enhancements and NOT manipulations to the image. The four points George Reis lined up concerning "best practise" together make up a good basic platform.

I always tell my new collegues: be aware of the thin line between manipulation and enhancement. Even USM sharpening for me is one step too far (theoritically of course): your image is focused or it isn't.

@ Andrew: using a 2000ppi image and stretching it in one direction (so without constrain proportions) will indeed leave you behind with at least some kind of interpolation. Most probably this won't give any problems for the visual aspect for the AFIS system, but nevertheless you add pixels which were not part of the original image (or cutting pixels out of the original image). I assume, but this is my personal opinion, if you keep a history log (I like to call it "digital chain of custody ) there's no problem what so ever. Shoot me if you do not agree...

Kindest regards,

Peter

This thread has gone into a few subjects making it interesting, but also making it more difficult to reply. I'd love to reply to about twenty different comments, but it would simply be too difficult to make sense of (or, at least it would for me).

This thread started with the idea of guidelines or best practices or rules for using image processing in a forensics environment. There are many different ideas as to what are valid procedures in forensics. Some concepts in developing SOPs are to list specific things that can be done and some that can't be done. I never liked that approach because none of us knows enough to know what will be available in the future, or whether some processes are great for things we simply aren't aware of, etc. So, I've never been a fan of rules that list specific processes, hardware, software, etc. On the other hand, without some sort of SOP, guideline, or best practice untrained individuals may do some rather scary processes and form opinions based on bad practice (I know of some examples). This is a topic worthy of much discussion - I tend to think that overly specific rules don't work well but following a general best practices concept as I mentioned above makes sense to maintain the integrity of our evidence, and to answer any legitimate challenges in court.

Another topic of this thread is interpolation. I think that this term isn't well understood. It really just involves the recalculation of pixel values. This happens if you change the size of an image by one pixel or by many pixels (not by cropping, but by retaining the same image content, but changing the number of pixels that make up that content). If you change the number of pixels in an image from 1000 X 1000 to 999 X 999, it is possible that every single pixel will change in value. Interpolation happens a lot in digital imaging. For instance color interpolation happens within the camera when we take a digital photo (unless we use a raw file format, then it happens when we open the image). When we scan a fingerprint at 1000 PPI with a scanner that has a native resolution of 1200 PPI interpolation is happening. When we send an image to our printer, chances are that interpolation is happening. And, this isn't necessarily bad, but it generally means that in any of these processes we're losing some sharpness. I think that some of us think that interpolation is somehow bad - I think it's just part of the nature of digital imaging, just as grain is part of the nature of film.

Steve brought up the concept of warping a latent print to fit the exemplar image. I avoided the question originally, but I think this is a really bad idea (and I think Steve thinks so too). But, there may be some instances where warping an image is a legitimate method. One example is in correcting pixel aspect ratio in video files. Another may be in correcting perspective distortion in a footwear, blood spatter, or fingerprint image.

There is a lot going on in a digital file from the time a shutter is released or a scan in initiated until that image is displayed. There is also a lot going on in a film based image from the time a shutter is released until a print is viewed. Whatever process we're using, having an understanding of the process is essential. The more knowledge we have, the better our evidence.

Sorry that I'm not posting any answers here, but I think the best thing for all of us is discussing and sharing of information - not rules that you can do this, but you can't do that.

George

This is one seriously thought-provoking thread.

I think we should forget about what we could and could not have done in a conventional darkroom and focus rather on how well we can document/explain/reproduce whatever it is we do to the image.

Surely it is for us to convince the court that we know what we are doing. It has been said in here somewhere that even if one pixel value changes we have changed the content so it is my opinion that "preserving the original image content" can never be the standard to work by. It is simply not possible to do unless we use the untouched scene capture in court straight off the camera.

It is further my opinion that the photoshop history tool is limited in how many transactions it will store and so proper documentation is invaluable. Tools in different sequences also give different results and so, again, proper documentation is essential.

As for the free transform tool I would say that it could be handy during comparison but IMO would not amount to more than a nice party trick in the courtroom.

Just my 2c worth.