Segmentacija rukom pisanih znakova korištenjem klasifikatora

master's thesis

• Title (english): Handwritten Character Segmentation Using Classifiers
• Author: Ivan Čapalija
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2016-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu su opisana i testirana dva sustava segmentacije rukom pisanih znakova. Prvi sustav se temelji na standardnim metodama računalnog vida kao što su: pomični prozor, predlaganje regija interesa i pretjerana segmentacija. Drugi sustav je baziran na dubokom učenju i gotovo je u cijelosti implementiran standardnim slojevima konvolucijskih neuronskih mreža.
Rezultati testiranja oba sustava su prikazani i objašnjeni. Pokazalo se da su lokalizacija i klasifikacija drugog sustava usporedivi s onima iz prvog sustava, ali brzina segmentacije drugog sustava je skoro dva reda veličine veća i to znači da se on može koristiti u stvarnom vremenu. Testiranjem je utvrđeno da oba sustava imaju prostora za poboljšanje, ali veće poboljšanje je moguće ostvariti u drugom pristupu. Izrazito bi teško bilo osposobiti prvi sustav da radi brzo kao drugi.

Abstract (english)

In this thesis, two systems for segmentation of handwritten characters have been elaborated and tested. The first system is based on standard computer vision methods such as: sliding window, region proposal and over-segmentation. The second system is based on deep learning and almost all components of this system are implemented as standard layers used in convolutional neural networks.
Testing results of both systems are shown and explained. It is concluded that classification and localization results of both systems are similar, but segmentation process of the second system is almost 100 times faster and that means that the second system can be used in real time. Results of testing showed that both systems can be improved, but bigger improvements can be made on the second system. It would be extremely difficult to improve the speed of the first system to match the speed of the second system.

• Keywords: računalni vid; optičko raspoznavanje znakova; segmentacija; duboko učenje; konvolucijske neuronske mreže
• Keywords (english): computer vision; optical character recognition; segmentation; deep learning; convolutional neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:168:642609
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2018-02-06 08:16:06