JPH0147829B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention is characterized by normalizing the sum of black bits for each character frame division unit area in each subpattern representing a stroke in a specific direction of an input character pattern by the character line width and the size of the character frame corresponding to each subpattern. The present invention relates to a character recognition method in which elements are extracted, a division format of a character frame division unit area is determined according to the complexity of an input character pattern, and characters are recognized by referring to a dictionary corresponding to the division format. As a form of character recognition method, a feature matrix is created with the quantities expressed by the above formula as feature elements F (k, k _i , k _j ), and the similarity with a standard character mask expressed in a similar format is calculated. There is a character recognition method that recognizes characters by measuring them. F (k, k _i , k _j )=B _k (k _i , k _j )/WL·WP _k WL represents the character line width of the input character pattern. WP _k represents WP _h , WP _v , WP _r , WP _l , etc.
For example, WP _h is the horizontal character frame size,
Similarly, WP _v , WP _r , and WP _l are the sizes of the character frame in the vertical direction, the 45-degree diagonal right direction, and the 45-degree diagonal left direction, respectively. B _k (k _i , k _j ) is B _h (h _i , h _j ), B _v (v _i ,
v _j ), B _r (r _i , r _j ), B _l (l _i , l _j ), etc. For example, B _h (h _i , h _j ) represents the divided unit area (h i , h _{i )} of the horizontal sub-pattern HSP. is the sum of black bits in h _j ),
h _i is the number of the divided unit area of the character frame in the X-axis direction, and h _j is the number of the divided unit area of the character frame in the Y-axis direction. Similarly, B _v (v _i , v _j ), B _r (r _i , r _j ), B _l
(l _i , l _j ) are the divided unit areas (v _i , v _j ), (r _i , r _j ), ( l _i
,
is the sum of black bits in l _j ). Also, v _i , r _i , l _i
is the number of the dividing unit area in the X-axis direction of the character frame of each sub-pattern, and v _j , r _j , l _j is the number of the dividing unit area in the Y-axis direction of the character frame of each sub-pattern. In this method, the positional information of character strokes is held in the divided unit areas, and the setting of the divided unit areas is one factor that influences the recognition rate. X-axis of character frame,
If divided unit areas are set by equally dividing each area in the Y-axis direction, the influence of stroke position fluctuations based on individual differences will be significant. Figure 1 shows two examples where the horizontal stroke position of the kanji ``eye'' is different.
ORGB is the input character pattern for the kanji "eye",
HSPA and HSPB are their horizontal subpatterns. In the example shown in Figure 1, if the Y-axis direction of the character frame is equally divided into four and divided unit areas h ₁ to _{h 4} are set, as shown by the dashed line, some black parts related to the horizontal sub-pattern Bit sum B _h (h _i , h ₃ ), B _h (h _i ,
_h4 ) is 1/ between both horizontal subpatterns HSPA and HSPB.
The character matrices of the two input character patterns ORGA and ORGB are also different, resulting in a lower recognition rate or the need for many dictionaries that take into account character deformation. Become. An object of the present invention is to reduce the influence of stroke position fluctuations based on individual differences by setting dividing coordinates to follow the stroke position fluctuations.
