Restoring image orientation and detecting inverted image for printed music document

Abstract

Optical Music Recognition (OMR) is a technique for converting printed music documents into computer readable formats. For many OMR systems, the skewness of an input document is a critical issue and if the image is skewed, the recognition accuracy of these OMR systems may drop dramatically. In this paper, we introduce an image orientation restoration approach based on image contour and apply it to OMR application. This technique in conjunction with our another technique, the inverted image detection, can achieve image orientation restoration for arbitrary image rotation. From a series of experiments, the error for our orientation angle estimation is below 0.20 for any possible document rotation and the accuracy of inverted music score detection is up to 98.89%. Our approach can effectively restore image orientation and therefore gets rid of the problem of image rotation in OMR applications.