JP3445039B2 - Music score recognition device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a musical score recognizing device, and more particularly to a musical score recognizing device capable of efficiently recognizing notes as constituent elements of a musical score by simple processing.

[0002]

2. Description of the Related Art In a conventional score recognition apparatus, for example, score image data read by a scanner is
There has been one that recognizes staffs, notes and various symbols and generates performance data such as MIDI file data. And
In the recognition processing of various symbols included in a musical score, for example, there has been a method of detecting a staff and erasing the detected staff to separate labels (images of various symbols).

[0003]

Usually, when a player reads a musical score, the pitch is judged by a black ball (notehead) written on the staff, and a vertical thin line extending from the notehead ( It seems that the thin and thick parts of the score are unknowingly separated, as the length of the note is judged by the number of thick flags (clams or continuous hooks) connected to the stem. Be done.

On the other hand, the biggest problem in the conventional musical score recognition apparatus is the separation of symbols. In the music score, symbols are drawn on the staff, and during the recognition processing, the labels of the respective music symbols are connected by the staff, which makes recognition difficult. In many score recognition devices, a process of erasing staffs is performed as a pre-recognition process, taking care to avoid unnecessary label separation as much as possible. However, even if such processing is performed, there are still many labels in which symbols are in contact with each other, such as temporary symbols in contact with additional lines, due to the nature of the musical score in which a large number of symbols are arranged so that they can be viewed in a musical manner.

Further, as a characteristic property of the musical score, there is a problem of a symbol whose shape is not constant. Although notes have regularity in their structure, their shapes as labels are infinite. Therefore, it is impossible to recognize them by a method such as matching with a dictionary for each label. Need to take. However, in the conventional erasure method, which is a symbol separation method, labels cannot be separated when they are in contact with each other, and it is difficult to recognize those whose shape changes like notes. There was a point. An object of the present invention is to solve the above-mentioned problems of the prior art, and to provide a musical score recognition apparatus capable of efficiently recognizing notes with simple processing.

[0006]

The present invention SUMMARY OF THE INVENTION, in part in the musical score recognition device to recognize the note from the input musical score image data, which constitute the thin line in accordance with a threshold corresponding to a musical score image data in a predetermined line width separating means for separating the image data, fine to detect the narrow end line on the basis of the image data of the portion constituting the thin lines separated by the separation means
The thin line detection means includes line detection means and thick symbol detection means included in the musical score image data for detecting a thick symbol based on image data other than the image data of the portion forming the thin line. Means and the bold symbol detection means
A musical score recognition device characterized by recognizing a note by using a detection result obtained by . With such a configuration, the present invention separates a thick portion, a thin vertical line, and a thin horizontal line in a musical score image as preprocessing for musical score recognition, so that it is sensuously similar to that when a human reads a musical score. A recognition method can be configured, the notes can be efficiently recognized from the separated images, and the recognition rate is improved .

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of the musical score recognition apparatus of the present invention. This device is a general computer system such as a personal computer to which a scanner and a MIDI interface circuit are added.
The CPU 1 is a central processing unit that controls the entire score recognition apparatus based on a program stored in the ROM 2 or the RAM 3. In addition, C at a predetermined cycle set in advance
It has a built-in timer circuit that interrupts PU1.
The RAM 3 is used as an image data buffer, a work area, etc. in addition to the program area. The hard disk device HDD4 and the floppy disk device FDD5 are
Stores programs, image data, performance data, etc. The CRT 6 displays the video information output from the CRT interface circuit 7 under the control of the CPU 1, and the information input from the keyboard 8 is taken into the CPU 1 via the keyboard interface circuit 9. The printer 10 prints the print information output from the printer interface circuit 11 under the control of the CPU 1.

