Bet­ter im­ages go­ing mo­bile

Smart­phones us­ing new apps rec­og­nize ob­jects, faces and dig­i­tal im­ages.

Los Angeles Times - - BUSINESS - By Tracey Lien

“OK, this one is pretty cool,” said 6-foot-tall Omar Tayeb as he held his left hand in front of him­self and snapped it with his iPhone.

He didn’t move the phone away. He left it there as an app de­tected each of his fin­ger­nails and pro­ceeded to paint each one. On the phone’s screen, Tayeb’s nails turned a daz­zling shade of May­belline pur­ple.

“So the app was able to de­tect where my nails were, paint them in one of May­belline’s nail pol­ish col­ors, then ad­just to the room’s light­ing so it looks nat­u­ral,” Tayeb said. “It’s quite an im­pres­sive us­age.”

And it is im­pres­sive, given that as lit­tle as five years ago the smartest smart­phones would have strug­gled to rec­og­nize the most ba­sic of im­ages, let alone iden­tify in­di­vid­ual fin­ger­nails and ac­cu­rately color them.

Early fash­ion and beauty apps used im­age-recog­ni­tion tech­nol­ogy to crudely su­per­im­pose hair­styles and clothes onto peo­ple’s bod­ies, of­ten to hi­lar­i­ous ef­fect: shirts and dresses sit­ting at awk­ward an­gles, hair­styles that looked more like clown wigs.

But tech­nol­ogy moves fast, and of all the re­search in ar­ti­fi­cial in­tel­li­gence tech­nol­ogy, im­age recog­ni­tion is mov­ing the most quickly through pop­u­lar cul­ture.

Neu­ral net­works — com­puter pro­grams roughly inspired by the work­ings of the hu­man brain — are es­pe­cially good at rec­og­niz­ing ob­jects, faces and dig­i­tal im­ages. Google uses such tech­nol­ogy to try to tell a trout from a salmon in a Google Im­age search.

Now im­age recog­ni­tion is go­ing mo­bile. Small but so­phis­ti­cated smart­phone cam­eras serve as mo­bile ma­chine eyes, scour­ing the world for im­ages with the help of their hu­man own­ers, and pro­cess­ing them right there on the phone.

More than 1 bil­lion smart­phones were sold in 2013, ac­cord­ing to re­search firm IDC. And where there are peo­ple, there are mar­keters try­ing to sell them stuff.

Like ketchup. Blip­par users can snap pic­tures of a Heinz ketchup bot­tle and get recipes and ketchup vid-

eos. Or ce­real. Kids can point their phones at a box of Lucky Charms and in­ter­act with lep­rechauns through the iPhone screen in a kind of re­al­ity-meets-com­put­er­screen mash-up known as aug­mented re­al­ity.

But Tayeb, a 28-year-old devel­oper turned founder, has loftier goals. Blip­par is build­ing a mo­bile im­age search en­gine, never mind the di­rect mar­ket­ing. Blip­par is teach­ing its plat­form to rec­og­nize ev­ery­day ob­jects.

Point your phone at a dog and it will tell you what breed it is. Point it at a book and you can read the re­views. Point it at a movie poster and it will pull up the rel­e­vant IMDB and Rot­ten Toma­toes Web pages.

Blip­par wants to make “blip­ping” — the act of scan­ning an ob­ject through the app — as ha­bit­ual as open­ing Google Maps in a search for di­rec­tions or open­ing Yelp for res­tau­rant re­views.

Blip­par is one of many com­mer­cial oper­a­tions rid­ing the im­age-recog­ni­tion wave. Semi­con­duc­tor com­pany Qual­comm Inc. in San Diego al­ready makes many of the chips that run mo­bile de­vices. It has also de­vel­oped Vu­fo­ria, an im­age-recog­ni­tion soft­ware plat­form for de­vel­op­ers that un­der­lies more than 15,000 apps with an aug­mented or vir­tual-re­al­ity com­po­nent.

Neu­ral net­works, points out Tim McDonough, Qual­comm’s vice pres­i­dent of mar­ket­ing, are able to ac- cept feed­back and learn new things.

“We call it cog­ni­tive com­put­ing, which is kind of a dorky term that makes you think of a per­son play­ing chess, but the ba­sic idea is ma­chines can learn, and they can make de­ci­sions based on what they learn,” McDonough said.

“You can train a com­puter to rec­og­nize im­ages by giv­ing it re­ally large num­bers of im­ages to learn.”

Blip­par’s im­age-recog­ni­tion tech­nol­ogy works in a sim­i­lar way: A per­son feeds the com­puter large num­bers of an­no­tated im­ages and, over time, the com­puter be­comes more ac­cu­rate at iden­ti­fy­ing what it sees. A com­puter may only need to see one im­age of a movie poster to be able to rec­og­nize that movie poster ev­ery time, but for other, more generic ob­jects, it may need to see tens of thou­sands of im­ages be­fore it can iden­tify a spe­cific type of tree or breed of an­i­mal.

Advertising is an ob­vi­ous use, with brands such as Lego, Pepsi and McDon­ald’s al­ready find­ing ways to in­cor­po­rate im­age recog­ni­tion and aug­mented re­al­ity into their mar­ket­ing ma­te­ri­als.

Com­pa­nies like Face­book are us­ing fa­cial-recog­ni­tion tech­nol­ogy to iden­tify and tag peo­ple’s faces in photos.

The tech­nol­ogy helps make driver­less cars pos­si­ble: Is that a per­son at the cross­walk or a lamp­post?

Some of the pos­si­ble uses are semi-com­i­cal. Peo­ple wear­ing vir­tual-re­al­ity head­sets, deeply im­mersed in pre­tend worlds, might not re­al­ize they’re about to trip over a real-life elec­tri­cal cord. Ma­chine eyes could pass the warn­ing — or tell them that a stranger has just en­tered the room.

Jay Wright, vice pres­i­dent at Vu­fo­ria, has his own vi­sion: He’d like to sum­mon email to show up in his field of view, type by sim­ply hold­ing his fin­gers out in front of him and to never need to use a de­vice with a screen.

“I think it’s on the five- to 10-year hori­zon,” he said.


QUAL­COMM BUILT a ro­bot equipped with its im­age-recog­ni­tion tech­nol­ogy that can track and fol­low peo­ple around.


BLIP­PAR is teach­ing its im­age-recog­ni­tion plat­form to rec­og­nize ev­ery­day ob­jects, then di­rect users to in­for­ma­tion about those ob­jects.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.