Extracting Notes from Chinese Gong-che Notation Musical Score Image Using a Self-adaptive Smoothing and Connected Component Labeling Algorithm

Abstract

In this paper, we present a novel algorithm to segment a Chinese Gong-che Notation (GCN) musical score image and extract notes from the image. First, a self-adaptive smoothing algorithm was used to segment the image with an X-axis function that shows the number of object pixels in each pixel column of the image. The X-axis function uses X-projection to compute the number of flex points. Next, the algorithm iteratively smooth the function and computes the number of flex points of the next smoothed function, until the number of flex points between both functions are equal. Finally, the image is segmented into several sub-images using the X-axis values for flex points in the function. For extracting notes from the image, we first obtain all connected components in the image using a conventional connected component labeling algorithm and compute the minimum boundary-box of all connected components. Next, the algorithm decides which connected component belongs to which subimage. The 50 GCN images used for testing are musical score images from Nashu Studio Theatrical Music. Experimental results show that the correct rate of segmentation is 98.9%, and the loss rate of notes is 2.705%.