The scanner 12 optically scans a (printed) musical score and converts it into binary or grayscale image data, and may be of any type such as a flatbed type, a handy type, and a feeder type. You can use one. The image information read by the scanner 12 is taken into the RAM 3 or the HDD 4 via the scanner interface circuit 13. The MIDI interface circuit 14 is a circuit that transmits and receives MIDI data to and from an external MIDI device such as a sound source module.
The bus 15 connects each circuit in the musical score recognition device. In addition to this, a pointing device such as a mouse, RS
A serial interface circuit such as 232C may be provided.

FIG. 3 is a flow chart showing the main processing of the CPU 1. In S1, the image of the musical score is taken into the RAM 3 by the scanner 12. The image is captured as a binary image. In S2, image quality smoothing processing such as figure fusion is performed in order to reduce blurring and dot noise. In S3, an image quality check process is performed. In the image quality check process, first, the line width of the staff and the space between the staffs are detected in order to obtain the magnification and the density information and the reference data for the staff detection in the subsequent stage. In order to obtain the line width and the space width, first, scanning is performed in the vertical (y) direction at several points in the horizontal (x) direction, and all the lengths of black runs (continuous black pixels) and white runs are calculated. Seeking
Frequency distribution (histogram) data is created for each length.

Since the most common symbol on the score is the staff,
By detecting the peaks of the created black run length histogram and white run length histogram respectively, the line width of the staff,
The space can be estimated. Then, the magnification of the image data can be estimated, for example, from the space, and the density can be estimated from the ratio of the line width and the space. In the musical score recognition processing, the recognition rate decreases if the magnification and the density deviate from the predetermined ranges. Therefore, in S3, these values are
It is checked whether it is within a predetermined range. S
In 4, it is determined whether or not the check result in S3 is the image quality OK, and if the result is not OK, S1
Return to and change the magnification and density to re-capture.

In S5, staff recognition is performed. The staff recognition processing is roughly divided into staff scanning start position detection processing and staff shift amount detection processing. The outline of the staff scanning start position detection processing will be described. At a certain position in the x-axis direction, the widths of the black pixels and the white pixels are sequentially obtained, and the positions where the obtained line width and the space width are arranged in a staff line are determined. , It is detected in consideration of some error. Then, in order to remove the influence of additional lines (horizontal lines added to describe notes that have run off from the staff), a condition is added that there is a white pixel width larger than the gap between both sides of the staff line arrangement. The midpoint of each black run at the x position where the arrangement of black and white pixels that meets this condition is set as the staff scanning start position.

Next, the outline of the staff shift detection processing will be described. The position is changed left and right one dot at a time from the staff scan start position (five black pixel positions) at the obtained x position. When the number of black pixels is less than a certain number (for example, 3 or 4) of the 5 points, the number of black pixels is checked by shifting the 5 points up and down, and the y coordinate is increased in the direction of increasing the ratio of black pixels. Shift. Then, the shift amount from the start position is set as the staff shift amount. The staff is detected by scanning from the staff scanning start position to the left and right until the number of black pixels becomes zero.

In S6, paragraph recognition processing is performed. This process is roughly divided into a paragraph recognition process and a bracket recognition process. In the paragraph recognition process, the staff is detected in the entire image, the set of staffs whose left ends are almost at the same position among the staffs is searched, and it is checked whether the ends of the staff are connected by black pixels. And recognize paragraphs. If the rectangles surrounding the paragraph are arranged side by side, this is also processed in chronological order. Note that the existence of a paragraph may be estimated by taking a histogram of black pixels in the x-axis and y-axis directions in advance and detecting a blank portion of the histogram. If the staves are connected by square brackets, there may be notes that cross the staves, so it is better to process the staves connected by the brackets in one unit. . In the bracket recognition, recognition is performed in a predetermined range to the left of the paragraph line by a method similar to the standard symbol recognition described later. In this recognition, it is only necessary to recognize the brackets and the large arch line.

