Mobile Device Remote Identity Proofing Part 4 – Best of the Biometrics

Download PDF of complete paper

VII. Fingerprints

 

There are two national fingerprint specifications; the FBI's Integrated Automated Fingerprint Identification System (IAFIS) Image Quality Specifications (IQS) Appendix F and NIST’s PIV-07 1006.  Appendix F has stringent image quality conditions, focusing on the human fingerprint comparison and facilitating large scale machine many-to-many matching operation.  (FBI Biometric COE, 2010)  Our focus however will be based on the PIV-071006 standard, a lower-level standard designed to support one-to-one fingerprint verification.  The class resolution requirements for fingerprint capture and use for Personal Identity Verification (PIV) at Fingerprint Application Profile (FAP) level ten or above are 500 PPI with a maximum tolerance variation of ± 2%.  Class resolution refers to the resolution required for acquisition or imaging related use. (Wing, 2011)

Most of the complexity related to resolution pertains to the friction ridges of the fingerprint.  A friction ridge is a raised section of the epidermis of the skin. A fingerprint is a trace image of the ridges in a human hand or foot to include the fingers and toes.  Traditionally fingerprints were captured by rolling the pad above the last joint of the finger and thumbs on an ink pad and then rolling the inked pad onto a piece of smooth card stock.  Impressions of fingerprints are left behind on various surfaces when the natural secretions of the body, or cosmetic oils and body lotions, gathered on the ridges are left behind when in deliberate or accidental contact with any smooth surface. These are referred to as latent prints.  While not always immediately visible these impressions could be lifted by dusting the print with specialized powders or exposing the print to chemicals like silver nitrate or cyanoacrylate ester  and capturing the image by pressing it to a specialized paper or plastic media. The latent print could then be compared to the inked print with a reasonable chance for a match determined by an experienced examiner in dermatoglyphics.  Although effective this method precludes its use in identity management based on the sheer volume of prints and comparisons required.  In other words it is not practically scalable.  What is needed is a means of digitally capturing the fingerprint and storing the resulting record.  Live Scan is the most widely used method of accomplishing this.  A live scan involves pressing or rolling a finger onto a specially coated piece of glass or platen and then imaging the fingerprint using optical, ultrasonic, capacitive or thermal imaging to capture the ridges of the finger and the valleys between them.  Optical imaging is in essence a specialized form of digital photography. The major difference between a digital camera and an optical imager for capturing fingerprints is the presence of a light-emitting phosphor layer which illuminates the surface of the finger increasing the quality of the resulting image.

There are challenging problems when developing fingerprint recognition systems that use a mobile camera. First, the contrast between the ridges and the valleys in images obtained with a mobile camera is low.  Second, because the depth of field of the camera is small, some parts of the fingerprint regions are in focus but some parts are out of focus. Third, the backgrounds, or non-finger regions, in mobile camera images are very erratic depending on how the image captures place and time. (Lee, Lee, & Kim, 2008)  So is there an insurmountable challenge with using a smart phone camera to capture a fingerprint?  Image quality is determined by light quality, lens quality and type, and shutter speed.  Smart phones do not fully address each of these important elements trading size and ease of use for function.  Because of this you will get a better picture from a low end Digital Single Lens Reflex (DSLR) camera than you will from a high end smart phone camera.  Shutter speed is not an applicable issue with fingerprint capture but light and lens quality and type are. 