Find the black bit number distribution by projecting it onto each axis,
Find the barycenter coordinate series for each black bit number distribution, use this barycenter coordinate series as a division coordinate candidate, and determine the division coordinates by associating the division coordinate number and barycenter coordinate number almost equally, and divide the division coordinate series of these two systems. This was achieved by setting the dividing unit area of the character frame. FIG. 2 shows an embodiment of a character recognition device according to the present invention. A detailed explanation will be given below based on FIG. 2. 1 is optical input from the form. This optical input 1 is input to a photoelectric conversion section 2. The photoelectric conversion unit 2 decomposes one character planned area into a 128 x 128 image and converts each pixel into a binary digital signal (hereinafter referred to as input character pattern). One character is represented by an input character pattern of about 60 x 60 bits. The pattern register 3 stores the input character pattern in a format capable of reproducing the X and Y coordinates of each pixel in the expected character area, and has a capacity of 128×128 bits corresponding to the expected character area. The character line width calculation unit 4 has a shift register configuration similar to a well-known filter circuit, and receives an input character pattern to calculate an individual line Q in which all pixels in a 2×2 shift register window become black bits. and the black bit sum A of the input character pattern, and calculate the line width WL using the known line shown below. WL=A/A-Q The character frame detection unit 5 detects the circumscribed frame of a character by expressing it in the left end coordinate X _l , right end coordinate X _r , upper end coordinate Y _t , and lower end coordinate Y _b in the pattern register, and then detects the circumscribed frame of the character. It detects the size of the frame. The size of the character frame in the horizontal direction is WP _h = X _r −X _l
+1, as the vertical dimension WP _v = Y _t − Y _b −
1. The size in the 45-degree right diagonal direction and the 45-degree left diagonal direction is detected as WP _r =WP _l =WP _h +WP _v /2. The horizontal, vertical, right diagonal, and left diagonal subpattern extraction units 6 to 9 extract subpatterns HSP and VSP representing strokes in the horizontal, vertical, right diagonal, and left diagonal directions based on the input character pattern and line width WL. , RSP, and LSP, respectively, by extracting continuous black bits that are sufficiently longer than the line width corresponding to each direction. For example, the sub-pattern HSP is created by horizontally scanning the entire surface of the pattern register 3, detecting the number of consecutive black bits for each scanning line, and extracting the black bits whose length is greater than 2WL. Extract subpatterns. Similarly, vertical, diagonal right, and diagonal left subpatterns are extracted by scanning the pattern register 3 in the vertical direction, diagonal right direction, and diagonal left direction, respectively. An example of subpatterns extracted in this way is shown in Fig. 3. In the ORG in Fig. 3, the input character pattern HSP is a horizontal subpattern, VSR is a vertical subpattern, RSP is a right diagonal subpattern, and LSP is a horizontal subpattern.
is a left diagonal subpattern. Character projection creation section 1
0 is the input character pattern of the pattern register
The distribution of the number of black bits is obtained by projecting it onto the axis and the Y axis (horizontal and vertical directions in the two-dimensional coordinates of the pattern register, respectively), and the distribution of the number of black bits SX(x), SY
Create (y). SX (x) = _y=Yt 〓 ^y=Yb P (x, y) SY (y) = _x=Xr 〓 ^x=X1 P (x, y) However, x and y are each 0 to 127 in the pattern register 3 Y _t , Y _b are the upper and lower coordinates of the character frame in the Y-axis direction, X _l , X _r
are the leftmost coordinate and rightmost coordinate in the X-axis direction, and P(x,
y) means a black bit or a white bit; in the case of a black bit (significant color), P(x, y)=1, and in the case of a white bit (background color), P(x, y)=0. Black bit distribution SX for the two input character patterns ORGA and ORGB of the kanji ``eye'' shown in Figure 1
(x) and SY(y) are shown in Figure 4. Center of gravity detection unit 11
is the entire range X _l to X _r of the character frame in the X and Y axis directions,
The distribution of the number of black bits for each of the input character patterns is calculated for each range obtained by dividing the ranges X _l - _{X r} and Y _b - _{Y t} by Y _b - Y t and the centroid coordinates detected in the previous process.
Centroid coordinate series X (M _p ), Y of SX (x), SY (y)
(M _q ) is obtained by dividing the sum of the first moments of each range by the sum of black bits of that range. However, M _p and M _q are the barycenter coordinate numbers assigned in order of the size of the coordinate values,
M _p =1 to MX (MX is the number of centers of gravity in the X-axis direction and is an odd number) M _q =1 to MY (MY is the number of centers of gravity in the Y-axis direction and is an odd number). It is desirable to adopt a relatively large number (compared to the number of divisions) of about 15 as the number MX of the center of gravity coordinates in the X-axis direction.