In S7, the recognition result of the paragraph is displayed, and the user is checked whether or not the paragraph recognition result is correct, whereby it is determined whether the result is OK. If the result is not OK, the process proceeds to S8. After the transition, the paragraph recognition result is corrected. In the score staff, in addition to the parts having the same part structure in each paragraph, parts may be omitted or added in the middle, or a single staff and a grand staff may change from paragraph to paragraph in the same part. Correspondence of such parts is performed by comparing the brackets or the like, but since the correspondence of parts may not be uniquely determined, the paragraph recognition result can be corrected in advance. If the staff recognition fails, the subsequent processing cannot be performed, so it is necessary to change the magnification and the density and capture the image again. Therefore, in step S7,
First, display the recognition result of the staff, and let the user judge whether it is correct. If it is not correct, go back to S1 to re-import the image, and if the staff is recognized correctly, May display the paragraph recognition result for checking.

In step S9, processing rectangle determination processing is performed. A rectangle that is wide to some extent including the determined staff (in the case of a grand staff, the staff in the staff) is taken as the recognition processing rectangle. The size of the rectangle is larger than the maximum area where the musical symbols related to the staff exist, and the necessary symbols are not deleted by the staff inclination correction. The subsequent recognition is performed within this rectangle.

In S10, a staff slope correction process is performed. In brief, the pixel row is vertically shifted for each row of the rectangular image based on the staff shift amount obtained previously. It is more accurate to calculate the shift amount for each staff and perform shift correction in the rectangular image, but one shift amount may be calculated for the entire captured image and the entire image may be shifted. Thereafter, the graphic labels (independent black pixel areas) in contact with the upper and lower ends of the rectangle are deleted as constituent elements of the upper and lower parts. Finally, the upper and lower blank areas are detected to reduce the rectangle.

In S11 to S15, various symbols are recognized. There are roughly three types of score symbols in terms of shape and position. (1) A fixed type whose top and bottom positions are almost fixed (clefs, time signatures, etc.).
(2) A fixed type that has flexibility in the vertical position (temporary symbols, rests, etc.). (3) Those of indefinite type and indefinite position (notes, slurs, ties, etc.) Each of them is recognized in the order of clef, time signature, note recognition, fixed symbol recognition, character string recognition, slurs, and tie recognition.

The clef and time signature recognition are performed first because the recognition processing with low processing cost is performed first and the processing cost of subsequent recognition is reduced by deleting this symbol. This is because by recognizing such a thing, false recognition in later recognition is reduced. Moreover, the fixed symbol recognition is performed after the note recognition, which is a recognition method that is not easily affected by the contact of the label, and by deleting this note, it is possible to recognize the accidental symbols etc. that touch the note. This is because The slur and tie recognition is the last to reduce the number of labels subject to slur recognition, which requires high processing cost. Further, by making only the labels around the previously detected notes the target of slur and tie recognition, it is possible to further reduce the processing cost of slur and tie recognition, and to reduce erroneous recognition of slur and tie.

In S11, a clef and a time signature are recognized as symbols at fixed positions with respect to the staff. In the processing, first, a histogram of black pixels is taken vertically in a rectangular area including the determined staff, and a band in which the amount of black pixels is greater than or equal to a certain threshold is set in a place where a symbol may exist. As the target of matching. Matching is performed by peripheral peripheral features at several points between staffs. Peripheral feature is the first (first) distance from the left and right edges of a rectangular area that includes only the symbol to be matched to the white pixel area between the staff and the black pixel area. Alternatively, it is obtained up to several orders (from the second time onward). If matching fails, the adjacent bands are merged and recognized again. And
The recognized symbol is deleted from the image data.

