Optična razpoznava znakov v slikah naravnih scen

PETEK, ROK (author), Šajn, Luka (mentor)

V magistrskem delu so predstavljene in opisane sodobne metode optične razpoznave znakov v slikah naravnih scen. Izbrane so bile metode, ki dosegajo visoko točnost in so robustne na osvetlitev ter ostale geometrijske spremembe. Naše delo temelji na implementaciji treh različnih metod za pridobivanje značilk. Osnovna metoda HOG, na kateri temeljita tudi ostali dve metodi je ena izmed bolj popularnih metod pridobivanja značilk. Metoda HOG je bila primarno uporabljena pri detekciji ljudi, vendar je pri razpoznavi znakov v slikah naravnih scen modificirana, za doseganje boljših rezultatov. Na metodi HOG bazira metoda PHOG, ki pretvori osnovni HOG v piramidalni sistem, ter obenem vključuje bilinearno interpolacijo. Zaradi piramidne strukture PHOG-a, je ta metoda počasnejša od metode HOG, vendar bolj natančna, saj je tudi vektor značilk večji. Tretja metoda, ki smo jo implementirali je metoda Co-HOG, ki od metode HOG podeduje vse dobre lastnosti, kot je invariantnost na različno svetlost in lokalne geometrijske spremembe. Co-HOG se razlikuje po tem, da značilke vsebujejo tudi prostorska razmerja med slikovnimi elementi, s čimer se znak bolj natančno opiše in razpozna, med drugim je tudi hitrejša metoda pridobivanja značilk. Zaradi različnih naravnih faktorjev v znakih naravnih scen je razpoznava znakov s tradicionalnimi sistemi optične razpoznave znakov nenatančna, saj ti predpostavljajo, da se znaki ne razlikujejo v pisavi in barvi, ter na monotono ozadje znaka. Pri pridobivanju robustnih značilk znakov naravnih scen se uporablja metode, ki so invariantne na velikost znaka, šum ozadja, tip pisave ter na vizualne efekte, ki pritegnejo pozornost, kot je npr. prelivanje barv v posameznem znaku. Zgoraj opisane metode ne potrebujejo klasičnega predprocesiranja in binarizacije slike znaka, kot to počnejo tradicionalni sistemi, saj te zajemajo značilke z metodami, ki opišejo videz objekta in obliko z intenziteto gradientov ter smermi robov. Metode pridobivanja značilk so bile evalvirane na različnih podatkovnih bazah, kot so ICDAR, Chars74K, CVL OCR DB. Generirali smo tudi sintetično podatkovno bazo, ki imitira znake v naravnih scenah, tako da vključuje množico različnih pisav ter šumov v slikah. Sintetična podatkovna baza znakov je bila generirana z namenom povečanja učne množice, ter izboljšanja rezultata klasifikacijske točnosti.

• Language: Slovenian
• Keywords: Optična razpoznava znakov, naravne scene, računalniški vid, HOG, PHOG, Co-HOG, SVM, ANN, K-NN
• Work type: Master's thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2016
• PID: 20.500.12556/RUL-81522
• Publication date in RUL: 11.04.2016

Secondary language

• Language: English
• Title: Optical character recognition in images of natural scenes

Abstract

This masters thesis presents and describes modern methods of optical character recognition in natural scenes. Methods with high classification results and are robust to illumination and geometric transformations were selected for the thesis. Our work is based on the implementation of three different methods for obtaining features. The basic HOG method, which also underlies the other two methods is one of the most popular feature extraction methods in object detection and character recognition. HOG method was primarily used in connection with human detection, but was adapted for character recognition also. PHOG method, which is based on HOG, converts the basic HOG algorithm into a pyramid scheme and also includes bilinear interpolation. Due to the pyramid structure of PHOG, the method is slower than the HOG algorithm, but more precise, since the feature vectors are larger. The third feature extraction method, which we have implemented is Co-HOG algorithm, which inherits all the good qualities of HOG method, such as invariance to illumination and geometric changes. Co-HOG is differs from HOG and PHOG, by its feature representation, where it also captures the spatial relationship of neighbouring pixels in order to describe the character more accurately. Among other things Coo-HOG is also a computationally faster than HOG and PHOG. Due to various factors in natural scene text images, the traditional character recognition systems produces inaccurate results, because it assumes that the characters do not differ in fonts and colors and presumes a monotonous background of images, whereas in obtaining features from natural scene images, the algorithms should be robust and invariant to character sizes, background noise, different fonts, local illumination changes and visual effects that draw attention, such as color blending. The above described methods do not require preprocessing and segmentation as traditional systems do, since the extract features with methods that describe the appearance of the object and the shape with gradient intensity and edge directions. Feature extraction methods were evaluated on a variety of databases such as ICDAR, Chars74K, CVL OCR DB. We have also generated a synthetic database of character images, that simulates characters in natural scenes, by including large variety of different fonts and noises in images. Synthetic image database was generated with the aim of increasing the training set and the improvement of classification accuracy.

• Keywords: Optical character recognition, natural scenes, computer vision, HOG, PHOG, Co-HOG, SVM, ANN, K-NN