An additional challenge is the probability that one can spoof or fool an optical camera with an image or impression of a fingerprint.  This is resolved within the industry by using various live finger detector technologies.  One means of live finger detection is accomplished “by measuring the unique electrical properties of a living finger that not only characterize the finger print but measure what is underneath it. This technology has the capability to process the acquired data, that is, characterize and classify the results in a way that enables the system to verify a living finger with a very high degree of confidence.” (Clausen & Christie, 2005)  It is unlikely that this type of fraud prevention technology can be integrated into widely available smart phones in the near future so the risk of fraudulent fingerprints in a mobile identity management program will have to be addressed through policy or another more easily implementable technology enhancement.  Despite the obvious challenges, capture of a useable fingerprint image with a cell phone camera is not impossible.  The operator must take into account the fixed focal length of the camera lens and make sure the auto focus is disabled in order to get close enough to capture an image with prominent ridges. Lighting also remains a challenge.  An informal test while this paper was being written used an I-Phone® 4 both with a flash and without.  A distance from the camera of four inches with no flash in a brightly lit room resulted in the best image with clearly defined ridges in the left index finger of the test subject.  By importing the image into paint.net and using the color inversion tool an image just as clear to the naked eye as one caught on a live scan was produced.  This test was by no means scientific but serves as an indicator that it is not a far stretch to utilize off the shelf cell phone technology.   The methodology of the image capture is not necessarily a limiting factor even taking into account challenges with optics and lighting.  The recognition algorithms used in the associated databases can counter or resolve some of the issues.  Many fingerprint recognition algorithms perform well on databases that had been collected with high-resolution cameras outperforming feature only searches by trained examiners. (Indovina, Hicklin, & Kiebuzinski, 2011) 

VIII. Face

Facial recognition is considered to be the most immediate and transparent biometric modality when it comes to physical authentication applications.  Why is it that many people are inclined to give up their facial image without question while the concept of giving up a fingerprint causes them great discomfort and angst.  Facial recognition is a modality that humans have always depended on to authenticate other humans.  We are in essence hardwired for facial recognition.  Therefore the addition of facial recognition through or enhanced by technology is an easy one to accept.  “Whether or not faces constitute a [special] class of visual stimuli has been the subject of much debate for many years. Since the first demonstrations of the Binversion effect…it has been suspected that unique cognitive and neural mechanisms may exist for face processing in the human visual system.” (Sinha, Balas, Ostrovsky, & Russell, 2006)

Facial recognition as a technology is one of the most mature of the biometric modalities.  It is also relatively simple from the image capture standpoint.   Capture of a facial image requires little or no cooperation from the subject making it the technique of choice for passive applications like those used in airports and casinos.  On the surface it seems as though all of the issues are algorithm related but as our concept is focused on a cell phone camera as our capture device this is not really the case. 

We previously discussed the megapixel issue but megapixel capability has no discernible impact on the biggest challenges with facial recognition which are image capture and pose correction.  Image capture is a light and optics issue.  One of the biggest drawbacks to smart phone cameras is the size of the sensor.  Camera technology has changed but the basic principles have applied since the first tin types were produced in the mid 19^th century.  The sensor is the replacement to the emulsion based films.  The larger the sensor the more light it can detect resulting in better picture quality.  Smart phone cameras have a much smaller sensor than the traditional 35mm film size and as a result have a smaller angle of view when used with a lens of the same focal length.  This results in an image that is essentially cropped.  In order to adjust for this the camera must be further back from the subject posing problems related to lighting and detail.  

Facial recognition software analyzes a number of structural facial elements.  Examples of these distinctive surface features include shape of the eyes and the eye sockets; the width, length, and structure of the nose; the thickness of the lips, and the width of the mouth.  What is common about all of these elements is that they are three dimensional.  A camera captures images in two dimensions. The difference between a three dimensional subject and the two dimensional output of the cameras is handled by the software but pose issues including expressions, external features, background, and lighting all add variables that decrease the effectiveness of the algorithms. In the home environment it may be difficult to deal with lighting and background issues but this is not an insurmountable challenge.  In the same manner external features such as beards, glasses, jewelry, and piercings can all pose problems.  The author of this paper has endured lengthy picture sittings in front of DSLR cameras for PIV credentials.   It seems his white goatee gives the capture software conniptions.  This serves to demonstrate that issues of facial capture are not necessarily specific to smart phone cameras. 

Many of the issues in facial image capture would be solved if the images could be captured in 3D.  Of course this would eliminate the use of smart phones as a capture device, or would it?   Fujitsu continues to refine a way for phones that just have one rear camera to shoot three-dimensional videos with the aid of a special attachment.  The attachment uses mirrors to send two different images to the camera’s sensor and is smaller than a stick of Chap Stick.  In June of 2011 Sprint released the HTC Evo 3D 4G 'Gingerbread' Smartphone.  This phone had two integrated cameras capable of taking 3D pictures.  With the potential of standard 3d capture technology on the horizon it may not be long at all before changes in lighting and camera angles become irrelevant.  Three dimensional image captures can only serve to enhance the potential of fingerprint capture as well.  Even the issue of software sensitivity to expressions, one not mitigated by 3D technology, could soon be eliminated.  As far back as 2004 Technion, the Israel Institute of Technology, a public research university in Haifa researched using metric geometry to address the issue of expression sensitivity.  The approach was to use metric geometry isometrics to create an expression invariant three dimensional face recognition solution. (Bronstein, Bronstein, & Kimmel, 2004)