To simplify the explanation _, we will describe the case of detecting seven _{barycenter coordinates} Black bit count distribution of input character pattern SX
By dividing the sum of the first moments of (x) by the sum of black bits in that range, we can calculate the central barycenter coordinate number.
Find the center of gravity coordinates X (M ₄ ) of M ₄ (see Figure 4). X(M ₄ )= _x=Xr 〓 ^x=Xl SX(x)・x／ _x=Xr 〓 ^x=Xl SX(x) Next, each range divided by the center of gravity coordinates X(M ₄ ), Two barycenter coordinates X( _M2 ) and X( _{M6 )} are obtained for _l ~X( _M4 ) and X(M4)~ _Xr . X(M ₂ )= _x=X(M4) 〓 ^x=Xl SX(x)・x／ _x=X(M4) 〓 ^x=Xl SX(x) X(M ₆ )= _x=Xr 〓 ^{x=( M4)} SX(x)・x/ _x=Xr 〓 ^x=(M4) SX(x) Next, the centroid coordinates detected so far X(M ₂ ),
Range X _l to X divided by X (M ₄ ), X (M ₄ )
( _M2 ), X( _M2 )~X( _M4 ), X( _M4 )~X( _M6 ),
Four centroid coordinates X for X (M ₆ ) ~ X _r
Find (M ₁ ), X (M ₃ ), X (M ₅ ), and X (M ₇ ). X(M ₁ )= _x=X(M2) 〓 ^x=Xl SX(x)・x／ _x=X(M2) 〓 ^x=Xl SX(x) X(M ₃ )= _x=X(M4) 〓 ^x=X(M2) SX(x)・x／ _x=X(M4) 〓 ^x=X(M2) SX(x) 〓 X(M ₇ )= _x=Xr 〓 ^x=X(M6) SX(x )・x／ _x=Xr 〓 ^x=X(M6) _SX (x) Detection of barycenter coordinates The centroid coordinates Y (M ₄ ) of the black bit distribution SY (y) of the input character pattern are detected in the range Y _b to Y _t , and then the range is divided into two by the centroid coordinates of the character frame.
The barycenter coordinates Y (M ₂ ), Y of the black bit distribution SY (y) for Y _b ~ Y (M ₄ ) and Y (M ₄ ) ~ _{Y t} , respectively.
(M ₆ ) and further divide the character frame in the Y-axis direction by the centroid coordinates Y (M ₂ ), Y (M ₄ ), and Y (M ₆ ) detected so far, each range Y _b ~ Y( _M2 ), Y( _M2 )
By detecting the barycenter coordinates of the black bit distribution SY(y) for ~Y( _M4 ), Y( _M4 )~Y( _M6 ), Y( _M6 )~ _Yt , a total of 7 Coordinates of the center of gravity Y (M ₁ )
~Y( _M7 ) is detected. The character frame dividing point determination unit 12 sets the number of divisions in the X and Y axis directions corresponding to each subpattern to NX and NY, and sets the division coordinate series in each direction of the X and Y axes corresponding to each subpattern to X _d (I), As Y _d (J), the barycenter coordinate series X (M _p ), Y (M _q ) in each direction of the X and Y axes are used as division coordinate candidates, and the barycenter coordinate numbers M _p and M _q are divided coordinate numbers I and J.
The divisional coordinates X _d (I) and Y _d (J) are determined in almost equal correspondence to . The division format of the division unit area in this example is all subpatterns HSP, VSP, LSP,
It is the same for RSP, and is divided into the same number of divided regions on both the X and Y axes, but the number of divisions in the X-axis direction can take four formats: NX = 4, 5, 6, and 8. , the division number NY in the Y-axis direction can take four formats: NY = 4, 5, 6, 8, and the division coordinate number in the X-axis direction is I.
