| 序号 | 专利名 | 申请号 | 申请日 | 公开（公告）号 | 公开（公告）日 | 发明人 |
|---|---|---|---|---|---|---|
| 1 | 基于卷积神经网络的光学乐谱识别方法 | CN201910787063.9 | 2019-08-25 | CN110598581B | 2022-09-27 | 马学健; 董瓒; 郭玲 |
| 本发明公开了一种基于卷积神经网络的光学乐谱识别方法，包括：对乐谱图像进行谱线检测；根据谱线的位置进行谱线删除；音符分割，得到一系列音符图像；将音符图像输入到已经训练好的神经网络中完成识别。本发明采用基于图论的谱线检测算法进行谱线检测，不受乐谱图像的质量、谱线扭曲形变等影响，可以准确检测出谱线的位置；采用基于线轨迹高度+局部游程直方图算法进行谱线删除，可以有效避免过删除现象；采用基于层次分级+模板匹配的分割算法，可以有效进行音符分割且保证音符的完整性；利用卷积神经网络对分割之后音符进行识别，结果具有较好的识别精度和识别速率。 | ||||||
| 2 | 基于卷积神经网络的光学乐谱识别方法 | CN201910787063.9 | 2019-08-25 | CN110598581A | 2019-12-20 | 马学健; 董瓒; 郭玲 |
| 本发明公开了一种基于卷积神经网络的光学乐谱识别方法，包括：对乐谱图像进行谱线检测；根据谱线的位置进行谱线删除；音符分割，得到一系列音符图像；将音符图像输入到已经训练好的神经网络中完成识别。本发明采用基于图论的谱线检测算法进行谱线检测，不受乐谱图像的质量、谱线扭曲形变等影响，可以准确检测出谱线的位置；采用基于线轨迹高度+局部游程直方图算法进行谱线删除，可以有效避免过删除现象；采用基于层次分级+模板匹配的分割算法，可以有效进行音符分割且保证音符的完整性；利用卷积神经网络对分割之后音符进行识别，结果具有较好的识别精度和识别速率。 | ||||||
| 3 | 一种基于CRNN的端到端光学乐谱识别方法及系统 | CN202310681836.1 | 2023-06-09 | CN116645669A | 2023-08-25 | 姚俊峰; 单子豪; 王钰菲 |
| 本发明提供了乐谱识别技术领域的一种基于CRNN的端到端光学乐谱识别方法及系统，方法包括如下步骤：步骤S10、获取大量的乐谱图片以及对应的文本文件，基于各所述乐谱图片以及文本文件构建乐谱数据集；步骤S20、基于卷积神经网络、循环神经网络以及CTC算法构建端到端的乐谱识别模型；步骤S30、利用所述乐谱数据集对乐谱识别模型进行训练；步骤S40、利用训练后的所述乐谱识别模型进行乐谱识别。本发明的优点在于：极大的提升了光学乐谱识别的泛化性。 | ||||||
| 4 | 一种基于残差注意力Transformer的光学乐谱图像识别方法 | CN202111522531.3 | 2021-12-14 | CN114359946A | 2022-04-15 | 温翠红; 朱龙娇; 刘嘉怡 |
| 本发明提出了一种基于残差注意力Transformer的神经网络方法用于光学乐谱图像识别。该识别过程的步骤为：通过建立基于残差注意力Transformer的乐谱图像识别模型，提取乐谱图像中的音符序列特征；根据捕获的音符序列特征，将乐谱图像中的音符序列数字化。本发明采用预训练的浅层残差卷积神经网络初始化参数，并提取音符基本特征，接着利用循环神经网络对获得的音符特征和标签序列分别进行编码和解码，提取音符序列的关联信息；而基于残差注意力Transformer结构，对重点关注区域加强关注，抑制无关区域的关注，进一步提取音符序列上下文特征；同时，当训练模型时，使用并行计算的统一掩码语言模型，有效地降低了音符序列错误率，缩短了模型训练时间。 | ||||||
| 5 | 一种多章节和乐段的乐谱播放数据交互处理显示办法以及计算机程序 | CN202411617008.2 | 2024-11-13 | CN119540973A | 2025-02-28 | 王立辉 |
| 本发明公开了一种多章节和乐段的乐谱播放数据交互处理显示办法，包括如下步骤：通过光学乐谱识别技术对乐谱文档的扫描图像进行处理、分析和识别，最终获得乐谱图形及音乐语义的数字表达，并记录位置信息，并且遍历所有乐章、乐段、小节位置信息；对乐谱的音符进行基元抽取，包括符干抽取、符头抽取和符渠抽取，得到乐谱识别的核心和关键数据信息；根据单声部和多声部乐理规则，设置重复的起点位置和终点位置以及播放顺序；加载多媒体资源，根据上述步骤起点和终点以及顺序位置信息(X、Y，Z)，计算出单声部与多声部播放顺序。本发明可以实现多行乐谱和多声部数据交互处理，整合之后显示播放。 | ||||||
| 6 | 一种基于低秩结构的五线谱谱线检测和删除方法 | CN201610937180.5 | 2016-10-25 | CN106548168B | 2019-10-18 | 孟凡奥; 关欣; 李锵 |
| 本发明公开了一种基于低秩结构的谱线检测和删除方法，首先，将输入的乐谱图像进行二值化等预处理操作，得到乐谱图像，接着对乐谱图像进行低秩图像恢复和差分图像获得操作，由这两个图像矩阵进行“或”运算，得到谱线图像，最后谱线图像与乐谱图像进行“异或”运算，得到结果符号图像。与现有技术相比，本发明实现了将音符从谱线中分离出来，最有效的克服了现有大多数光学乐谱识别系统中对于中音符分离存在的技术障碍。 | ||||||
| 7 | 一种基于低秩结构的五线谱谱线检测和删除方法 | CN201610937180.5 | 2016-10-25 | CN106548168A | 2017-03-29 | 孟凡奥; 关欣; 李锵 |
| 本发明公开了一种基于低秩结构的谱线检测和删除方法，首先，将输入的乐谱图像进行二值化等预处理操作，得到乐谱图像，接着对乐谱图像进行低秩图像恢复和差分图像获得操作，由这两个图像矩阵进行“或”运算，得到谱线图像，最后谱线图像与乐谱图像进行“异或”运算，得到结果符号图像。与现有技术相比，本发明实现了将音符从谱线中分离出来，最有效的克服了现有大多数光学乐谱识别系统中对于中音符分离存在的技术障碍。 | ||||||
| 8 | 一种乐谱识别方法 | CN202111388016.0 | 2021-11-22 | CN114092946B | 2024-08-20 | 冯欣; 戴培元; 王思平; 龙建武; 兰利彬; 薛明龙 |
| 本发明公开了一种乐谱识别方法，涉及乐谱识别技术领域。本发明首先将一幅乐谱图像通过目标识别分割为通过树状结构组织的行、小节、音序；而后使用多标签分类方法处理对音序进行识别，最终通过一定的逻辑处理将提取到的树状结构组合成数字乐谱文件并输出，音序的概念：每一声部的每一节拍内音符和修饰符的集合。本方法流程有效的避免了使用同一网络而无法忽视的类间数据量差距，从而均衡各个类的样本，使每个网络都有其特异性；与基于数字图像处理的光学乐谱识别相比，使用本发明方法流程中目标检测网络进行宏观块分割，能够有效地提升对于小节间信息模糊不清，统计学边界不分明的情况，从而增强系统的健壮性。 | ||||||
| 9 | 一种乐谱识别方法 | CN202111388016.0 | 2021-11-22 | CN114092946A | 2022-02-25 | 冯欣; 戴培元; 王思平; 龙建武; 兰利彬; 薛明龙 |
| 本发明公开了一种乐谱识别方法，涉及乐谱识别技术领域。本发明首先将一幅乐谱图像通过目标识别分割为通过树状结构组织的行、小节、音序；而后使用多标签分类方法处理对音序进行识别，最终通过一定的逻辑处理将提取到的树状结构组合成数字乐谱文件并输出，音序的概念：每一声部的每一节拍内音符和修饰符的集合。本方法流程有效的避免了使用同一网络而无法忽视的类间数据量差距，从而均衡各个类的样本，使每个网络都有其特异性；与基于数字图像处理的光学乐谱识别相比，使用本发明方法流程中目标检测网络进行宏观块分割，能够有效地提升对于小节间信息模糊不清，统计学边界不分明的情况，从而增强系统的健壮性。 | ||||||
| 10 | Pen type musical score symbol input device | JP2006268859 | 2006-09-29 | JP2008089812A | 2008-04-17 | KUZUOKA HARUKI; TSURUMI ATSUSHI; SUGA KUNIHIRO; YUZUKIZAKI SAYURI |
| <P>PROBLEM TO BE SOLVED: To provide a pen type musical score symbol input device which has high operability and portability. <P>SOLUTION: The pen type musical score symbol input device comprises a pen tip 12 for writing musical score symbols to a printed medium (adaptive form 30) where positional information indicating a position on the medium and the staffs are printed, a read means (C-MOS camera 14) of optically reading the position information when predetermined musical score symbols are written to the printed medium with the pen tip, a recognition means (CPU 20) of recognizing the musical score symbols written to the printed medium referring to the positional information read by the read means, a converting means (CPU 20) which converts the musical score symbols obtained by the recognition means into reproducible musical piece data, and a reproducing means (sound source unit 26) which reproduces the musical piece data as a sound. <P>COPYRIGHT: (C)2008,JPO&INPIT | ||||||
| 11 | Dictionary pattern normalizing method in musical score recognition | JP22186083 | 1983-11-25 | JPS60114985A | 1985-06-21 | TODA AKIRA; SAITOU MASA |
| PURPOSE: To attain musical score recognition with less mis-recognition by normalizing a dictionary pattern based on the interval of 5 lines so as to make the size of a musical score data after extraction of a feature pattern coincident with that of the dictionary pattern surely. CONSTITUTION: A interval L between five lines is obtained together with a musical score data ESD from which a characteristic pattern is extracted from a musical score data SD read optically. A dictionary pattern DP prepared in advance is a normalized dictionary data NOP normalized into the size of the musical score data ESD based on the interval L of five lines, and the normalized dictionary data NOP and the feature extracting dictionary data are compared so as to recognize the musical score. COPYRIGHT: (C)1985,JPO&Japio | ||||||
| 12 | 촬영된 음악 악보 영상의 자동연주를 위한 가사 영역 추출장치 및 방법 | KR1020160082018 | 2016-06-29 | KR101793184B1 | 2017-11-03 | 양형정; 민딩콩 |
| 본발명은촬영된음악악보영상의자동연주를위한가사영역추출장치및 방법에관한것으로, 촬영된악보영상의자동연주를위해빈도수, 투영, 오선의높이, 위치등의정보를이용하여악보의광학악보인식(Optical Music Recognition:OMR) 과정중에가사영역을추출하는촬영된음악악보영상의자동연주를위한가사영역추출장치및 방법에관한것이다. 상술한바에의하면, 악보의가사영역추출에서적은계산량으로도높은인식률을낼 수있고, 오선과마디선정보를이용해왜곡을보정하여가사영역을추출할수 있는효과가있다. | ||||||
| 13 | Musical score recognition device | JP27514195 | 1995-09-29 | JPH0997060A | 1997-04-08 | NAKANO SEIJI; SUMIDA REN; HINO TETSUO; OOBA ATSUSHI |
| PROBLEM TO BE SOLVED: To efficiently recognize a musical score by separating parts which constitute thin lines from musical score image data, detecting symbols composed of thin lines from the separated thin-line images, and detecting thick symbols from the remaining thick image. SOLUTION: A scanner 12 optically scans a musical score and generates binary or gray-scale image data, and the read image information is inputted to a RAM 3 or HDD 4 through a scanner interface circuit 13. Here, the parts constituting the thin lines are separated from the musical score image data, and the symbols composed of thin lines are detected from the separated thin-line images. Then the thick symbols are detected from the remaining thick image. Thus, the thick part, thin longitudinal lines, and thin lateral lines in the musical score image are separated as the preprocessing of musical score recognition to constitute a recognizing method which is sensuously close to a read of the musical score by a human, and the symbols or components of the symbols can efficiently be recognized from separated images to improve the recognition rate. | ||||||
| 14 | Music recognition device | JP27513395 | 1995-09-29 | JPH0997058A | 1997-04-08 | NAKANO SEIJI; SUMIDA REN; HINO TETSUO; OOBA ATSUSHI |
| PROBLEM TO BE SOLVED: To shorten a recognition processing time and to improve a recognition rate by scanning musical score image data at right angles to the five staffs, and finding the line width and intervals of the five staffs from pixel length frequency distribution data and calculating resolution and/or density. SOLUTION: A scanner 12 optically scans a musical score and generates binary or gray-scale image data, and the read image information is inputted to a RAM 3 or HDD 4 through a scanner interface circuit 13. Here, the music scale image is scanned at right angles to the five staffs to find the length of successive white or black pixels, and frequency distribution data by the length are generated. The line width and intervals of the five staffs are found from this frequency distribution data, the resolution and/or density of the inputted musical score image data are calculated from the found line width and intervals, and the picture quality information is displayed on a CRT 6. Consequently, when the resolution and density of the musical score image data are not within a range required for recognition, they are displayed and image data can be taken in again. | ||||||
| 15 | System for recognizing musical score | JP17690382 | 1982-10-07 | JPS5966784A | 1984-04-16 | NAGAO MAKOTO; SAITOU MASA; TODA AKIRA |
| PURPOSE:To improve the recognizing efficiency, by setting a window in response to a recognition object of a musical score to recognize the object with a histogram in the window or its movement. CONSTITUTION:The density data of the musical score 1 for one record's share is read optically by a scanner or the like to obtain the density distribution curve (histogram) 2 in the abscissa direction X. Then, the density distribution curve 2 in the direction X read in this way is converted into the density distribution curve (histogram) 3 in the ordinate direction Y. That is, the density distribution curve 3 in the direction Y which is correspond to the X distribution is obtained by accumulating the values of the curve 2 in the direction X and moving sequentially the accumulation upward or downward. | ||||||
| 16 | Music recognition device | JP27514295 | 1995-09-29 | JPH0997061A | 1997-04-08 | NAKANO SEIJI; SUMIDA REN; HINO TETSUO; OOBA ATSUSHI |
| PROBLEM TO BE SOLVED: To separate the paragraphs of a score, recognizes the correspondence of parts, and generates score-recognized data in time series, track by track, by recognizing a paragraph according to the connectivity of the ends of the five staffs detected from musical score image data. SOLUTION: A scanner 12 optically scans the musical score and generates binary or gray-scale image data and the read image information is inputted to a RAM 3 or HDD 4 through a scanner interface circuit 13. Here, the paragraph is recognized from the connectivity of the ends of a plurality of the five staffs in an image area, and whether or not there is the connectivity of the paragraph with precedent and following images at the end of the image is detected. Further, whether or not there is a large parenthesis or large slur on the left side of each paragraph is recognized to separate respective parts, thereby making the parts correspond, paragraph by paragraph. Consequently, the five staffs can accurately be grouped by the paragraphs and the parts can be made to correspond between the programs. | ||||||
| 17 | Toy for learning musical score | JP2004171792 | 2004-06-09 | JP2005352087A | 2005-12-22 | KIRIFUCHI CHIZUKO |
| <P>PROBLEM TO BE SOLVED: To provide a toy for learning musical score enabling a user to clearly recognize a musical piece corresponding to a musical score without depending on the user's skill. <P>SOLUTION: The toy 1 for learning musical score comprises: a light shielding sheet member 19 on which a plurality of musical note patterns 23 are drawn and each note head 25 is made optically transparent or light transmissive; and a main body 3 having a sheet fitting face 5 to which the sheet member 19 is attachably and detachably fitted. The main body 3 comprises: a plurality of array switches 7 arrayed on the sheet fitting face 5; a plurality of light emitting elements L01-L32 provided in correspondence with the array switches 7 and make the array switches 7 gleam; a loud speaker 11; and a control section 27 which makes the plurality of light emitting elements L01-L32 emit light in a prescribed order and makes the speaker 11 output sounds having the length and pitch corresponding to the prescribed order in response to the operation of the array switches 7 being made to gleam by the light emitting elements. When the sheet member 19 is fitted on the sheet fitting face 5, each note head 25 aligns the position of the plurality of array switches 7. <P>COPYRIGHT: (C)2006,JPO&NCIPI | ||||||
| 18 | Recognizing system of musical score | JP12151482 | 1982-07-13 | JPS5911470A | 1984-01-21 | NAGAO MAKOTO; SAITOU MASA; TODA AKIRA |
| PURPOSE:To recognize a note and a bar line from the ratio of a peak area to the area of a prescribed width at the left and right side of the peak area, by reading density data for one code's portion of a musical score with an optical scanner and obtaining a density distribution curve in the lateral axis for detecting the peak position from the distribution curve. CONSTITUTION:The density data for one record's portion of the musical score 1 is read by a scanner to obtain the density distribution curves 2-4 in the lateral direction X. In obtaining the curves 2-4, the density data of the locations N1, N2 having notes are accumulated along the axis to output peaks PC1, PC2 of the density data. The maximum specified value of the distributing curve 3 of the longitudinal direction Y is assumed as a threshold value X0, the distributing curve 3 is cut at the level and the effect of the density of five lines is eliminated. The peak values PC1, PC2 are obtained from the distribution curves 2-4 in the cut lateral direction X, thereby discriminating the notes and the bar lines from the ratio of the area of the peak position to the area of a prescribed widths D1, D2 at the left and right side of the peak area. | ||||||