At S12, note recognition described later is performed. In brief, first, image data is separated into vertical thin line, horizontal thin line, and thick line images. Then, the symbols and the parts of the symbols are detected from each image, and these are combined to recognize symbols such as musical notes. In S13, the fixed symbol recognition is performed.
In this process, first, a publicly known contour line load direction index is taken, the matching degree between the label size and the contour line load direction index is calculated for each symbol data in the dictionary, and the matching degree is normalized and integrated. Outputs the highest symbol. In addition to the size and load direction index, other features such as peripherals may be used. In addition, as a measure against the label cut by the staff deletion, the label cut by the staff deletion is registered in the dictionary, and when it is recognized as this symbol, the labels around it are combined and recognized again. . The recognized symbol is deleted from the image.

In S14, character string recognition is performed. In order to recognize character strings such as speed symbols, the alphabets and other symbols recognized by the fixed symbol recognition are used, and the rectangles enclosing the symbols are extracted in a character string form. By performing matching, a character string-like symbol can be recognized even if each constituent character is slightly wrong.

In step S15, slur tie recognition is performed. In this process, of the remaining labels, the labels around the detected note are thinned and subjected to multi-arc approximation. Then, since the line may be broken due to the previously erased symbol, the obtained multi-arcs are connected to each other. Finally, the slurs and ties are recognized based on the calculated arc shape, the thickness of the original image, and the relationship with the notes. Since this is the last routine in recognition, it is not necessary to delete the recognized symbols from the image, but if you delete the recognized slurs and ties and perform the fixed symbol recognition again after this, the slurs and ties will be deleted. You will be able to recognize the touched symbol.

In S16, for example, based on the recognition result, the musical score image data is combined and displayed, and the user is checked whether it is correct or not to determine whether it is OK. If the result is not OK, Shifts to S17, and the recognition result is manually corrected using a mouse, a keyboard, or the like. In S18, performance data creation processing is performed. In the processing, MIDI file data, which is a well-known performance data format, is generated based on the recognized various symbols and note information.

FIG. 4 is a flow chart showing details of the note recognition processing of S12 of FIG. In S20, a short horizontal black run is detected and the image is separated. The length of the black run is determined based on the line width of the previously detected staff, for example, twice the line width. FIG. 2 is an explanatory diagram showing an example of image separation processing. 2A is an original image, and FIG. 2B is an example of a vertical thin line image separated from the original image in S20 in S20. Bar line 20 and stem 2
1 is cut into a plurality of labels at, for example, staff positions. In addition, a thick image part such as a chained part of a stem disappears.

In S21, a vertical line is detected from the separated image data. FIG. 5 is a flowchart showing details of the vertical line detection processing in S21 of FIG. In S40, the vertical center line of each label (independent black area) is obtained from the horizontally short set of black runs. The center line may scan the image with a line that connects the midpoint of the horizontal run at the top of the label and the midpoint of the horizontal run at the bottom of the label, and determine the probability of black pixel existence. It is also possible to obtain the midpoint coordinate sequence from the horizontal black pixel sequence at the vertical position of, and obtain the midline by the least square method. In S41, the center lines are connected to each other. This determination is performed based on, for example, the inclination between line segments, the distance between the end points (in the x coordinate), the existence probability of one pixel in the original image of the line segment, and the like. In S42, the line is extended from the tip of the connected line to the outside until the black pixel does not exist in the original image. In S43, the lines whose x coordinates are close to each other are integrated and recognized as one line.

Conventionally, there has been a method of detecting a stem from a histogram in the vertical direction of an image. However, in such a method, the peak of the histogram becomes low when the inclination of the image is large, and the stem is accurately represented. Becomes difficult to detect. On the other hand, according to the stem detection method of the present invention, even if the stem has an inclination, it can be accurately detected. In addition, since the stems are accurately recognized, including the inclination, in this method, especially when the noteheads of many notes are close to each other, the stems and the noteheads can be more accurately associated with each other. . Moreover, since the method of connecting the median lines detected from the cut thin labels is adopted, it is possible to recognize even when the stem is slightly bent like a handwritten score.