(I = 1 to NX-1, NX = 4, 5, 6, 8), and the division coordinate number in the Y-axis direction is J (J = 1 to NY
-1, NY = 4, 5, 6, 8) and set the character frame to
The division coordinate series X _d (I) and Y _d (J) to be divided into NX and NY division unit areas are determined. The barycenter coordinate numbers M _p and M _q in each direction of the X and Y axes are almost equally matched with the barycenter coordinate numbers I and J in the X and Y axes directions to form a divided coordinate series X _d (I), Y _d (J). In order to determine this, a table as shown below is prepared, and the barycenter coordinate numbers M _p and M _q are read out from this table corresponding to the number of divisions NX and NY in each direction of the X and Y axes, and the barycenter coordinates are determined. Centroid coordinates M (M _p ), Y corresponding to numbers M _p and M _q
(M _q ) is determined as the division coordinates X _d (I) and Y _d (J).

[Table] This table is for the case where the number of barycenter coordinates MX and MY detected by the barycenter detection unit 11 is seven.
In the general case, the number of barycenter coordinates is set to be equal to the number of divisions in each of the X and Y directions, and if there are any extra barycenter coordinates remaining, the number of barycenter coordinates is divided in order from both ends. It can be created by making the correspondence so that many centroid coordinates are included. Figure 5 shows the number of divisions NX in each direction of the X and Y axes,
When the number of divisions NX=NY=5 is specified as NY, the correspondence between the division coordinate series X _d (I), Y _d (J) and the barycentric coordinate series X (M _p ), Y (M _q ) It also shows the divided unit areas k _i and k _j set by these divided coordinate series X _d (I) and Y _d (J). In addition, the number of divisions NX, NY detects the complexity of the input character, and determines the number of divisions NX, NY according to this complexity, or if it is rejected, change the number of divisions NX, NY and try again. Although character recognition is performed, detailed explanation thereof will be omitted. Here, it is assumed that the number of divisions for each sub-pattern is all determined to be 4×4 by the division number determining unit 8 for the input character pattern of the kanji character "me", and the division in the X-axis direction is performed accordingly. X as coordinate
(M ₂ ), X (M ₄ ), and X (M ₆ ) are divided coordinates in the Y-axis direction, and Y (M ₂ ), Y (M ₄ ), and Y (M ₆ ) are determined. The feature matrix extraction unit 13 extracts the coordinates of both ends in the X-axis direction and the division coordinates X _l , X (M ₂ ), Y (M ₃ ), X (M ₄ ), X _r in the Y-axis corresponding to the number of divisions of the sub-pattern Coordinates at both ends of the direction and division coordinates Y _b , Y (M ₂ ), Y
(M ₄ ), Y (M ₆ ), Y _t and each sub-pattern HSP,
In response to VSP, RSP, LSP, the size of the character frame WP _h , WP _v , WP _r , WP _l corresponding to each sub-pattern, and the character line width WL, each sub-pattern is determined from the division coordinates. The number of black bits for each divided unit area B _k (k _i , k _j )
, and calculate the line width W and the character frame size WP _h , WP _v , corresponding to each sub-pattern as shown in the formula below.
Normalized _by WP _r and WP _{l , the} feature element is calculated for each _divided unit area of _each sub-pattern _.
k=h, v, r, l) and create a feature matrix. F (h, h _i , h _j ) = B _h (h _i , h _j ) / WL・WP _h F (v, v _i , v _j ) = B _v (v _i , v _j ) / WL・WP _v F (r, r _i , r _j ) = B _r (r _i , r _j )/WL・WP _r F (l, l _i , l _j )=B _l (l _i , l _j )/WL・WP _l Calculation of the black bit sum B _k (k _i , k _j ) for each divided unit area (k _i , k _j ) of each sub-pattern is performed using the divided unit area
This is done by reading each sub-pattern at the dividing coordinates corresponding to k _i , k _j ) and adding up the number of black bits. For example, the black bit sum B _h (1, 1) corresponding to the feature element F (h, 1, 1) is the X coordinate X _l , X (M ₂ ) and Y corresponding to the divided unit area (1, 1).
