Research On The Detection And Removal Algorithm Of Staff Lines On Printed Score

Optical music recognition is the process of converting paper score into digital score.The main purpose is to process,identify and analyze the image of the scanned score and finally obtain the symbol on the score automatically and accurately.The obtained digital score is the form of music score which can be directly synthesized by computer or processed with other good application by computer.Since staff lines in the score often intersect or overlap with most musical symbols,the unique characteristics of the symbols are destroyed,resulting in difficulties in subsequent symbol recognition.Therefore,before the score recognition,it is generally necessary to first detect the staff lines in the score and remove them as many as possible without breaking the music symbols.The effect of the line deletion directly affects the accuracy of subsequent symbol recognition.In this thesis,the printed-score is taken as the research target.The detection of staff lines and the deletion of the lines are studied in depth.The staff lines are widely observed in the score image.From the global observation,the distribution of the staff lines has low rank characteristics compared with the whole score.And from the local observation,there are differences between the lines and the local texture of the notes.According to the global characteristics and local details,the staff lines detection and removal method based on low rank structure and on local binary pattern are proposed respectively.Staff lines under ideal conditions can be well deleted by using these two methods both from global and local observation,while staff lines under non-ideal conditions like curving can be deleted also by using method based on local binary pattern.The methods proposed in this thesis are verified on the existing database and the results are compared with the run length analysis and other methods.The results of the contrast experiment are further improved on the pixel error and the segment error.