JPS6336037B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character recognition device having a function of making a determination using character information before and after a certain character or average character information of a line including a certain character.

When recognizing characters, for printed text, the input characters are quantized into a binary mesh pattern and each mesh is given an appropriate weight, or only the meshes necessary for recognition are selected. A commonly used method is to determine the category name by superimposing standard patterns and determining the degree of similarity or dissimilarity.

However, a difficult problem arises when the size and shape of printed characters cannot be specified. One of them is the issue of distinguishing between uppercase and lowercase letters.
(C, c), (O, o), (P, p), (S, s), (
U,
u), (V, v), (W, w), (X, x), (Z, z
)
An example is the distinction between Furthermore, when it comes to lowercase letters and numbers, the lowercase letter q in Q and the number 9 are similar in shape, so it may be difficult to distinguish them.

The purpose of the present invention is to refer to information such as the height and position of characters in predetermined categories of the preceding and following characters when recognizing a specific character included in a character string consisting of uppercase letters, lowercase letters, and numbers. The object of the present invention is to provide a character recognition method that enables determination based on the above.

The present invention provides a quantization separation means for quantizing characters in a form into binary quantization patterns and separating them into quantization patterns for each character, and an upper end position T _i and a lower end position B _i of the quantization pattern P _i . upper and lower end position detection means for detecting the upper and lower end positions; category determination means for determining a category name C _i by comparing the quantization pattern P _i with a standard pattern prepared in advance _;
The upper end position T _i and the lower end position B _i are set as a set,
Buffers to be stored in alphabetical order, average height α when the category name C _i belongs to a predetermined category set S _A , and average height α when the category name C _i belongs to a predetermined category set S _a . an average height detection means for calculating a height average β; and when a category name C _i in the buffer belongs to a predetermined category set S _P , the category C _i
The uppermost position T _o and the lowermost position of the category C _o belonging to the caligorie set S _B detected closest to
B _o , the category name C _i , the top end position T _i and the bottom end position B _i , and the heights α and β, using protrusion determining means for recognizing or changing the category name C _i ; This is a character recognition method characterized by being constructed by. The present invention also provides quantization separation means for quantizing characters in a form into binary quantization patterns and separating them into quantization patterns for each character;
Upper end position T _i and lower end position of the quantization pattern P _i
upper and lower end position detection means for detecting B _i ; category determining means for determining a category name C _i by comparing the quantization pattern _{P i} with a standard pattern prepared in advance; _i and the lower end position B _i are stored as a set in alphabetical order, the average height α when the category name C _i belongs to a predetermined category set S _A , and the category name C _i average height detection means for determining the average height _β when belonging to a predetermined category set S _{a ;} a size determining means for certifying or changing a category name C _i by comparing the position T _i with the lower end position B _i and the height average α and β; and a category name C _i in the buffer that is a predetermined category. A category C _o belonging to the category set S _B that is detected closest to the category C _i when belonging to the set S _P
the top end position T _o and the bottom end position B _o , and the category name C _i , the top end position T _i and the bottom end position B _i
and a prominence determining means for recognizing or changing the category name C _i using the height averages α and β.

FIG. 1 is a block diagram showing one embodiment of the character recognition system of the present invention.

91 in Figure 1 is a quantization method that scans the characters printed on a form, converts them into a binary pattern such as "1" for black parts and "0" for white parts of the paper, and separates each character one by one. It is a separate circuit. 92 is “1” at the top of the quantization pattern separated into individual characters.
The height position of is the top end T, and “1” is at the bottom.
This is a character uppermost and lowermost end detection circuit which detects the height position of B as the lowest end B and sets it in a part of the register 95. The coordinates are 0, 1, 2, starting from the top.
Suppose that 3... is given. This realization means is possible, for example, in Japanese Patent Publication No. 43-126. A register 93 superimposes each of the standard patterns stored in the standard pattern storage device with the quantized pattern to determine the degree of dissimilarity (or similarity), and stores the name of the standard pattern with the smallest degree of dissimilarity as the judgment category name in the register 95. This is a category judgment circuit that outputs This implementation means can also be achieved by, for example, Japanese Patent Laid-Open No. 156322/1983. In multiplexer 5,
While the category judgment circuit is operating, the signal line 90
The contents of the register 95 are passed through according to the signal on the register 95, and when the category judged by the category judgment circuit is set in the register 95, the register 95 is passed through at the address of the buffer 6 specified by the address counter 8. The contents of 95 are written by a write signal on signal line 905. Before one line on the form is scanned, the buffer 6 and address counter 8 receive a signal 901S (signal 90
1 shows the signal sent from 1. The signal on the signal line lmn is hereinafter referred to as lmnS), and the multiplexer 7 is controlled to pass the contents of the address counter 8. Every time one character is recognized, the contents of the address counter 8 are incremented by 1 by the signal 902S, and the top end, bottom end, and judgment category name of the input quantization pattern are
One line is sequentially stored at different addresses in the buffer 6.

