Portrait-building software Pehchan Pro might not be able to add ear hair,but it recreates the human face from its parts and helps catch an accused.
Last saturday,Delhi residents saw a person of somewhat dubious sex peering out of their morning papers. It had thin lips,faint smile lines,fainter forehead lines,and hair that appeared to be hastily pasted-on velcro. A closer look at the caption revealed that it was one of a pack of assailants inebriated labourers,said the police whod reportedly abducted and gang raped a call centre worker whose office cab had dropped her in Dhaula Kuan early on November 24. The sketch was made from the description of her colleague,whod also been seized,but had managed to wrest herself free.
At the computer wing of the State Crime Records Bureau (SCRB),Delhi,which also uses the same portrait-building software,Pehchan Pro,12-year-old Divyanshu Malhotra watches impatiently as it loads. Hes here to make photofits of three men who tried to abduct him earlier that week,close to his school in north Delhi. Hell choose from disembodied facial components 1,200 in total to build the portraits,starting with the shape of the face.
His face was a little fat, Divyanshu says. Fat lips,too. What shape of chin? the lady officer asks him,as she adds a pair of pendulous lips. All chins look alike, the boy replies breezily. No, the officer tells him,Some are triangular,some are round Round,then. What sort of eyes? Like this? she clicks on a narrow pair. No,no. They werent so thin. She widens them a little. Add thin eyebrows to that, he says,shifting forward eagerly in his chair,peering at the thick-lipped early hominid taking shape onscreen. And make his hair a little long. Now,can you add spikes in the middle? And little hairs on his ears? No, says the officer,still intent on adjusting the hair. Well,how about that? It looks a little like him, the boy says,with an exaggerated yawn. I guess.
It might have its limitations like the lack of ear hair in its database but police officers at the SCRB,whove been using it officially since 2003,think its great. Its user-friendly,effective,and practical, says an officer. It helped in famous cases,like the bomb blasts in Jaipur,the firing at Akshardham. Its also apparently been used to identify the accused in the murders of Haren Pandya and Madhumita Shukla,the Telgi paper scam,and comes with the effusive endorsement of CBI officers,who praise its ability to convincingly age culprits whove been at large for several years. Its being used by the police in Delhi,Haryana,Kerala,Maharashtra and Gujarat,and considered to be a vast improvement over the National Crime Records Bureaus own software,introduced in 1995. That one has a limited database of facial features,all cut out from photos of convicted criminals. Its portraits looked like collages,with blackened noses sprouting out of blanched faces. It was patchy and difficult to manipulate, an officers recalls. In Pehchan Pro,everything merges beautifully.
In a way,the police nearly wished this software into existence. Tamaal Roy,director of Surat-based Gujarat Computer and Softwares Ltd,which provided software to textile industries,traces its inception to 10 years ago,when a local policeman tried out his Shringar,a makeover software that invited men and women to try new hairstyles and clothes and observed that itd be very useful as a facial composite,for detecting criminals in disguise. I consulted a variety of sources, says Roy,starting with beat officers opposite my Crawford Market office in Mumbai,who tried the software and gave me feedback to improve it. He also consulted his plastic surgeon friend,for basic parameters,like the span of the bridge of the nose,and even local sculptors who made Durga Puja idols,as they make everything lifelike,in 3D.
Later versions boasted gerontological features,for which he consulted psychologists,including his wife. He supplemented expert inputs with internet research,and what he learned at seminars on biometry and face recognition. I read up about types of ageing,and regional differences in the way character lines and smile lines develop, says Roy. For instance,I learned that paan-chewers develop strong lines around the mouth,and that,past-40,a Punjabi has lines on the forehead,but a Manipuri doesnt. He spent several fruitless weeks trying,and failing,to sell it to the NCRB. In 2003,after the Gujarat Police took his help in making a composite of Haren Pandyas killer,police departments began to buy the software from him,and got him to train their officers to use it.
The SCRB in Delhi makes an average of four photofits a day,and these are circulated in 174 police stations in the National Capital Region. But they arent sure how many help crack cases,as they claim investigating officers are reluctant to share the credit. Theres one sensational case theyre sure it helped crack,though,in 2009. That October,seven newborn boys were kidnapped from jhuggis across west and south-west Delhi. Two photofits,made from the descriptions of two young mothers,produced an identical,forbidding face: a woman with darkened pits beneath her eyes,and a scarf bound about her head. Early in December,a source spotted this person in a park in Rajouri Garden,and informed the police. A raiding party was dispatched,and the culprit was caught,along with her accomplice,who had driven the getaway auto.
The officer who put together the photofit of the woman was called to the court during the trial,as a witness. When I saw her,I knew it was her, she says proudly. She looked just like the picture I made. But would anyone be able to tell who the thin-lipped rapist was? Its all down to the witness narration, says the investigating officer. In the case of the (sunken-eyed) lady,shed befriended the witnesses over several visits,and sold them costume jewellery. But the girl in Dhaula Kuan mustve been in panic,just wanting to escape,and hadnt probably registered the culprits features. She glances at the photofit again. He looks a little ladylike. Perhaps hes pahadi?
This reveals several limitations,borne out by research. Police officers forced to find leads in an image depicting a person of indeterminate gender may draw prejudice-charged associations with photofit features. For instance: Chinky-minki,Mohammedan-jaisa,Pahadi. In a 2004 test,cognitive psychologists in the US found subjects more likely to associate violent criminal acts with those who had phenotypically black facial features. Worse,were poorer at distinguishing between faces unlike those weve grown up surrounded by something that psychologists term the other-race effect. In addition,we dont process faces sequentially,one component at a time; we process them holistically. Which might be what made little Divyanshu exclaim,midway into the face-building exercise for the second man who attacked him: Saare aise the!
Talking to a hard-bitten police detective may also bring you to the same conclusion. One,posted in a ritzy south Delhi neighbourhood,says,You know whatd be really useful? A software that simulates,not ageing,but decomposition! We find so many long-dead bodies. Theyve turned green or black; the face is bloated,double,triple; the skin has come off these cases stay unsolved because theres no way to identify the cause of death,or the bodies. He pauses,and laughs,But then again,police attitude is the best software in the world. Thats what we should really work on.