Prepoznavanje rukopisa strojnim učenjem i primjena u identifikaciji popunjenih polja na uplatnici

undergraduate thesis

• Title (english): Handwriting recognition with machine learning and its application for text identification on payment slips
• Author: Josip Mrđen
• Mentor: Domagoj Jakobović (mentor)
• Committee member: Domagoj Jakobović (predsjednik povjerenstva)
• Committee member: Marko Đurasević (član povjerenstva)
• Committee member: Martin Soldić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-13, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Napretkom u računarskoj znanosti te razvojem dubokog učenja, područje računalnog vida je u zadnjem desetljeću dobilo velik zamah. U moru primjena, detekcija i prepoznavanje teksta (bilo tiskanog ili pisanog) je jedno od istraženijih područja te domene. Ovaj završni rad bavi se upravo prepoznavanjem teksta, za potrebu ekstrakcije teksta na rukom pisanim uplatnicama. Za potrebe toga korištene su konvolucijske neuronske mreže za čisto prepoznavanje teksta te elementi pretprocesiranja slika i detekcije pravopisnih elemenata kao dio izvan domene strojnog učenja. Klasifikacija slika u slova i brojke uz pretprocesiranje pokazala se kao uspješan pothvat s visokom stopom točnosti, dok je detekcija zbog različitog šuma na svakoj uplatnici te jednostavnih detekcijskih metoda bila glavni uzrok nezadovoljivosti dobivenih podataka na određenom broju uplatnica.

Abstract (english)

With the development of computer science (particularly deep learning) in the recent years, the computer vision gained a big swing of advancement. In the sea of applied techniques, detection and text recognition (wether printed or handwritten) is one of the most explored fields of that domain. This paper deals with text recognition in the aim of its extraction on handwritten payment slips. For that purpose, convolutional neural methods were used for plain text recognition while the methods for preprocessing of images and detection of the ortographic elements were excluded from machine learning domain. Classification of pictures into numbers and letters with a little preprocessing showed to be a successful method for text recognition with a high rate of accuracy. On the other hand, detection of ortographic elements was the main reason for unsatisfiable results on some of the slips, due to the noise on the cover of each slip and the simplicity of the detection methods used.

• Keywords: Prepoznavanje teksta; Konvolucijske neuronske mreže; Duboko učenje; Računalni vid; Segmentacija slika; Detekcija objekata
• Keywords (english): Optical character recognition; Convolutional neural networks; Deep learning; Computer vision; Image segmentation; Object detection
• Language: croatian
• URN:NBN: urn:nbn:hr:168:320512
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-13 23:17:02