Returning to FIG. 4, in S22, a short black run in the vertical direction is detected and the image is separated. The length of the black run is also determined based on the line width of the staff, but S2
It may be different from the value of 0. FIG. 2C is an example of the horizontal thin line image separated from the original image of FIG. The additional line 22 is cut into a plurality of labels at the notehead position. In S23, horizontal lines are detected in the same manner as vertical lines from the separated image data.

In step S24, the ellipse of the notehead candidate is detected from the image with the thin line erased. FIG. 2D is an example of the remaining thick line image obtained by separating the thin line from the original image of FIG. From this thick line image, noteheads (hereinafter referred to as black ball noteheads) of notes shorter than quarter notes, hooks, and continuous hooks can be separated. The black ball is recognized by extracting the boundary line coordinates at a certain interval from the boundary line chain of the thick label, calculating the elliptic formula by a known method, and taking the matching degree with this shape or image. In some cases, the elliptic expression may fail to be derived. Therefore, the elliptic expression is recalculated with different intervals, and the one with a high matching is selected.

As a conventional note head detection method, there is a method of holding coordinate information of a point where a black pixel exists in a note head area as a dictionary. In such a method, or even if the dictionary has other feature quantities, it is necessary to prepare a large dictionary to support various note fonts or feature quantities. It was necessary to recognize and build a dictionary. On the other hand, in the method of detecting the note head as an ellipse type, the shape of the ball is represented by the ellipse type. Therefore, depending on whether the shape of the ellipse type, that is, the coefficient of the equation is within a predetermined range. Note heads can be recognized, a large amount of dictionary is not required, and the elliptic coefficient range can be sensuously determined, so that it is not necessary to recognize a large number of note samples. Further, by expanding the elliptic coefficient range, it is possible to recognize handwritten musical scores. Furthermore, even when erasing notes, accurate erasure can be performed based on the shape determined by the elliptic formula.

In order to deal with chords, first, in order to recognize chords arranged in the horizontal direction, thick labels are cut by vertical lines of stem candidates. To deal with chords in the vertical direction, indentations in the thick label are detected, a pair of indentations on the left and right are made, and the thick label is cut by a line connecting the depressions. If the set of dents is not found, such as when the image is crushed, the position of the dents is estimated. FIG. 7 is an explanatory diagram showing an example of a chord note cutting process. FIG. 7A is an original image including chord notes, and FIG. 7B is an image in which thin lines are deleted. At this stage, the chord note heads are connected. (C) is the result of the cutting process, the left notehead is connected by the vertical line of the stem shown by the dotted line in (b), and the right notehead connects the left and right indentations shown by the arrow in (b). Each is cut by a line.

At S25, a note is detected by combining with the vertical line of the stem candidate previously obtained. For the connection, it is possible to use musical knowledge such as "a ball at the end of a stem has a fixed side to be connected". In S26, the 2nd note,
Detects the noteheads of whole notes (hereafter referred to as the hollow noteheads). The white notehead checks all the chains of the boundary lines of the original image, detects the white holes, and calculates the elliptic formula. Then, the degree of matching with the shape of the ellipse or the image is taken for recognition. For notes whose notes are on the staff, the elliptic formula is calculated from the combination of two chains. Instead of calculating and recognizing the elliptic formula, the bold label of the notehead portion may be simply held as a template in the dictionary. In this case, the bold label itself of the touched note head may be held in the dictionary instead of cutting the bold label in order to recognize the chordal note. Further, instead of template matching, recognition may be performed by feature extraction such as peripheral features. Even when using the elliptic formula, instead of calculating the elliptic formula from the chain of all the boundary lines, hold an elliptic formula calculated in advance with a notehead as a template, and normalize this with the interval between staffs. You may match with a thick label.

At S27, a note is detected by combining with the vertical line of the stem candidate previously obtained. In S28, a continuous hook is detected from the detected note and the thick line image. The continuous hook detects thick labels around the stems of the notes that do not consider the flag that has been obtained so far, and calculates the number of continuous hooks from the shape. In addition, other notes connected to this thick label are also detected. If there are no other notes to connect, it is considered to be a single flag or a half-hook. The note length information is determined by the number of hooks. It should be noted that, due to the intersection with the staff, there may be a case where several continuous hooks are one thick label, and after taking this into consideration, the number of continuous hooks is calculated as follows.