It is obtained by reading out the horizontal sub-pattern HSP in the range determined by the coordinates Y _b and Y (M ₂ ) from the horizontal sub-pattern extraction section 6 and counting the number of black bits. The identification section 14 includes a standard character mask FS.
(k, k _i , k _j ) is the same as the feature matrix in the input character pattern, and the sum of black bits of each divided unit area k _i , k _j for each sub-pattern is calculated by the character line width and the character frame corresponding to the sub-pattern. It is expressed by the characteristics of standard characters normalized by size, and multiple dictionaries regarding the division format of the division unit area are prepared.
The degree of similarity between the input character pattern and the standard character mask is measured, and the character code of the most similar standard character mask is recognized as the input character pattern name, and the character code is output to the character code output terminal. The division format here is common to all subpatterns, and the same number of subpatterns in the X and Y directions is 4, 5, 6, or 8.
The area is divided into divided unit areas. As is clear from the above explanation, in this embodiment, when dividing the character frame in each direction of the X and Y axes into 4, 5, 6, 8, etc. division areas and setting division unit areas, the input character pattern Since it is based on the barycentric coordinate series X (M _p ), Y (M _q ) of the black bit number distribution in each direction of the X and Y axes, it follows stroke fluctuations due to individual differences and determines the dividing unit area. There are advantages. Referring to Figure 4, two input character patterns
There is a difference in the horizontal stroke position between ORGA and ORGB, but both input character patterns ORGA,
Since the black bit distribution SX(x) obtained by projecting ORGB onto the X-axis is the same, the division coordinate in the X-axis direction X _d (1),
X _d (2) and X _d (3) are also the same, and the division coordinates in the X-axis direction are not affected by the difference in horizontal stroke position. Divided coordinates in the Y-axis direction X _d (1), X _d (2), X _d
(3) is input character pattern ORGA, ORGB as Y
Since it is determined based on the barycentric coordinate series of the black bit number distribution SY (y) projected onto the axis, it moves according to the change in the position of the horizontal stroke in the Y-axis direction, and the division coordinates of both input character patterns ORGA and ORGB are Y _d (1)
~Y _d The relative positions of the vertical divided area determined by (3) and the horizontal stroke are approximate. In this way, when dividing coordinates are determined based on the barycentric coordinate series, there is an advantage that feature matrices can be extracted stably and variations in character patterns due to individual differences can be absorbed. Note that the subpattern division format in the above embodiment was such that both the X and Y axes were divided into the same number of subpatterns, and all subpatterns were divided in the same format. The present invention can be applied by storing barycenter coordinate numbers for each sub-pattern and for each X-axis and Y-axis in accordance with the format. Furthermore, in the above embodiment, the center of gravity coordinates and the divided coordinates were associated with each other by employing a table, but the correspondence can also be established by executing the arithmetic processing of the flowchart shown in FIG. can. Note that all the results of removal in FIG. 6 are rounded down to the nearest whole number. In FIG. 6, in step S1 (the number of centroids
MX+1) divided by the number of divisions NX to find the number M _a , and in steps S2 and S3, the remainder R ₁ of (MX+1)/NX is calculated.
and seek. In addition, k _a is determined in step S4, and division number k _i and center of gravity number M _p are set to 0 in steps S5 and S6.
Set to . Also, steps S7, S8, S9
Then, each time the division number k _i is increased by one, the previously set R ₂ is decreased by one, and the center of gravity number M _p is increased by M _a . In step S10, it is checked whether the remainder _R2 is non-negative, and unless the remainder _R2 is negative, the process proceeds to step S10.