FIG. 2 is a diagram specifically showing the operation of the present application. Figure 2 a shows Susan opened the box.
This is to show the upper and lower end positions of each character when the sentence "quickly" is quantized by the quantization separation circuit 91. 0, 1, 2,…20…40 from top
...It is divided into 100 meshes numbered 0 to 99, such as 60...80...99. For example, the first S is between 40 meshes with the top position T at 20 and the bottom position B at 60, and the character P in open is at the top position T at 40,
The lower end position B is 80.

Such a quantized input character string is detected by the upper and lower end position detection circuit 92 to determine the upper end position T _i of each character.
(i represents the i-th character), the lower end position B _i is determined, and the category determination circuit 93 recognizes and outputs the category C _i of each character. This category judgment circuit cannot distinguish between uppercase and lowercase letters when the categories are the same in shape but differ in size, such as S and s, or U and u, so uppercase letters are output as the judgment output. . FIG. 2b shows the contents of the register 6 corresponding to the input character string a. For simplicity, T _i and B _i were set to 15, 20, 40, 60, or 80. The q in "quick" is written with the lowercase alphabetic letter q and the number 9, which cannot be determined.

When buffer 6 stores data for one row, registers 12 and 13 are selected in multiplexer 10 by a signal on signal line 904, and address counter 8 is cleared by signal 901S. The search circuit 14 reads out the content (1 or 0) of the memory 141 specified by the category name part of the register 12, and when the content is "1",
The search signal “1” and the contents of the register 12, that is, the uppermost and lowermost positions, are input to the average value calculation circuit 16. In memory 141, uppercase letters
ABCDEFGHIJKLMNQRTY, lowercase
bdghklpqy, the content of the address corresponding to the number 1.2.3.4.5.6.7.8.9. is 1. Average value calculation circuit 1
6, when the judgment category is these category names (hereinafter referred to as category set S _A. In other words, this category set S _A includes characters with the largest vertical character width among alphanumeric characters). Add the heights of the quantization pattern (top and bottom tables). After reading out the contents of the buffer 6, the sum of the output signals "1" from the search circuit 14 is
Divide the sum of heights to find the average and set the category.
Let the height of the character S _A be α.

Among the categories C _i of register 6 in Figure 2b,
The registers 14 that output "1" are q, d, h, b, in order from the left as shown in FIG.
There are 7 characters: q, k, l, and y. Since all these heights are Bi-Ti=40, the average height α of the characters in the category set S _A is also 40.

register 13, search circuit 15, memory 151,
The average value calculation circuit 17 corresponds to the register 12, the search circuit 14, the memory 141, and the average value calculation circuit 16, respectively. However, the memory 151 stores lowercase letters a, e, m, n, r, (hereinafter referred to as category set S _a . In other words, this category set S _a contains letters with the smallest vertical character width among alphabetic characters. (includes)
Since the content of the address corresponding to is "1", the average value calculation circuit 17 calculates the category set.
We will find the height β of the character S _a .

Among the categories C _i of register 6 in Figure 2b,
As shown in FIG. 2d, the registers 15 output "1" from the left: a, n, e, n,
There are 6 characters: e and e. All these heights are
Since B _i −T _i =20, the average height β of the characters in the category set S _a is also 20. (uppercase P and lowercase p) or (lowercase q and number 9), etc., are almost the same in shape and size, but the bottom B of these letters is the same as the specific Letters from category sets, such as uppercase letters A, B, C, D, E,
F, G...Y, Z, lowercase letters abcdeh, i, k, l,
If one of the characters m, n, o, r, s, t, u, v, w, x, z does not appear below the bottom edge (the set of characters that meet this condition is defined as a category set S _B ), an uppercase P, and a lowercase P if it appears. Specifically, the address counter 8 is reset and starts reading from the buffer 6. control signal 90
4 controls multiplexer 10 to send the read information to register 30 in turn. The contents of the category section of the register 30 are registered in the memory 310 as a category set S _P such as the capital letter P and the number 9.
(A set of alphanumeric characters that have the same shape and size as other alphanumeric characters, but differ only in the position of the characters)
, the search circuit 31 outputs a signal “1” to the signal line 32. The signal line 32 is connected to the control section 9, and the control section 9 sets the contents of the address counter 8 to the address counters 50 and 60. The address counter 50 is
The address counter 60 is incremented by one by the signal 51S, and the address counter 60 is decremented by one by the signal 61S. Overflow of the address counters 50, 60 is signal 52
The control unit 9 checks through S and 62S. After the content of the address counter 50 increases by one, the multiplexer 7 sets the signal 53S to the address of the buffer 6, and the content of the address counter 60 increases by one.
After the signal 63S has been decreased by
is the address of the buffer 6, and the signals 53S and 63S are generated alternately. A control signal 903 is generated so that address counters 50 and 60 are selected alternately.