The maximum value and the minimum value in the y-axis direction of the thick label having a possibility of continuous hooking are obtained, and the standard deviation of this length is obtained.
If the standard deviation is less than a certain value, this thick label is considered to be a combination of one or several continuous hooks of the same length. After that, by measuring the vertical gap (gap) of the thick label, the number of continuous hooks is measured and reflected in the note length information. If the standard deviation is greater than a certain value, this thick label is considered to be a combination of hooks of different lengths.In this case, the stems of other notes connected to this label are , And calculate the gap. If a thick label existing within a certain distance from the end of the stem is detected and no other notes related to the label are present, it is considered as a note having a single flag.

In S29, the pitch of the note is detected. The pitch of a note can be determined from the coordinates of the center of the ellipse in the case of the notes between the staves, otherwise it may be obtained from the coordinates of the center of the ellipse. It may be different from the staff interval, so it is better to consider the additional line. The additional line is detected by horizontal line detection. However, in the additional lines of the four-notes on the line, the horizontal line may not be detected because the thin line portion is small. Therefore, in this case, it is possible to determine whether the line is on the line or between the lines by the existence probability of black pixels in a certain range at both ends of the ellipse. Also,
For a note between two staves on a grand staff, depending on whether there is an additional line at the expected position, the height of the note is based on the upper staff or the lower staff. It is necessary to determine whether it is the standard. In S30, the image of the note detected so far is deleted from the image. Recognized notes can be removed from the image along the shape of the detected note parts, such as ellipses and vertical and horizontal lines, to separate other labels that touch the recognized notes. S3
In 1, the repeated parentheses are recognized from the remaining horizontal and vertical lines and the image is erased. In S32, the crescendo, the decrescendo, etc. are recognized from the remaining horizontal lines and erased.

In step S33, the brackets of the tuplet are recognized and erased from the remaining horizontal and vertical lines. In S34, the bar line can be recognized from the remaining vertical lines. A thick vertical line of a double vertical line can be detected by performing vertical line detection with a thick label. The bar line is recognized based on the relationship with the staff and the positional relationship between the bar line candidates. The recognized symbol can be removed from the image along the shape of the recognized symbol to separate other labels that have come into contact with the recognized symbol. Thereafter, the undeleted image is subjected to staff deletion in the same manner as a general musical score recognition apparatus, and other symbols are recognized while extracting labels.

FIG. 6 is a flow chart showing the second embodiment of the thin line separation processing. The second embodiment adopts a method of separating a thick portion and a thin portion by image shrinkage and expansion. First, S50
In, the original image is saved. In S51, the process of reducing the periphery of the black pixel area dot by dot (degeneration process) is repeated n times to erase the thin portion. In S52, the process (expansion process) of adding black pixels to the periphery of the degenerated figure by one dot is repeated m times, and only the thick portion is returned to the original thickness. It should be noted that the number of shrinkage processes n and the number of expansion processes m do not have to be equal.

In S53, a thick line image is obtained by taking an AND (logical product) for each pixel (black is 1) of the original image stored in S50 and the processed image. S
At 54, a thin line image is obtained by subtracting a thick line image from the original image. For thin line images, vertical lines,
A horizontal line may be detected, or a vertical line and a horizontal line may be separated.

This method has a thick and thin separability better than the short run erasing method, especially in the oblique direction. However, the processing time will be somewhat longer. Further, by changing the number of connections (4 connections, 8 connections) at the time of shrinkage and expansion processing, it is possible to change the extraction method of the thick portion according to the shape of the target part. For example, it is possible to combine degeneration with 8 connections and expansion with 4 connections, or with degeneration with 8 connections and 4 connections alternately, and expansion with 8 connections. It is also possible to perform expansion. For example, in the case of 2-linked degeneracy, the pixel is set to black only when the pixel of interest and the three pixels above and below (or left and right) are black in the original image. It is also possible to perform thin line erasing only in the vertical or horizontal direction by this two-link process.