11, the number of the center of gravity is increased by one, and the process proceeds to step S12.
Then, the centroid number M _p is associated with the division number k _i and the division coordinates X _d (k _i ) are determined. If the remainder R ₂ is negative,
In step S13, it is determined whether the current division number k _i is larger than k _a , and if it is larger, the center of gravity number is incremented by one, and if it is smaller, the center of gravity number set in step S9 is associated with the division number and divided. Coordinates X _d (k _i )
Then, in step S14, the division number k _i is (NX-
1) is detected and ends. As is clear from the above explanation, since the dividing coordinates of the character frame are determined based on the barycenter coordinates of the black bit number distribution in each direction of the X and Y axes of the character pattern, there may be movement of character strokes, etc. Even in this case, the position of the divided unit area containing the stroke is stable and character fluctuations can be absorbed, which has the advantage that the dictionary is simple and a high-speed and stable character recognition device can be realized.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an example of an input character pattern, FIG. 2 is a functional block diagram showing an embodiment of a character recognition device according to the present invention, FIG. 3 is a diagram showing an example of a sub-pattern in FIG. 2, and FIG. The figure shows the relationship between the input character pattern example and the barycenter coordinate series in Figure 2, Figure 5 shows the correspondence between the barycenter coordinate series and the divided coordinate series, and Figure 6 shows the relationship between the barycenter coordinate series and the divided coordinate series. It shows a flowchart for another determination method. 1... Optical input, 2... Photoelectric conversion section, 3... Pattern register, 4... Character line width calculation section, 5... Character frame detection section, 6... Horizontal sub-pattern extraction section, 7
... Vertical sub-pattern extraction section, 8 ... Right diagonal sub-pattern extraction section, 9 ... Left diagonal sub-pattern extraction section, 10 ... Character projection creation section, 11 ... Center of gravity detection section, 12 ... Character frame division point Determination unit, 13...Feature matrix extraction unit, 14...Identification unit, 15...Character code output end. SX(x), SY(y)...Black bit number distribution, _Xl , _Xr ...Character frame coordinates in the X-axis direction,
X _d (1) to X _d (4)...Divided coordinates in the X-axis direction,
(M ₁ ) to X (M ₇ )...Coordinates of the center of gravity in the X-axis direction, Y _t ,
Y _b ...Character frame coordinates in the Y-axis direction, Y _d (1) ~ _{Y d}
(4)...Divided coordinates in the Y-axis direction, Y (M ₁ ) ~ Y
(M ₇ )...Coordinates of the center of gravity in the Y-axis direction.

Claims (1)

[Claims] 1. Obtained by normalizing the number of black bits of an arbitrary divided unit area in a subpattern representing a stroke in a specific direction of an input character pattern by the character line width and the size of a character frame corresponding to the stroke direction. A feature matrix is created by extracting the feature elements from the input character pattern for each of a plurality of sub-patterns with different stroke directions and for each divided unit region obtained by dividing the character frame, using the quantity as a feature element. In a character recognition method that creates a standard character mask and recognizes an input character pattern by referring to a dictionary in which the standard character mask is described in the same format as the feature matrix, the input character pattern is projected onto the First, the barycentric coordinates of each black bit number distribution are determined for the range of the character frame, and then the character frame is divided at each barycenter coordinate detected so far. repeating the process of determining the barycenter coordinates of each of the black bit number distributions multiple times for each of the ranges in which the barycenter coordinates of the husband and husband are determined; One barycenter coordinate in the barycenter coordinate series is divided into one by almost equally corresponding the barycenter coordinate number in the series and each division coordinate number in the X-axis and Y-axis split coordinate series, which are assigned in the order of the size of the coordinate value. A character recognition method comprising the step of determining each division coordinate series in association with the coordinates, and setting an area obtained by dividing a character frame by two systems of the division coordinates as the division unit area.