In multiplexer 10, the data read from buffer 6 is set in register 40 by control signal 904. The contents of the category section of the register 40 are registered in the memory 410.
The search circuit 41 checks whether it is in the category set S _B such as o, r, s, t, u, v, w, x, z, and if it is registered, the signal line is Outputs “1” to 42. Since the address counters 50 and 60 are selected alternately, the fact that "1" is output to the signal line 42 means that the memory 410
This indicates that among the characters of the categories registered in , the characters closest to the characters of the category name set in the register 30 have been retrieved. When “1” is output from the search circuit 41 to the signal line 42, the protrusion determination circuit 34 selects the register 40.
Determination is made using the lowest position B _o , the category name C _i of the register 30, the lowest position B _i , α indicating the size of uppercase letters, β indicating the size of specific lowercase letters, and constants such as 0.2, 0.3, 0.4, etc. do. Specifically, in the caligori name P, B _i −B _o ＞0.3α or B _i −B _o
If >0.4β then lowercase p Category name 9, B _i −B _o >0.3α or B _i −B _o
If >0.4β, lowercase q Category name P, B _i −B _o <0.2α or B _i −B _o
If <0.3β then capital P Category name 9, B _i −B _o <0.2α or B _i −B _o
If <0.3β, it can be realized by calculation like number 9. However, although the topmost position is not used here, it can be used if necessary.

In the case of FIG. 2, as shown in FIG. 2e, first, P of OPened causes the search circuit 31 to output "1". This “1” becomes the trigger, and the second
As shown in FIG. f, the next character e causes the search circuit 41 to output "1". The bottom position B _i of category P is
80, since the bottom position B _o of category e is 60,
In the salient circuit 34, B _i −B _o =80−60=20>0.3×40=12 Therefore, category P is a lowercase letter. next
I still don't know whether the q in quickly is the alphabetic letter q or the number 9, but this q or 9 pulls out "1" from the search circuit 31, and the next letter u pulls out the number 9 from the search circuit 41.
Pull out “1” from. The bottom position B _i of category q or 9 is 80, and the bottom position B _o of category u is 60.
Therefore, B _i −B _o = 80 − 60 = 20 > 0.3 × 40 = 12, and the category is lowercase q. In this way, the lowercase letters p and q are output from the protrusion determining circuit 34 as shown in FIG. 2f. Top position
T _i and the lower end position B _i are also output.

Through the above operations, it is determined whether it is an uppercase letter P, a lowercase letter p, the number 9, or a lowercase letter q, and the result is sent to the signal line 35 together with the top end T _{i and} the bottom end B _i . The control section 9 outputs this information to the signal line 906 and further writes it to the address of the buffer 6 specified by the address counter 8 via the multiplexer 8.

By the means described above, it is possible to determine the uppercase letter P and the lowercase letter P, as well as the alphabetical letter Q, the lowercase letter q, and the number 9, in multi-font type, which has been difficult in the past.

By the way, if input forms are simply to be sorted, such as the address of mail, there is no need to distinguish between uppercase and lowercase letters, regardless of the distinction between alphanumeric characters and numbers. However, when you send the read result to someone, it is necessary to distinguish between uppercase and lowercase letters. To do this, address counter 8 is reset, and data from buffer 6 is set in register 20 in multiplexer 10 by control signal 904S. In the memory 210, the category set S _c
As, capital letters C, O, S, U, V, W, X,
"1" is stored in the addresses corresponding to and Z, and in the search circuit 21, when the contents of the register 20 are included in these category sets _S , a signal “1” is sent to the signal line 22. Size judgment circuit 2
4, when the signal "1" appears on the signal line 22,
Category C _i of register 20 Top end T _i and bottom end
B _i , heights α and β of average value calculation circuits 16 and 17
is input, and it is determined whether the category C _i is an uppercase or lowercase letter based on these T _i , B _i , α, β, and other constants 0.8, 0.9, 1.2, and 1.3. Specifically, if |T _i −B _i |0.8α or |T _i −B _i |1.2β, change to lower case. If |T _i −B _i |0.9α or |T _i −B _i |1.3β, change to upper case. It makes a decision such as leaving it as is, and sends the confirmed or changed category name C _{i ,} top end position T _i , and bottom end position B _i to the control unit 9 via the signal line 26 . The control unit 9 sends this result to the signal line 906, passes it through the multiplexer 5 according to the control signal 907S, and writes the result to the address of the buffer 6 indicated by the address counter 8 according to the write signal on the signal line 905. , write. This operation is continued until the contents of the buffer 6 have all passed through the register 20. the result
Categories such as COSUVWXZ, in which uppercase and lowercase letters are the same, are input into the buffer 6 with a distinction between sizes.

