TeMu-app: Music characters recognition using HOG and SVM

Conventionally, music sharing has been done through two ways: aural transmission and in the form of written documents which is normally called musical scores. As many of these paper based scores have not been published they are subjected to be damaged. To preserve the music an application that has the capability of digitalizing these symbolic images and creating new scores is required. Meanwhile, learning how to read a music score and, then, playing it on a musical instrument are difficult tasks to most beginner music learners. Therefore, an automatic system to understand the music score and to play its rhythms would ease their learning process. In this paper, a mobile application is developed to reach the mentioned aims. Proposed algorithm consists of several key steps including: (1) preprocessing in which the skewness and illumination issues are fixed, (2) segmentation in which the symbols are extracted followed by staff line detection and erosion, (3) feature extraction where the HOG discriminative features make the feature space, and, (4) recognition to which a multi-class SVM is applied. It was observed in the course of experiments that the propose method is resists against affine transformation and reach the accuracy of 94.24% in recognition process.