Is there any way to identify the fingerprint contents in an image using matlab?

If the image is mixed with fingerprints and certain characters, is there any way to pick up the fingerprints and leave the characters

Answer 1

[Cedric]

Here is a lightweight approach that is absolutely not solid/stable; .. could be a starting point?

 I = imread( 'fingerprint001.jpg' ) ;
 bw = ~im2bw( I ) ;

 % - Extract large blobs as mask.
 mask = bwareafilt( bw, [400 Inf] ) ;
 bw2 = bw & ~mask ;

 % - Use bw2 as mask for I.
 I2 = I ;
 I2(~repmat(bw2,1,1,3)) = 0 ;

 % - Display
 figure() ;
 set( gcf, 'Units', 'normalized', 'Position', [0.1, 0.1, 0.8, 0.8] ) ;

 subplot( 2, 2, 1 ) ;
 imshow( I ) ;
 title( 'Original' ) ;

 subplot( 2, 2, 2 ) ;
 imshow( bw ) ;
 title( 'BW' ) ;

 subplot( 2, 2, 3 ) ;
 imshow( mask ) ;
 title( 'Mask' ) ;

 subplot( 2, 2, 4 ) ;
 imshow( bw2 ) ;
 title( 'Mask''ed' ) ;

 figure() ;
 set( gcf, 'Units', 'normalized', 'Position', [0.1, 0.1, 0.8, 0.8] ) ;
 imshow( I2 ) ;

==== FORMER =============================================================

I cropped bands, but it is counter productive.

 % - Crop bands.
 id  = find( bw(:,1), 1 ) ;
 bw  = bw(id:end,:) ;
 idl = find( ~bw(fix(end/2),:), 1 ) ;
 idr = find( ~bw(fix(end/2),:), 1, 'last' ) ;
 bw  = bw(:, idl:idr) ;

[Walter Roberson]

This solution is an example of changing the question: the original question wanted the characters taken out, but this solution takes out the round arc and the widget as well as the characters.

If I recall bwareafilt() did not exist in R2013a. It is not difficult to write, however, by using regionprops to ask for Area and using logical indexing to select the range of areas to preserve.

[Cedric]

No actually, this is an example of misunderstanding the question! I should have read it twice .. I was trying to get the fingerprints :-/

Well, I am reading it again and finally I am not sure what "pick up the fingerprints and leave the characters" means.

[D Joseph]

"pick up the fingerprints and leave the characters" means one image showing only the fingerprint contents and other image showing the rest like characters and so on.

[D Joseph]

Thanks@Cedric. But could you please try some heavy method

[Image Analyst]

Why do you want both? What is the real world use case here? If it's forensics, I can see wanting to get the fingerprint, but why would you want or care about anything else like characters or other clutter? What is your "use case"?

[D Joseph]

its definitely forensics. Characters are needed just to display as cartoon components. Fingerprint is used for further processes. Can we able to find ridge contents in such obtained fingerprints?

[Walter Roberson]

I have been trying to push the point in a few different Questions lately: that if you want to detect "characters" then you need optical character recognition, and in particular you need to define ahead of time what the valid "characters" are (or are not) because if you do not then anything is potentially a "character" in some language you didn't happen to think about (like Horta or Reaver)

What Cedric has done here in detecting blobs that are much larger and removing those is very reasonable for the situation where the problem is not defined in terms of "characters" but in rather in terms of "removing parts that do not belong to fingerprints of this type". The "things that do not belong" can be defined by "relatively large dark object" for this image. And that being the case you do not need to recognize whether it is a character because you don't care that it might be a Greek Lowercase Xi, you just want the squiggle gone.

[D Joseph]

I just want to extract the texture component ie., the fingerprint from that image. Is there any algorithm for the same

[D Joseph]

excellent work by Cedric Wannaz. But the function bwareafilt is not builtin in MATLAB2013a. Is there any code available for the same

[Walter Roberson]

lowerlimit = 400;  %for example
upperlimit = inf;  %for example

stats = regionprops(YourImage, 'Area', 'PixelIdxList');
areas = [stats.Area];
toosmall = areas < lowerlimit;
toobig = areas > upperlimit;
unwanted = toosmall | toobig;
unwantedpixels = vertcat(stats(unwanted).PixelIdxList);
NewImage = YourImage;
NewImage(unwantedpixels) = 0;

[Cedric]

Keep in mind that these are just tricks. I am not particularly knowledgeable in image analysis, and it is likely that you will have to learn quite a bit (e.g. on VisionBib) if you want to implement a solid approach. The next things that I would do if I had to develop quickly a lightweight approach are

• Convert the original image to HSV, and try to define an HSV volume which picks the tones of the fingerprints.
• Mask small blobs (like a few pixels large) as what defines a fingerprint is a set of elongated blobs.

[D Joseph]

Can you please provide the code for these processes

[Cedric]

Use RGB2HSV for converting the image to HSV, then read about the range of H, S, and V (e.g. on Wikipedia), and create a mask based on their values. No guarantee that it will lead anywhere though. For the second, use my example and what Walter wrote, where you define e.g. lowerlimit=10 and upperlimit=Inf. You will have to tweak the lower limit though, and it is likely to be image-specific, so I would say that you will invariably be back to a situation where you have to spend a lot of time learning/testing if you want to implement something robust.