In register 6 of FIG. 2b, S,
Although it remains unclear whether U, S, O, O, X, U, and C are uppercase or lowercase letters, they extract "1" from the search circuit 21 as shown in FIG. 2k. In the first category S, T _i =20 and B _i =60, but in the size determination circuit 24, |20−60|=40≧0.9α=0.9×40=36 Therefore, the uppercase letter S is output as is. All letters other than S are T _i = 40 and B _i = 60, but in the size judgment circuit 24, |40−60|=20≦0.8α=0.8×40=32 Therefore, all letters are changed to lower case and output. be done.

These results are shown as in Figure 2i.
Finally, the contents of the register 6 are changed as shown in FIG. 2j, and the recognition is completed.

By adding such means, it is possible to distinguish between uppercase and lowercase letters that have the same shape but differ only in size.

The character recognition method described above can be applied not only to alphanumeric characters, but also to distinguishing between katakana characters such as ``dakuten'' and ``ha'', and handakuten ゜ and ro.

In addition, by using similar means to recognize handwritten characters, it is possible to distinguish between C and c, W and w, etc., and between uppercase P and lowercase p, lowercase q of Q, and the number 9. It is possible.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the character recognition system of the present invention. FIG. 2 is a diagram for explaining the operation of the present invention. In the figure, 91 is a quantization separation device, 92 is a top end/bottom end detection circuit, 93 is a category determination circuit, 94 is a standard pattern memory, 5, 7, 10
is a multiplexer, 6 is a buffer, 8, 50, and 60 are address counters, 9 is a control unit, 1
2, 13, 20, 30 and 40 are registers, 1
4, 15, 21, 31 and 41 are search circuits, 14
1, 151, 210, 310 and 410 are memories, 16 and 17 are average value calculation circuits, 24 is a magnitude determination circuit, and 34 is a protrusion determination circuit.

Claims (1)

[Claims] 1. A character recognition device that recognizes alphanumeric characters, which quantizes characters in a form into a binary quantization pattern,
quantization separation means for separating each character into quantization patterns P _i ; upper and lower end position detection means for detecting the upper end position _{T i and} lower end position B _i of the quantization pattern P i;
Category determining means for determining _a category name C _i by comparing the quantized pattern _{P i} with a standard pattern prepared in advance _; , the buffer to be stored in alphabetical order, the average height α when the category name C _i belongs to the category set S _A , which is a set of alphanumeric characters with the largest vertical width, and the category name C _i average height detection means for calculating the average height β when belongs to a category set S _a which is a set of letters with a small vertical width among lowercase English letters _; When belonging to the category set S _P , which is a set of characters that have the same shape and size but differ only in position among numbers, the character that does not have a protrusion at the bottom among the alphabetic characters detected closest to the category C _i . The uppermost position T _o and the lowermost position B _o of the category C o belonging to the category set S _B which is a set of , the category name _{C i ,} the uppermost position T _i and the lowermost position B _i
and a prominence determining means for recognizing or changing the category name C _i using the height averages α and β. 2. Quantization separation means for quantizing characters in a form into binary quantization patterns and separating them into quantization patterns P _i for each character, and an upper end position T _i and a lower end position B _i of the quantization pattern P _i upper and lower end position detection means for detecting the upper and lower end positions, category determining means for determining the category name C _i by comparing the quantization pattern P _i with a standard pattern prepared in advance, and the category name C _i and the upper end position T _i and the lower end position B _i as a set, a buffer to be stored in alphabetical order, and the category name C _i
The average height α when belongs to the category set S _A , which is a set of characters with a large vertical width among alphanumeric characters, and the category name C _i is a character with a small vertical width among lowercase alphabetic characters. and an average height detection means for calculating the average height β when belonging to a category set S _a that is a set _of , the category name C _i is determined by comparison with the upper end position T _i , the lower end position B _i , and the height averages α and β _.
size determination means for recognizing or changing the size _of C _i
The top end _position T _o and the bottom end position B _o of the category C _o belonging to the category set S _B , which is a set of alphabetic characters detected closest to , the uppermost position
T _i , the lowest end position B _i , and the average height α and β
and a salient determination means for recognizing or changing the category name C _i using the above character recognition method.