Gender Estimation in Face Recognition Technology: How Smart Algorithms Learn to Discriminate

Please download to get full document.

View again

of 10
27 views
PDF
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Document Description
Gender Estimation in Face Recognition Technology: How Smart Algorithms Learn to Discriminate
Document Share
Document Tags
Document Transcript
  Gender Estimation in Face Recognition Technology : How Smart Algorithms Learn to Discriminate Sarah KemberFace recognition technology is becoming central to a naturalized—embedded and invisible—ontology of everyday control. As a marketing and surveillance-based biometric and photographic technology, one of its main advantages over other biometrics such as finger printing or iris-scanning is that it operates at a distance and does not reuire consent or participation. Face recognition is a default setting on social net!orking sites like Facebook, offering automatic tagging suggestions as the user uploads photographs of friends and family. "t is becoming ubiuitous in international airports and other social environments !here security and#or commerce are at stake. $he ob%ective of face recognition is to be able to pick out a face from a cro!d and identify the target by comparing it !ith a database. &here this ob%ective is hard to achieve, another goal entails learning to discriminate bet!een classes of faces based on gender, race, and age. $his is easier in that it relates to groups rather than individuals and appeals to biological differences. Media Fields Journal   no.7 (2013)  2 Gendered Estimation in Face Recognition Technology Face recognition systems seek to overcome the division bet!een human and machine vision or, specifically, bet!een human and machinic capacities for appearance-based face recognition and identification. 'uestions of system accuracy and performance come to the fore because the comparison remains unfavorable. Subseuent performance an(iety serves to legitimize a range of technological innovations designed to close the gap, and among   them is the use of A") *A" approaches utilize tools suchas neural net!orks and machine learning techniues to recognize faces.+ o! do machines learn $he issue has been !idely debated but in this conte(t it is clear that in addition to techniues of pattern recognition and sorting, the principle mechanism of machine learning is reductionism. atthe! $urk and Ale( /entland have made a significant contribution to the development of face recognition. For them, *developing a computational model of face recognition is uite difficult, because faces are comple(, multidimensional, and meaningful visual stimuli.+ 0  Face recognition systems substitute the meaning of faces for a mathematics of faces, reducing their comple(ity and multidimensionality to measurable, predictable criteria. oreover, face recognition technology reuires a reduction in the variation of face images and environments and must ultimately replace faces !ith vectors 1principal components of faces2 or !ith standardized templates in order to learn anything at all. System accuracy and performance depend on *constrained environments such as an office or a household.+ 3  $he face image presented to the system for recognition must be centered, *the same size as the training images,+ and fully frontal or in profile, so reproducing—as input—the mug shot photograph generated by nineteenth-century !ays of seeing. 4  An elision of labor secures the illusion of autonomy in face recognition technology. $here is also an inventory of technological failures that, combined !ith reductionism, delimit the claim to smartness implicit in the system5s ability to learn.  6  o!ever, !e cannot simply dismiss the claim, as it is manifested in the very architecture of the system. ere, smartness materializes in pattern-recognizing and sorting algorithms that are learning to identify faces by discriminating among them, generating ontological and epistemological divisions—bet!een male and female, black and !hite, old and young—that in this case must remain un-reconciled, reduced to a set of essentialized categories that guarantee system performance by ensuring that input 1a recognizable face2 is euivalent to output 1a recognized face2. $he aim of a facial recognition system is either to verify or to identify someone from a still or video image. Follo!ing the acuisition of this   3 Media Fields Journal  *probe+ image, the system must first of all detect the face or distinguish bet!een the face and its surroundings. $o do this it selects certain landmark features in order to compare them !ith the database. 7ither that or it generates !hat are called standard feature templates—averages or types. 8nce detected, the face is normalized, or rather the image is standardized !ith respect to established photographic codes such as lighting, format, pose, and resolution. Again, this aids comparison !ith thedatabase. o!ever, the normalization algorithm is only capable of compensating for slight variations, so the probe image must already be as close as possible to a standardized portrait. "n order to facilitate face recognition, the already standardized image is translated and transformedinto a simplified mathematical representation called a biometric template.$he trick, in this process of reductive computation, is to retain enough information to distinguish one template from another and thereby reduce the risk of creating *biometric doubles.+ 9 8ne of the algorithms used in face recognition is /rincipal :omponent Analysis 1/:A2. "t produces images akin to Francis ;alton5s nineteenth-century eugenicist photographic composites by removing e(traneous information, including the outline of the face itself . <  /:A reduces faces to their vectors and refigures them as eigenfaces. "n *7igenfaces for =ecognition,+ $urk and /entland e(plain that the system functions *by pro%ecting face images onto a feature space that spans the significant variations among kno!n face images.+ >  Significant features are referred to as eigenfaces *because they are the eigenvectors 1principal components2 of the set of faces.+ $hey may correspond to familiar features like eyes and noses !hose geometric relation is then measured and computed. 7ach input, or individual face image, is *a !eighted sum of the eigenface features, and so to recognize a particular face it is necessary only to compare these !eights to those of kno!n individuals.