Pengenalan Huruf Braille Menggunakan Radially Average Power Spectrum Dan Geometri

Abstract

Education is very important in the growth of children, including children with special needs such as people with visual impairments. Not only at school children can also develop their potential in learning anywhere and anytime. One of the most important learning processes is reading and writing. People with visual impairments carry out reading and writing activities using Braille. The problem is the lack of family knowledge about braille, so the activities accompanying blind children cannot be conducted. This study aims to convert braille into text, so it can make it easier for families to understand braille. The introduction of braille letters is done by extraction of frequency and spatial features. The methods proposed in this study are fast fourier transform (fft), radially average power spectrum (rapsv) and wavelets as well as several spatial features, namely local binary pattern (lbp), segmentation based on fractal analysis (sfta), first order statistics, gray level co-occurrence matrix (glcm), moment invariant and geometric features. In this study, the classification process was carried out using Bayes Net, Naïve Bayes, SVM, KNN and Random Forest. From several experiments it was found that the Random Forest classification method gave the best results. The rapsv method provides an accuracy of 93.91%, frequency feature extraction produces the same accuracy as the combination of rapsv with geometry, which is 94.04% and of all features, an accuracy of 97.18% is obtained.