Mask Matching for Low Resolution Musical Note Recognition

In this paper, we deal with the recognition of musical scores in a low resolution image captured by the digital camera of a mobile phone. Compared to clean, scanned images, low resolution images contain music scores with severe distortions or irrecoverable discontinuities. Most of existing approaches for scanned music scores cannot be applied to the scores with such damages. In this paper, we present mask based approach to cope with incomplete information in the low resolution image. We mainly focus on musical notes in this paper. After the separation of objects in the score into symbols and musical notes, musical notes are identified by mask matching. For the head of a musical note, a 2D head mask is employed while 1D mask is used to extract features for tails. Experimental results show the effectiveness of our approach with the recognition rate of close to 100% in sample images.