Staff-line Removal for Music Score Images using U-net

국문 초록 (Abstract)

악보영상의 오선제거는 악보 인식의 향후 과정에 영향을 미치므로 중요한 문제이다. 본 논문에서는 두가지 단계로 구성된 악보영상의 오선제거에 대한 새로운 방법을 제시한다. 먼저 super-resolution을 이용하여 입력영상의 화질을 개선한다. 오선제거 성능은 많은 경우 입력영상의 화질에 좌우된다. 그리고 나서 전통적인 심층학습네트워크인 CNN 대신 U-net을 이용하여 오선제거를 시도한다. U-net은 객체분할에 좋은 성능을 보이는 심층신경망으로 알려져 있으며 본 논문에서는 이러한 객체추출 및 분할성능에 탁월한 U-net을 활용함으로써 성능을 개선하고자 한다. 제안된 방법은 ICDAR/GREC 자료셋에 대하여 실험을 수행하였으며 기존방법보다 우수한 결과를 생성하였다.

다국어 초록 (Multilingual Abstract)

Staff- line removal from the music score images is important because it directly influences the subsequent procedures for music score recognition. We propose a novel technique for staff-line removal in music score images, which is is composed of two steps. Firstly, a super-resolution method to enhance the quality of music photos is is used as a preprocessing step. The performance of the staff- line removal task is often dependent on the quality of the original music score image, hence, the preprocessing can enhance the staff- line removal performance. Then, a modified U-net model is is used, instead of conventional Convolutional Neural Network (CNN), to remove staff lines from previously enhanced images. U-net has been proved proven to be an effective deep neural network model, particularly in object image segmentation, which has been adopted for staff-line extraction and removal. The proposed approach is is evaluated on the ICDAR/GREC dataset and the experiment showss an improved results than compared to existing methods.

목차 (Table of Contents)

• 요약
• Abstract
• 1. Introduction
• 2. Related Works
• 3. Proposed Method
• 4. Experimental Result
• 5. Conclusion
• References

참고문헌 (Reference)

1 Olaf Ronneberger, "U-Net : Convolutional Networks for Biomedical Image Segmentation" 234-241, 2015

2 Alicia Forn´es, "The ICDAR/GREC 2013 Music Scores Competition: Staff Removal. International Workshop on Graphics Recognition" 2013

3 Minh-Trieu Tran, "Super-resolution in Music Score Images by Instance Normalization" (사)한국스마트미디어학회 8 (8): 64-71, 2019

4 Jorge Calvo-Zaragoza, "Staff-line detection and removal using a Convolutional Neural Network" 28 : 665-674, 2017

5 Aishik Konwer, "Staff line Removal using Generative Adversarial Networks" 1103-1108, 2018

6 Hyung Je Cho, "Recognition of Piano Score Using Skeletal Lines and Run-Length Information" 2 (2): 461-473, 1996

7 Christian Ledig, "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network" 105-114, 2017

8 Jorge Calvo-Zaragoza, "Music staff removal with supervised pixel classification" 19 : 211-219, 2016

9 K. Gregor, "Learning fast approximations of sparse coding" 399-406, 2010

10 C. Dong, "Learning a deep convolutional network for image superresolution" 184-199, 2014

11 Phillip Isola, "Image-to-Image Translation with Conditional Adversarial Networks" 5967-5976, 2017

12 C. Dong, "Image super-resolution using deep convolutional networks" 38 : 295-307, 2016

13 Bee Lim, "Enhanced Deep Residual Networks for Single Image Super-Resolution" 1132-1140, 2017

14 Lukas Tuggener, "DeepScores - a dataset for segmentation, detection and classification of tiny objects" 2018

15 Fornés Alicia, "CVC-MUSCIMA : A ground truth of handwritten music score images for writer identification and staff removal" 15 : 243-251, 2011

16 A. Dutta, "An Efficient Staff Removal Approach from Printed Musical Documents" 1965-1968, 2010

17 C. Dalitz, "A comparative study of staff removal algorithms" 30 (30): 753-766, 2008

18 Thierry Geraud, "A Morphological Method for Music Score Staff Removal" 2599-2603, 2014