Näo tuvastamine eriti madala kvaliteediga pildil
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Aastate jooksul on piltide töötlemine ja analüüs arenenud pakkudes nüüd igapäevastele väljakutsetele praktilisi lahendusi. Uute lahenduste ja ettepanekute sünd toob kaasa ka uusi väljakutseid, mis on paratamatult seotud innovaatiliste uuendustega. Olemasolevad näotuvastuse algoritmid on hästi toiminud ja neid on muu hulgas rakendatud sellistes lahendustes nagu sotsiaalmeedia kujutise märgistamine, mobiiltelefoni näo biomeetriline autentimine ja sisserände piirikontrolli näotuvastus. Põhjus miks need algoritmid on suutnud eelnimetatud stsenaariumides hästi toimida tuleneb sellest, et kasutuskõlblike kujutiste kvaliteet on tavaliselt kõrge eraldusvõimega [1].Teistes näidetes kus näotuvastus vajalikuks osutub nagu linna turvakaamerad, lennujaama kaamerad ja muud situatsioonid kus kujutise salvestuskvaliteeti ei saa kontrollida või manipuleerida, muutub jõulisema lahenduse leidmine pea kohustuslikuks, et oleks võimalik nägu tuvastada sõltumata kaadri suurusest, valgusoludest, rassist, vanusest, kehaasendist või muudest varieeruvatest faktoritest, mis võivad oluliselt muuta algoritmide võimet kujutistest aru saada.Käesoleva töö eesmärk on tuvastada ja testida alternatiivseid meetodeid näotuvastusülesannete täitmiseks äärmiselt madala kvaliteediga piltides.
Image processing and analysis have evolved over the years into providing practical solutions to everyday challenges. The birth of new solutions and proposals also create new challenges usually surrounding the new innovations.Existing face recognition algorithms have performed well and they have been deployed into solutions such as social media image tagging, mobile phone facial bio-metric authentication, immigration border control face matching among other solutions. The existing algorithms have been able to perform well in these scenarios because of the quality of the image from these use cases are usually of high quality with high resolution (HR) [1]. In other possible application of face recognition such as city camera surveillance, airport security surveillance and other related scenarios where image stream quality cannot be directly controlled or manipulated, it becomes imperative to seek a more robust solution that can deal with face recognition regardless of the frame size, lighting condition, race, age, pose and other varying factors that can significantly change the way the images are perceived by existing algorithms.The goal of this thesis is to identify and test alternative methods of performing face recognition task in extremely low-quality images.
Image processing and analysis have evolved over the years into providing practical solutions to everyday challenges. The birth of new solutions and proposals also create new challenges usually surrounding the new innovations.Existing face recognition algorithms have performed well and they have been deployed into solutions such as social media image tagging, mobile phone facial bio-metric authentication, immigration border control face matching among other solutions. The existing algorithms have been able to perform well in these scenarios because of the quality of the image from these use cases are usually of high quality with high resolution (HR) [1]. In other possible application of face recognition such as city camera surveillance, airport security surveillance and other related scenarios where image stream quality cannot be directly controlled or manipulated, it becomes imperative to seek a more robust solution that can deal with face recognition regardless of the frame size, lighting condition, race, age, pose and other varying factors that can significantly change the way the images are perceived by existing algorithms.The goal of this thesis is to identify and test alternative methods of performing face recognition task in extremely low-quality images.