+ $urk and /entland ackno!ledge that an eigenface is *an e(tremely compact representation+ not only of the face but of the srcinal face image. ?  "t is a practical rather than elegant solution to the problem of face recognition. @  inear Biscriminant Analysis 1BA2 is another key algorithm. "t creates classes of faces, much like avelock 7llis did in his nineteenth-century physiognomy of criminals. "n their survey of face recognition techniues, Cafri and Arabia e(plain that BA *ma(imises the ratio of the bet!een-class scatter+ and is better at classifying and discriminating bet!een classes of faces than /:A. 0D  $his may be partly because this approach starts by selecting faces that are already distinctive. As BA researchers Kamran 7temad and =ama :hellappa state, *First, !e need a training set  4 Gendered Estimation in Face Recognition Technology composed of a relatively large group of sub%ects !ith diverse facial characteristics. $he appropriate selection of the training set directly determines the validity of the final result.+ 00  Sorting algorithms discriminate bet!een classes and types of faces. Eoth BA and, increasingly, /:A are being used to discriminate on the basis of gender. :ontemporary face recognition systems differ from earlier analogue and digital systems in that they are e(clusively oriented to!ard recognition rather than recall. $hey are designed according to the surveillance and marketing imperatives of targeting, tracking, and location. o!ever, picking out one face in a cro!d is harder and more prone to error than identifying once class of faces as distinct from another, especially !hen that class appeals to the biological categories that inform gender, race, andage. $hese categories are naturalized through geometric coding techniues 1!here syntactic coding is reserved for face recall2 and the default sub%ect of these techniues is still the young !hite male.Cacue /enry5s /hotoF"$ pack came in to use in the 0@>Ds and consisted of photographic images of five features 1hair and forehead, eyes, nose, mouth, and chin2 mounted on card. 03  e included a male and female database but established !hat he claimed !as a universal—genderless—facial topography. $his !as actually derived from a norm, a young !hite male that face recognition systems continue to use, but !ith the aim, for e(ample, of *restricting access to certain areas based on gender+ or *collecting valuable demographics+ such as *the number of !omen entering a retail store on a given day.+ 04  $he segue from disciplinary to biopo!er is, for Foucault, contingent on the increasing use of demographics and statistics that orient governance more to!ards the populace than the individual. 06  Face recognition systems demonstrate both forms of po!er and perhaps even the shift from one to the other. $his becomes clearer as !e track back from the biopolitical uses and applications of face recognition technology to the disciplinary design and architecture of the technology itself. Koray Ealci and olkan Atalay present t!o algorithms for *gender estimation.+ 09  $hey point out that the same algorithms can be used *for different face specific tasks+ such as race or age estimation, *!ithout any modification.+ 0<  "n the first algorithm, the training face images are normalized and the eigenfaces are established using /:A. 0>  /:A is described here as a statistical techniue for *dimensionality reduction andfeature e(traction.+ 0?  $he performance of the system is improved by the subseuent use of a *pruning+ algorithm, !hich identifies statistical   5 Media Fields Journal  connections e(traneous to gender 1race or age2 estimation and deletes them. *After deletion, the system is re-trained+ and the pruning is repeateduntil *all the connections are deleted.+ 0@  A performance table is produced, sho!ing the relation bet!een each iteration of pruning, the percentage of deleted connections, and the accuracy of the system. $he accuracy of gender estimation in Ealci and olkan5s e(periment actually diminishes after the eighth iteration, albeit by only a fe! percentage points, allo!ing them to claim that the system is stable. $hey maintain that pruning or the deletion of statistical connections improves gender estimation not in a linear or absolute sense but by enhancing the process of classification itself. For ;eoffrey Eo!ker and Susan eigh Star, classification is a largely invisible, increasingly technological, and fundamentally infrastructural means of *sorting things out.+ 3D  "t is an instrument of po!er-kno!ledge that is productive of the things it sorts—things such as faces that are by no means *unambiguous entities+ that precede their sorting. 30  $he e(istence of a pruning algorithm that renders faces less ambiguous testifies to their elusiveness, or their inherent resistance to classification as one mode of representationalism. "t !ould, perhaps, be going too far to suggest that there is a crisis of representationalism in appearance-based face recognition systems. o!ever, their designers and engineers are clearly a!are that faces are things that *resist depiction+ 33  because they are *comple( and multidimensional+ 34  and not *uniue, rigid+ ob%ects. 36  $he advantage of a more dynamic and relational approach to the production of faces in face recognition technology !ould include recognizing representationalism as a claim, a defensive manoeuvre in the face of faces5 non-essential ontology and dynamic co-evolution !ith technological systems. Still, this defensive manoeuvre matters in a double sense) it is both meaningful and material, reproducing norms—for e(ample, norms of gender in a machine that is learning to classify, sort, and discriminate among the population—better than it could before. "f this is a last push to representationalism, it is one that reinforces it rather than sho!s it the door. Face recognition technology upholds a belief in thee(istence of ontological gaps bet!een representations and that !hich they represent. "t also re-produces the norms of nineteenth-century disciplinary photography even as photography becomes allied to the security-based biopolitics of computational vision and smart algorithmic sorting. "n this sense, Kelly ;ates is right to argue that ne! vantage points can underscore old visions as !ell as old claims to unmediated visuality. 39  ike her, " uestion the autonomy of face recognition systems !ithout denying that, in con%unction !ith human input of various kinds, they enact 
Similar documents
View more...
Search Related
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks