JPS6072087A - Music sheet recognizing device - Google Patents

Abstract

PURPOSE:To execute a music sheet recognition at a high speed and in parallel by using the respective exclusive processors for processings such as generation of a histogram, segmenting of a peak, matching, etc. which are required for recognizing a music sheet. CONSTITUTION:A density distribution in the horizontal axis direction of a density data of every one record of a music sheet, and a density distribution in the vertical axis direction are derived by an X histogram generating exclusive processor 2 and a Y histogram generating exclusive processor 3, respectively. A peak position of the density distribution is detected by a peak segmenting exclusive processor 5. From the density distribution in the vicinity of a segmented peak position, a black circle note and a bar line, and a music symbol and a white circle note are recognized by a one-dimensional matching exclusive processor 6 and a two-dimensional matching exclusive processor 7, respectively. An optical scanner 1 and the whole exclusive processor are controlled by a main processor 8.

Description

[Detailed Description of the Invention] This invention performs processing such as creating histograms, extracting peaks, and matching necessary for musical score recognition. IJj respectively
This invention relates to a music score recognition device that can perform high-speed parallel recognition by using a professional sensor.

Music score recognition involves optically reading the density data for the lou chord, determining the density distribution in the 11 horizontal directions (X histogram), converting it to the density distribution in the vertical axis direction (Y histogram), and converting it to the density distribution in the vertical axis direction (Y histogram). A predetermined ratio with respect to the maximum value of the density distribution is determined as Threnskjöld, and by using this Threnskjöld to count the e degree distribution in the horizontal axis direction (Geta I)J'J), the density influence of the -Ji line of the musical score is removed, and the staff From the concentration distribution in the horizontal axis direction canted by a predetermined ratio of the width of the peak, the device detects the peak and extracts the peak), and from the ratio of the area of the peak position and the area of the predetermined width on the left and right of the peak position, the γY mark of the black circle musical notes, residual notes, etc.) and bar lines (one-dimensional matching), and musical notation symbols (#,
shi, -, etc.) and white circle i'l? , T (whole note, three-way eight
Name 1) requires, for example, a dictionary r゛ consisting of data of 1°° and "-1°" for the target symbol, etc., and a dictionary f° consisting of data of "1゛° and "0°''. is prepared, and the target time of munching is extracted from the recognition target area with the data names "・l" and "O", and fX, 3=b or f
If 'Xg=c' is taken, and the difference region of the 1-year-old bacterium to be recognized is β, then the similarity is calculated using the formula Sfg=b# or Srg=cQ7 (two-dimensional matching). be.

Figure 1 shows a flowchart showing such a series of processing. Conventionally, the next record 1 is processed after completing the series of processing for the roux court, and many 11ν are required for musical score recognition. It took a while. Therefore, by using dedicated processors for row processing and performing 1r column processing, high-speed musical score recognition can be achieved by 11 steps. Therefore, an object of the present invention is to provide a musical score recognition device that performs each process necessary for musical score recognition in parallel at high speed using dedicated processors.

This invention will be explained below.

FIG. 2 is a diagram showing a schematic configuration of an embodiment of the present invention,
An optical scanner 1 that reads the density data for each code of the musical score, and the density of the density data in the horizontal axis direction! 1, a processor 2 dedicated to creating an X histogram, a Y histogram 1, creating a density distribution in the vertical axis direction, and a processor 3 for creating r/, f.
and a processor 4 dedicated to Getari Nori, which subtracts a certain value from the density distribution created by the processor 2 for creating X Hisdog Grad or a processor 3 dedicated to creating YHis I/gram to create a new density distribution, and a processor 4 dedicated to Getari Nori that creates a new density distribution. From the density distribution near the peak position, 1, j, which is extracted by the professional sensor 5 for extracting the detected peak and the processor for extracting the peak, the black circle γ is marked (quarter 1
′? A processor 6 dedicated to one-dimensional matching recognizes musical score symbols (#, etc.) and bar lines.

A two-dimensional munching-dedicated professional sensor 7 that recognizes notes (b atmosphere, ) and white circles (whole notes, three-way notes), and a main professional sensor 8 that controls the optical scanner l and the entire dedicated processor
It is configured by 1 to 1 to enable high-speed musical score recognition by performing the -temperament processing simultaneously and in parallel.

In such a configuration, the optical scanner 1 first reads the dark lW data of the first record of ff9, that is, record 1.l, as shown in FIG. 3(A), according to instructions from the main processor 8. and send this to
Start reading the concentration data of 1 and 2.

×His I Gram? The professional sensor 2 dedicated to the '1'' generation calculates the density distribution curve (X histogram) 10 in the horizontal +111+ direction X from the record J, +7) density data as shown in Figure 2 (B). 8) Between the places where there is a stroke like a stroke, at N2, the concentration data of the H mark needle is accumulated along the +11+, so in the same figure (B); iie -f !tn <Peak II of concentration data will be output as P I , P 2 , P 3. The X-Hist I/gram creation processor 2 will thus create the
When O is entered, it is sent to the Y histogram creation dedicated processor 2 and gap cut dedicated processor 4, and the creation of the X histogram of the record 2 that has already been read by the optical scanner 1 is started.

Processor 3 dedicated to Y histogram creation is X of record 1
The histogram 10 is converted into a density distribution curve (Y histogram) II in the vertical axis direction Y as shown in FIG. 3(C).

That is, (1 step) of the X histogram 10 in FIG. 3(B) is accumulated in the horizontal axis direction, and the cumulative operation is sequentially
■ By moving to three directions (or two directions), the same figure (
As shown in C), X histogram 11 is obtained for X histogram 10. Next, the Y histogram creation processor 3 generates a maximum of 4 fj of this Y histogram, , 11.
For example, 80χ of X ma is taken as Thrensholt xo, the corresponding ink Y0 is set, and this is cut by a dedicated process5. After sending it to Su4.

Start creating the Y histogram of the X histogram of record 2, which has already been created by the X histogram creation dedicated process, S2.

The processor 4 dedicated to gap cutting generates the X histogram lO of the record 1 created by the processor 2 dedicated to creating the X histogram.
Dedicated process for creating Y histogram from Y0
cuff as shown in Figure 3(D). As a result, a density distribution curve 12 containing only 59 pieces of music information at the notes (including rests) and bar lines, which have been removed from the staff density data, can be seen as shown in Figure 31 (E). .This means that when the density distribution of score 9 is determined by the cumulative value in the vertical direction as shown in the 31st Δ(A) and (B), the !j° line is always within the scanning range of score 9, and the 11th line This method takes into account the fact that the density data of the music score 9 is always accumulated and appears as the density 4j of the score 9, and by removing only the density data of the h-line from the whole, the notes written on the staff are The density data of only the music information such as 'l, 7 and small rrJ1 line is extracted.The brosé sensor 4 dedicated to Kutarinori creates the density distribution curve 12 of only the music information of record l in this way. Then, after sending this to a processor dedicated to peak extraction 5, a processor dedicated to one-dimensional matching 6, and a processor dedicated to two-dimensional processing 7, similar processing for record 2 is started.

The peak extraction dedicated processor 5 converts the density distribution curve 12 of only the musical score information of the record 1 into the width of the staff, for example, 8.
By comparing the size at 0% height xI, the fourth
As shown in the figure, the peaks PC and P of the concentration distribution curve 12
Detect O2. This means that the striking needle or bar line of the ¥r mark has been detected, and the peak extraction dedicated process 5 sends the positions of these peaks to the one-dimensional munching dedicated IT, I processor 6, and then records 2. Start similar processing for .

■Prosensor 6 dedicated to dimensional matching has gap cut +J,
Of the density distribution curve 12 of only the music information of the record 1 created by the professional processor 4, for the part with the peak detected by the dedicated processor 5, the black circle is determined from the density pattern of that part. Recognizing musical notes and small # lines 6 First, the ■dimensional machining processor 6 calculates the area S2 of the distribution curve within a predetermined range @D+ from the center to the left and right for the peaks PC1 and PO2.
Set them as shown in Figures (A) and (B). moreover,
Place 4 away from range Dl by distance #d to the left and right outside respectively
Then, find the areas S1 and Sj of the distribution curves within the range of distance M07, respectively. In this way, peak pc,
, the area S2 of the central portion D1 of the concentration distribution curve for PO2, and the area S, , S:I of the concentration distribution curve in the range of a predetermined pressure 111 (tD 4 DI) from the center. area, area S2 to II'i mouth J'j (St +
s3. ) compared in size.

s, + SA> s, ノ20% −・−・−(1)
In the case of , it is recognized that there is a dot mark at the peak position, and 20z of S, +33 ≦82... (2)
In this case, the peak position is recognized as a bar line.

This is based on the fact that there is no music information such as noteheads near the bar line, and there is music information such as special privileges near the notes. Note that Figure 5 (A) shows the case of a bar line because 5I = S3 = O and satisfies the above formula (2), and rA (B) satisfies the above formula (1), so if there is a note, Recognized. When a note and a bar line are identified as shown in H below, the one-dimensional matching processor 6 identifies the note head H and strike F of the note as shown in FIG. 6(A).
A rectangular readout window W is set in the region including Calculate the concentration distribution in the axial direction. Here, the concentration distribution curves of various notes (quarter note, half note, dotted quarter note ♀''J, etc.) along the horizontal axis and the valley concentration along the vertical axis (April 111) are tth one-dimensional munching. It is registered in the dictionary in the memory of the dedicated process 6, and the data in this dictionary and Figure 6 (A), ((
1:) are compared with each density data of the notes determined as shown in FIG. 6(A) and (B), and it is determined which note the density data of FIGS. In the case of a three-way note, the striking part is detected by the professional sensor 5 dedicated to peak 9 note detection and processed by the one-dimensional matching professional sensor 6, but the note head part is clearly defined as density data. fl) cannot be recognized, and is processed by the next professional sensor 7 dedicated to two-dimensional matching.

In this way, the one-dimensional → touching processor 6 recognizes the notes and bar lines of the black circles in the record 1, and then starts similar processing for the record 2.

The two-dimensional matching professional sensor 7 extracts the peaks of the density distribution curve 12 of only the music information of the record 1 created by the gap-cutting professional sensor 4, and detects the portions that are not detected by the peak cutting I17 river professional sensor 5, that is, musical score symbols such as #, b, etc. To recognize musical notes and 74th notes that cannot be recognized by the one-dimensional matching dedicated prosensor 6.

First, in the case of musical score symbols such as # and B, and whole notes, there is a special process for 2 soft element munching, and S7 is a process for Ij-
An area 14 for extracting a target object (# in this example) for machining is determined from an image area 13 in which tJa etc. have been deleted, and the cutout image defined by the cutout range 14 is matched. Move to 15. Note that this matching area 15 is slightly larger than the cutout image. Then, a dictionary of matching rivers previously prepared in the memory of a separate two-dimensional munching dedicated process 7 is moved within the matching area 15 to perform matching between the two.

Here, the image (#) shown in 7tIjB diagram (A) is replaced with the symbol (#) shown in the same diagram (B) registered in the dictionary.
-), the difference in shape is noticeable at p and Q. As for the image, as shown in FIG. 9(A), the image part is set to "1",
The other areas are set to "0°°," and the image part shown in Figure 9 (B) is set to "l" in the dictionary, and all surrounding areas are set to 0°.
8 (C), and data where the image part is set to "1'' and the surrounding area is all set to "-1" are prepared. and Q in the same figure (B)
The parts are compared using logical operations. That is, p, , p shown in FIG. 9(A), Q+ +' Q3 shown in FIG. 9(8), and Q2. shown in FIG. 9(C). The degree of similarity is determined by calculating the logical product with Q4 for each pixel data and subtracting the product f〆i from the degree of similarity described later. That is, in both cases of FIG. 9 (A) and (B), since P1×QI+P2×Q3=0, there is no influence on the degree of similarity. On the other hand, in the case of Figure 9 (A) and (C), since P, XQ2+P, do. Therefore, in the above case, even if the image is larger than the dictionary, the similarity (- does not change and the difference in shape is not evaluated. On the other hand, in the latter case 't), if the image is larger, the similarity is 7 times larger than the dictionary. Since the value based on the difference is subtracted from the class (μ degree), the difference from the dictionary is emphasized and it is evaluated as -C.

Here, the image in FIG. 9(A) is designated as g, the dictionary consisting of the data entry ``'l'' and 0°'' as shown in FIG. 9(B) is designated as f, and the image in FIG. 9(C) is 1" and -1" as shown in
If r′ is a dictionary consisting of data of
There is a difference in shape between # and parrot. This allows it to be distinguished from #. On the other hand, in the f'Xg mashing method, # and # are sensitive to the deviation of images g, and the degree of similarity is low and is the same as that between # and -, making it difficult to distinguish between # and a parrot. Also, in l:I matching, fXg
In the matching of - and #, - becomes #, the similarity becomes high, and it becomes the same as the similarity between - and -, and # and -
I can't tell the difference. On the other hand, in the f'Xg matching method, the difference in shape between - and # is emphasized and the degree of similarity is reduced, so that they can be distinguished from #.

Next, we will explain how to calculate 91 for class 19 degrees. As shown in Fig. 10, if the black part of the target image g is "the same", the white part is "0°", and the total black ("1") is -1!, then
For the symbol #, fXg = b ...... Using the dictionary in (3), calculate the similarity S4g as S, g, = b/, f X 100 (%)...
・(4) Decide. For symbols other than symbol #, for example, musical score symbols such as -, use f"Xg=c (5), and calculate the similarity Srg by !X l 00 (%)...
...Old...It can be determined by the λ1 calculation (6). Note that since the above image and dictionary are matched for each constituent pixel, the above equations (3) and (5) can be used to convert pixel data into G(i, j'l and F(i, j)). Then, the formula becomes as follows.

fXg −ΣΣF(i, j) Fist G(i, j) ・・・・
・・(7) J f'Xg−Σ′j, F'(i, j)・G(i, j)・
Old... (8) J On the other hand, recognition of the three-sided f sign is performed as follows. In other words, if the note head is unclear and is recognized by the one-dimensional matching processor 6 as described above, there is a possibility that it is a half-note, so the two-dimensional matching processor 7 processes the note head as shown in FIG. Cutting areas are provided in areas 21 and 22 on the left and right of the driving needle.
By moving the white circle dictionary data downward from ``2'' every 1 and 22, the above-mentioned similarity is calculated. In this case, two dictionary data, 0 and 0, are used. Trigon notes, etc. are determined from the above similarity calculation, but 0
determines whether the matching is correct between the staves, and O determines whether the matching is correct on the staff. In this way, the two-dimensional matching dedicated processor 7 recognizes the musical score symbols and white circle notes of record l, and then
Start similar processing for .

The main processor 8 controls the processing timing of each dedicated processor, as shown in the overall flowchart of FIG. 12, so that each dedicated processor performs processing simultaneously and in parallel. However, since each dedicated processor uses the result of the dedicated processor before l processing, for example, if we pay attention to record H, as shown in the time chart in Figure i13, we can see (A) in the same figure showing the processing of processor 2 dedicated to creating the X-histogram. A series of processes from 1 to 2 (F) showing the process of the two-dimensional matching dedicated prosensor 7 are performed sequentially. Then, when a series of processing power is completed for the records from 10 onwards, the main processor 8 totals the results and completes musical score recognition.

As described above, according to the present invention, each process necessary for musical score recognition is processed in parallel by a dedicated processor, so that high-speed and efficient musical score recognition is possible.

In the above embodiment, each dedicated processor is assigned to create an X histogram, create a Y histogram, create a getter νJ, peak cut out, one-dimensional matching, and two-dimensional matching, but it is not necessarily limited to this. Rather, if the processing is divided so that the processing capacity of each dedicated processor is equalized, even more efficient and faster musical score recognition can be achieved.

[Brief explanation of the drawing]

Fig. 1 is a flowchart 1 of a general musical score recognition method, Fig. 2 is a diagram showing a schematic configuration of an embodiment of the present invention, and Fig. 3 (
A) is an example of a lou chord for a musical score, (B) is a diagram to explain the creation of an X histogram, (C) is a diagram to explain the creation of an X histogram, (D) and (E) )
Ha geta! , 7J, Figure 4 is a diagram to explain peak extraction, Figures 5 and 6 are diagrams to explain one-dimensional matching, and Figures 7 to 11 are diagrams to explain two-dimensional matching. FIG. 12 is a flowchart of the processing of the embodiment of the present invention, and FIG. 13 is a time chart 1 showing the timing of each processing of the embodiment. ■...Optical scanner, 2...X histogram dedicated processor, 3...Y histogram dedicated processor. 4...Process for digit cutting, 5...Pro sensor for peak reno output 111, 6...One-dimensional, cutting "
Pro sensor for γ, 7...Pro sensor dedicated to two-dimensional matching, 8...Main pro sensor, 9...Music score, 10
...X histogram, 11...Y histogram. Gei l Figure 2 Figure

Claims (1)

[Claims] Easy a! an optical scanner that reads density data for each Luco I·t;
X histogram dedicated processor that calculates the density distribution, and Y histogram i, l, that measures the second 6 degree distribution in the vertical axis direction.
the processor for j, and the first concentration 11j of jiii.
-Wi (Processor for 17 to calculate the third concentration distribution by subtracting fi, and a processor dedicated to peak extraction to detect the peak position of +tFj/)J, and a processor dedicated to this peak extraction. A one-dimensional Munsing i, l, f processor that recognizes gap marks and bar lines in black circles from near the peak (, l, position (=1) of the third density distribution extracted by νJ, and a processor for musical score symbols and white circles. A professional sensor dedicated to 2D Manshinog that recognizes the Noza mark,
111j A musical score recognition device comprising an optical scanner and a main processor that controls the timing of the entire valley-dedicated processor, and is characterized in that processing necessary for musical score recognition is shared among the dedicated processors and processed in parallel.