Open AccessMaximally Stable Extremal Regions and Naïve Bayes to Detect Scene Text
Author(s) -
Ednawati Rainarli
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/879/1/012106
Subject(s) - naive bayes classifier , bayes' theorem , artificial intelligence , thresholding , pattern recognition (psychology) , bayes classifier , computer science , classifier (uml) , mathematics , image (mathematics) , bayesian probability , support vector machine
This study examines the performance of Maximally Stable Extremal Regions (MSER) and Naïve Bayes in detecting scene text. The variance of types and sizes of fonts, uneven lighting conditions, the text orientation, a complex background, occlusion, and the presence of objects that resemble text, make the scene text detection is quite challenging. The initial stage of the detection process is to use MSER to get the candidate characters in the image. The validation process of the candidate character uses the Naïve Bayes classifier, which we trained using char74k and CIFAR10 data sets. The classification process used HOG as the extracted features. The system validates the candidates by comparing the Naïve Bayes probability value with the specified threshold value. By using 100 images from ICDAR 2015, the research obtains a 50% reduction of the candidate with an accuracy increase of 8% for Naïve Bayes using threshold values. The result shows that Naïve Bayes with the thresholding value is better than the usual Naïve Bayes classification in selecting candidates.