Binarization of music score with complex background by deep convolutional neural networks

Abstract Binarization is an important step for most of document analysis systems. Regarding music score images with a complex background, the existence background clutters variety shapes and colors creates many challenges binarization. This paper presents model binarization by fusion deep convolutional neural networks. Our directly trained from image regions using pixel values as inputs binary ground truth labels. By utilizing generalization capability residual network backbone useful feature learning ability dense layer, proposed structures can differentiate foreground pixels clutters, minimize possibility overfitting phenomenon thus deal noises appearing in images. Comparing to traditional algorithms, generated our method have cleaner better-preserved strokes. The experiments captured synthetic show promising results compared existing methods.