| 19 | Electronic watch with musical scale display function | JP5565279 | 1979-05-09 | JPS55149087A | 1980-11-20 | TAKAHASHI NAOKI |
| PURPOSE:To make possible the distinct displaying of scores and display identification of the sounding scales by integrally optically displaying the scales of the plural notes being output by voice and displaying the notes being output by voice. CONSTITUTION:An FF is set by the concidence signal (f) at the reaching of the alarm set time from a clock circuit 2 and the scales of the plural notes of the musical piece are integrally optically displayed in the score display part 5 of a liquid crystal display device 3. At the same time, those under sounding out of the scales of the display part 5 according to the step count values of a duodecimal counter 11 by a sound length control circuit 10 are indicated by way of the subsequently selected and driven electrodes 6a, 6b... of a sounding condition display part 6. These indicated scales are subsequently read out from a ROM8 through an address part 9 and correspond to the scales being subsequently voice-output from a speaker 19 by way of a frequency select circuit 17, sound volume control circuit 18, etc. which are controlled by the scale codes and sound volume codes through a buffer 13, thus the distinct musical piece scale display free from color facing, decoloring, etc. is accomplished by the optical display which does not depend upon printing or the like and the identification of the sounding sclaes becomes possible. | ||||||
| 20 | Grouping system in recognition of musical score | JP17284982 | 1982-10-01 | JPS5962985A | 1984-04-10 | NAGAO MAKOTO; SAITOU MASA; TODA AKIRA |
| PURPOSE:To facilitate succeeding recognition processings, by detecting peaks on a basis of the density distribution in the direction of the horizontal axis which is cut by a prescribed rate of the width of a five staff and grouping information into independent notes, groups of notes, and symbols in accordance with the number of peaks. CONSTITUTION:Density data of one-record components of musical score is read optically, and a density curve 2 in the direction of the horizontal axis is converted to a density distribution curve 3 in a direction Y of the vertical axis. For example, 80% of a maximum value Xmax of the curve 3 is used as a threshold, and this threshold X0 is used to cut a quantity Y0 for X0 of the curve 2, and thus, a density distribution curve 4 only for note information of notes, bar lines is attained. This density data on the score attained in this manner is sampled and measured at intervals of a prescribed time to determine a peak value of the histogram. An independent tone is discriminated in case of one peak, and a symbol is discriminated when the interval between two peaks is <=2/3 of a five staff width GP, and a group of notes is discriminated when two or more peaks exist and the interval between adjacent peaks is >2/3 of the five staff width GP. | ||||||
QQ群二维码
意见反馈
意见反馈