[0039]

As described above, according to the present invention, for example, as a pre-process of score recognition, by separating a thick portion, a thin vertical line and a thin horizontal line in a score image, the score can be easily processed. Can be separated into symbols close to human perception, and the following recognition processing can be performed simply and intuitively. Then, the notes can be efficiently recognized from the separated image, and the recognition rate is improved .

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a musical score recognition apparatus of the present invention.

FIG. 2 is an explanatory diagram showing an example of image separation processing.

FIG. 3 is a flowchart showing a main process of CPU 1.

FIG. 4 is a flowchart showing details of a note recognition process in S12.

FIG. 5 is a flowchart showing details of vertical line detection processing.

FIG. 6 is a flowchart showing a second embodiment of thin line separation processing.

FIG. 7 is an explanatory diagram showing an example of a chord note cutting process.

[Explanation of symbols]

1 ... CPU, 2 ... ROM, 3 ... RAM, 4 ... Hard disk device, 5 ... Floppy disk device, 6 ... CRT display device, 7 ... CRT interface circuit, 8 ...
Keyboard, 9 ... Keyboard interface circuit, 1
0 ... printer, 11 ... printer interface circuit,
12 ... Scanner, 13 ... Scanner interface circuit, 14 ... MIDI interface circuit, 15 ... Bus

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Atsushi Ohba 200 Terashima-cho, Hamamatsu City, Shizuoka Prefecture Kawai Musical Instrument Mfg. Co., Ltd. (56) Reference JP-A-1-304498 (JP, A) JP-A-6- 102868 (JP, A) (58) Fields surveyed (Int.Cl. ⁷ , DB name) G10G 3/00-3/04 G10H 1/00 G06K 9/00-9/03 Z G06K 9/46-9 / 52 G06K 9/62-9/82

Claims (6)

(57) [Claims] 1. A score recognizing device for recognizing a note from input score image data, a separating means for separating image data of a portion forming a thin line from the score image data according to a threshold value corresponding to a predetermined line width. When a fine line detection means for detecting the narrow end line on the basis of the image data of the portion constituting the thin lines separated by the separation means, there is included in the musical score image data, the thin line structure and a thick symbol detection means for detecting a thick symbols based on image data other than image data of a portion, detected by the thin line detection means and the thickness symbol detection means
A musical score recognition device characterized by recognizing a note using a result . 2. The separation means detects a pixel row having a length equal to or shorter than a predetermined threshold value in the vertical and horizontal directions, and separates the pixel row to separate image data of a portion forming a thin line. The musical score recognition apparatus according to claim 1, wherein 3. The image obtained by repeating the degeneracy a certain number of times and then repeating the expansion a certain number of times,
The musical score recognition apparatus according to claim 1, wherein the image data of a portion forming a thin line is separated by taking a difference between the original images. 4. The musical score recognition apparatus according to claim 3, wherein the separating unit changes the number of connected pixels at the time of the contraction / expansion processing to separate the image data of the portion forming the thin line. . Wherein said fine line detection means, by connecting a midline extending in the line width direction central Said sub had lines in the image data of the portion constituting the thin lines separated by the separating means, the fine have line 5. The musical score recognition apparatus according to claim 1, further comprising means for detecting the. 6. The bold symbol detecting means is included in the musical score image data, and an elliptic expression is obtained from boundary line coordinate information of an image figure in image data other than image data of a portion forming the thin line. Elliptic formula calculating means to be obtained,
5. The musical score recognition apparatus according to claim 1, further comprising notehead detecting means for detecting a notehead from the elliptical shape obtained by the means.