IX. Why not?

 

There are other biometric signatures that have both been the focus of research and have seen increased use and acceptance from the physical and logical access communities.  Iris scans, hand geometry, and voice recognition are no longer the purview of James Bond and Ethan Hunt.  Although not practical for this smart phone centric premise they are worth mentioning and potential near future candidates.

Iris scans are based on the stability of the trabecular meshwork, an area of tissue in the eye located around the base of the cornea.  The patterns are formed by the elastic connective tissues which gives the iris the appearance of radial divisions which are unique and often referred to as optical fingerprints. Iris sampling offers more reference coordinates than any other biometric resulting in an accuracy potential higher than any other biometric.  Iris scans require a high degree of cooperation from the subject from whom the sample is being acquired.  Today specialized capture devices are required.  Despite their complexity these capture devices are nothing more than still cameras capturing very high quality images.  It is certainly not out of the realm of possibility that a smart phone camera could one day soon be capable of the required performance.

Hand biometrics is a fairly mature technology that lends itself to applications where the size of the capture device is not a factor.  Current devices are based on charge-coupled device (CCD) optical scanning and consistently deliver better quality images than fingerprint scanners.  This is largely due to the increased sample size, your hand being many times larger than a finger pad.  Three-dimensional photography may show some promise as an alternative method of hand biometric image capture in the future.  Current technology remains expensive and not at all compatible with the proposed smart phone format.

Voice recognition is perhaps the oldest form of biometric identifier. Not to be confused with speech recognition, which is the process of translating speech into text, voice recognition is the process of identifying someone from their voice patterns.  It is a phenotype, an observable behavior influenced by development, often with regional characteristics.  Of all of the fields of biometric research, speech development has seen the most modern day focus with significant research over the last four decades.  Voice recognition has some uniquely distinct advantages over other biometric signatures in that it can be combined with pass phrases, knowledge based verification, or can be used as a passive background tool.   Voice recognition is the least invasive and is easy on the user.  With all this it would seem like speech recognition should be the biometric of choice but has its disadvantages.  Voice recognition programs take the digital recording and parse it into small recognizable pieces called phonemes.  These phonemes may not be consistently reproduced as they can be influenced by behavior and health factors and even background noise. 

Works Cited

 

Bronstein, A. M., Bronstein, M. M., & Kimmel, R. (2004). Three-Dimensional Face Recognition. Technion, Israel Institute of Technology, Department of Computer Science. Kluwer Academic Publishers.

Clausen, S., & Christie, N. W. (2005). Live Finger Detection. IDEX ASA. Fornebu, Norway: IDEX ASA.

FBI Biometric COE. (2010, April 27). FBI Biometric Specifications FAQ. Retrieved May 31, 2012, from FBI Biometric Center of Excellence: https://www.fbibiospecs.org/iafis_FAQ.html

Indovina, M., Hicklin, R. A., & Kiebuzinski, G. I. (2011). Evaluation of Latent Fingerprint Technologies: Extended Feature Sets [Evaluation #1]. U.S. Department of Commerce, National Institute of Science and Tecnhology. Washington D.C.: US Government Printing Office.

Lee, S., Lee, C., & Kim, J. (2008). Image Preprocessing of Fingerprint Images. Biometrics Engineering Research Center at Yonsei University., Korea Science and Engineering Foundation, Seoul, Korea.

Sinha, P., Balas, B., Ostrovsky, Y., & Russell, R. (2006). Face Recognition by Humans: Nineteen Results All ComputerVision Researchers Should Know About. Proceedings of the IEEE , 94 (11), 1957.

Wing, B. (2011). Data Format for the Interchange of Fingerprint, Facial & Other Biometric Information. US Department of Commerce, National Institute of Science and Technology. Gaithersburg: US Government Printing Office.