[D Joseph]

Thanks@Cedric

[Cedric]

PS: I am generally not answering by email; please post specific questions on the forums. Also realize that it is a very complicated task that you are trying to achieve here, and it is unlikely that you will "get code" from people for doing the whole process. It would probably take weeks or months of work to people proficient with MATLAB for implementing a solution that may not even be that robust. This would imply reading the available literature, reading relevant toolboxes documentation, experimenting with examples, etc.

About your question "how to enhance an image", you should post a new question that includes the image and describes specifically the type of enhancement that you are looking for (knowing that "making the fingerprint appear" is not "one enhancement" but the outcome of a lengthy process that probably no one here can implement for you). This means that you will have to go through your toolbox documentation, experiment with the examples, see how they could apply to your case, and then come with much more specific questions like "how to eliminate blobs larger than a given amount of pixels" (which is what I did above, and what Walter nicely implemented for MATLAB 2013a).

[D Joseph]

Thank you sir

[Cedric]

There are a few things that you could precise

• How many images will you have to process?
• Are they coming from a database of scans, all with the same quality?
• Are all finger prints circled by hand the same way?
• Are all fingerprints at the center of the image?
• What is the purpose?

and you could attach 2 or 3 more pictures. Maybe if we know that there is some regularity that could be exploited, it will trigger an idea that could simplify the process.

[D Joseph]

I have to process 30 images. They are having same quality and from same database. The purpose is to extract the fingerprint content and to enhance it.

All images are in the format .jp2(compressed with Jpeg2000). The format is not supported here. So I'm attaching the screen shots of an image.

[Walter Roberson]

You can zip the .jp2 and attach the .zip

[D Joseph]

The .zip format is attached.

[Cedric]

Ok, then you can reduce the complexity of the question by stating that:

• The fingerprint does not need to be detected algorithmically. By this I mean that if there are only 30 images, you can pre-process them all by hand and keep only the area with the fingerprint.
• The algorithm for enhancing remaining areas does not need to define how successful it is if the final analysis is performed visually. You can check out the result visually and determine if it is good or not, and refine parameters.

If these two statements are correct, the question becomes simpler, because there is no need for algorithms that detect region(s) of interest, that evaluate what operations to perform in these regions, to evaluate how successful these operations were.

MATLAB may not be the best solution though; have you tried asking on a Photoshop or a Gimp forum? I am mentioning these two because you may realize that each image needs a specific treatment, and there may be no way to find one type of processing that enhance all of them. In such case, you will have to process them individually anyway.

Finally what is the purpose? Enhancing means enhancing for a given purpose. To illustrate, if you need to detect specific features with an algorithm, enhancing may mean increasing the contrast. Doing this, you loose details, but you make macroscopic features stand out. Now if the purpose is to analyze the outcome visually, then you may prefer keeping the details.

[D Joseph]

Purpose is to extract minutiae

Answer 2

[Walter Roberson]

Use an OCR (Optical Character Recognition) routine to locate the characters. Then you can remove them, or remove everything except them, depending upon your needs.

The OCR step is needed because there is a lot of variety in what characters can look like, and the way you have defined the process it is necessary to only locate some kinds of characters but not other kinds. (The general problem of locating all kinds of characters is not solvable, as any given mark could be a character in some language.

[D Joseph]

ocr is not accepted by the MATLAB 2013 version, as builtin function. The code for the same is too long. What to do

[Image Analyst]

Why don't you image it with a multispectral system?

[Walter Roberson]

You could upgrade. Or you could give up. Or you could change the question.

[D Joseph]

How can we image it with a multispectral system?

[Walter Roberson]

You would need to illuminate it with light of different frequencies and record it with devices sensitive to those frequencies. For example infrared

[D Joseph]

What is this infrared?

[Walter Roberson]

https://en.wikipedia.org/wiki/Infrared

Infrared (IR) is invisible radiant energy, electromagnetic radiation with longer wavelengths than those of visible light, extending from the nominal red edge of the visible spectrum at 700 nanometers (frequency 430 THz) to 1 mm (300 GHz) (although people can see infrared up to at least 1050 nm in experiments). Most of the thermal radiation emitted by objects near room temperature is infrared.

Answer 3

[Image Analyst]

I don't see why "cartoons" are needed in forensics. Anyway, there is no built-in fingerprint code. You're going to need some sophisticated algorithm way beyond what anyone here knows, especially for such bad fingerprint images as you showed here. Please go to VisionBib and find an algorithm there then code it up. Of course you must realize it won't be simple or short if it's smart enough to handle such poor quality images. And even the best algorithm out there might not work so great with your highly degraded images.

[D Joseph]

Thanks@Analyst. But I couldn't find any algorithm in VisionBib

[D Joseph]

Cartoons are not needed.Its just to display as noises. I want the fingerprint texture components.