CLPEX Chat Board

If you've ever been on the stand and mentioned that you use photoshop to prep latent prints for Analysis, AFIS or Charting, you've inevitably gotten questions (or accusations ) about how photoshop alters the image. I usually just mention that in essence it's just being used to enhance contrast not change content. Well with the new version of Photoshop employing Firefly AI, I decided to indulge my curiosity and see what I could do to change the content.

On the first image, I left the prompt blank which let's the AI interpret the image and add more 'image like' properties to the area you selected. The second image I gave it a prompt just for creative purposes to see how it would integrate into a fingerprint. (no, I'm not posting case images, these are my prints). The third image, I altered the second prompt with some lessons I learned along the way.

The first image (whorl to loop), I had to try this several times and limit the size of the lassoed area. Most of the time, it came out like a print with dysplasia. I did try prompts like 'arch pattern' and 'loop pattern' but with no real luck.

The second image I tried to recreate some of the features naturally found in fingerprints like faces, frogs and other glyphs, I used the prompt 'line art portrait' and an interesting result that integrated into the print.

The last image, I tried again but changed the print to 'spatter face' thinking the AI would interpret the lines less like portrait and more dysplasia-like as I had found in testing the first image with no prompt. I got lucky pretty early with the result I posted.

What would you try?

Great images Boyd.

Googles Bard now lets you upload images and then ask questions about the images. I uploaded an image of loop with a ruler. This is the response I got:

The image you sent me is a close-up of a fingerprint next to a ruler. The fingerprint is located in the upper left corner of the image, and the ruler is located on the right side of the image. The ruler is in centimeters, and the fingerprint is approximately 2 centimeters wide.

It got the position of the latent print and the ruler wrong. This sizing was correct.

I then asked with type of pattern this was, and this was it's answer:

The pattern of the fingerprint in the image is a whorl. Whorls are the most complex type of fingerprint pattern, and they are characterized by ridges that spiral or circle around a central point. Whorls make up about 35% of all fingerprint patterns

Which was wrong. So I asked it to regenerate its answer and this is what I got:

I apologize for my previous response. I have reviewed the image again, and I agree with you that the fingerprint pattern is a loop.

Loop patterns are the most common type of fingerprint pattern, accounting for about 65% of all patterns. Loops are characterized by ridges that enter the print on one side, curve around, and then exit on the same side. Loops are classified as either radial loops or ulnar loops, depending on the direction in which the ridges curve.

I have noticed in most of the model a lot of hallucinations when researching topics about the science of fingerprints, latent print examination, historical information about the science, etc. Not enough training data.