JPH0550785B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character pattern cutting device that cuts out a character pattern from a character string written in a document.

[Conventional technology]

To recognize characters, the characters written on paper are photoelectrically converted and converted into binary signals of 1 and 0 corresponding to the characters and the background.The character pattern is then cut out one character at a time. There must be. For example, Fig. 2 is a diagram showing the configuration of a conventional device of this kind, which was shown in Proceedings of the 28th National Conference of the Information Processing Society of Japan, pages 885-886, "English Document Input System Using Character Recognition." In the figure, 1 is a sheet of paper, 2 is a scanning means for optically scanning and photoelectrically converting a character string written on the sheet 1, and 3 is a pattern of the photoelectrically converted character string (hereinafter referred to as a "character string pattern").
(4) character string pattern storage means for storing
is the left end of the pattern obtained by dividing the above character string pattern (hereinafter referred to as the "basic pattern") based on the continuity of the marginal distribution values obtained by scanning the character string pattern in the direction perpendicular to the above character string. A basic pattern area detecting means detects a basic pattern area by determining the coordinates with the right end; 5 is a basic pattern area detecting means 4 that detects a single basic pattern and a plurality of consecutive basic patterns from the positional information of the basic pattern area obtained by the basic pattern area detecting means 4; a character recognition means that calculates a value (hereinafter referred to as "similarity") that quantifies the degree to which each pattern that combines regions (hereinafter referred to as "combined pattern") resembles a pattern of one character;
Reference numeral 6 denotes a similarity table for storing the similarity obtained by the character recognition means 5; 9 character extraction position determining means determines the position of the character pattern to be extracted based on the degree of similarity stored in the similarity table; 10 is a cutting means for cutting out a character pattern one character at a time from the character string pattern storage means 3 based on the character pattern cutting position determined by the character cutting position determining means 9, and 11 is a cutting pattern cross for storing the cut out character pattern. It's Hua.

FIG. 3 is a diagram showing an example of processing by the basic pattern area detection means 4 of FIG. 2. In the figure,
12 is a character string pattern, 13 is a marginal distribution value, 14
is a basic pattern, and a basic pattern area corresponding to the basic pattern is shown surrounded by a rectangle.

FIG. 4 shows the similarity table 6 shown in FIG.
15 is a diagram showing an example of the configuration of
16 is a basic pattern or a combination pattern corresponding to the degree of similarity shown to make the explanation easier to understand.
Further, the symbol "*" in the figure means that there is no pattern corresponding to the predetermined area.

FIG. 5 is a diagram showing an example of a combination of basic patterns or combination patterns generated to cut out characters by the character cutting position determining means 9 of FIG. 2.

Next, the operation of the conventional device of this kind shown in FIG. 2 will be explained using FIGS. 2 to 5.

First, a character string on paper 1 is photoelectrically converted by scanning means 2 and stored in character string pattern storage means 3. Next, the character string pattern "solution" 12 in the character string pattern storage means 3 is passed to the basic pattern area detection means 4.

The basic pattern area detecting means 4 uses the character string pattern "resolution" 12 based on the continuity of the area where the marginal distribution value 13 created by scanning the character string pattern in the vertical direction orthogonal to the character string exceeds a predetermined threshold value. and each basic pattern "corner" obtained by dividing,
The coordinates of the left and right ends of "〓", "〓", "〓" 14 are detected as the basic pattern area.

Next, the character recognition means 5 uses already known character recognition technology to identify individual patterns for all patterns, including a single basic pattern and a combined pattern that combines a plurality of consecutive basic patterns. The degree of similarity, which is a value that quantifies how much is likely to be a pattern of one character, is calculated, and the calculated degree of similarity is stored in the above-mentioned similarity table 6. For example _{, let the} feature vector extracted from _the input pattern P be _X = [ _{X 1 ,} The similarity S(P) of the time input pattern is calculated from equations (1) and (2).

S(P)=m a CxS _c (X) ……(1) S _c (P)=(X, F _c )/｜X｜・｜F _c ｜……(2
) Here, the inner product (X, F _c )=X・F _c T and the norm |X|=√(,) are shown.

The similarity S(P) takes a value between 0.0 and 1.0, and has a property that the more the input pattern is a character, the larger the value is.

The results obtained by the character recognition means 5 are as follows:
It is stored in the similarity table 6 shown in FIG. That is, the degree of similarity is stored in an area designated by the number i when numbering the basic patterns sequentially from the left side and the number M of basic patterns forming a predetermined pattern. For example, the similarity of 0.2915 for the combined pattern "Solution" 16 is stored in the area indicated by the address of i=1 and M=3.

Then, the character cutting position determining means 9 sets the boundary points between the basic pattern areas as cutting candidate positions of the character string pattern, and uses all possible combinations of these cutting candidate positions to
Find all possible combinations of cutout patterns shown in the figure. Next, for each set of cropping candidate positions, a cropping evaluation value is calculated based on the similarity value stored in the similarity table above corresponding to the basic pattern or combined pattern corresponding to the pattern cropped at the cropping candidate position. , determines an optimal set of character extraction positions based on the magnitude of this extraction evaluation value. The cutout evaluation value can be specifically determined by a method of calculating the arithmetic mean of the degrees of similarity corresponding to each pattern.

Next, the character cutting means 10 cuts out a character pattern one character at a time from the character string pattern storage means 3 based on the information on the set of character cutting positions determined by the character cutting position determining means 9, and stores the character pattern in the cutting pattern buffer. Output to 11.

[Problem that the invention seeks to solve]

Conventional character pattern extraction devices are configured to extract character areas based on the size of the extraction evaluation value calculated based on the similarity value obtained by recognizing the pattern as described above, so the free pitch Although it is possible to extract individual characters from a string, it is possible to cut out individual characters from a string, but it is possible to cut out individual characters from a character string.
The pattern of ``moon'' and 旁 ``moon'' is one character each, ``day'',
In the case of a character that also exists as ``month'' (hereinafter referred to as ``separable significant character''), the similarity values corresponding to the patterns ``明'', ``日'', and ``月'' are approximately the same. There was a problem that the rate of erroneously extracting significant characters such as ``Ming'' and ``Ming'' was high.

This invention was made in order to solve these problems, and it is possible to extract individual characters from Japanese character strings written in free pitch containing separable significant characters consisting of glyphs in which ``bias'' and ``旁'' each exist as one character. The purpose of this invention is to obtain a device that can correctly cut out characters.

[Means for solving problems]

The character pattern cutting device according to the present invention includes:
A value obtained by dividing a character string pattern to calculate a quantified value (hereinafter referred to as "differential similarity") to what extent the pattern resembles a single character based on the difference in similarity values calculated by the above character recognition means. Find combinations of candidate positions, calculate the average value of the above-mentioned differential similarity of each pattern cut out at the cutout candidate positions for the combination of the found cutout candidate positions as a cutout evaluation value, and calculate the optimal character cutout based on this cutout evaluation value. A character cutout position determining means for determining a combination of positions is provided.

[Effect]

In this invention, one character is a pattern that is generated by combining a pattern of one character existing in a meaningful Japanese character string with a pattern of another character adjacent to this character or a part of another character. This can only happen with a probability close to 0, and on the other hand, there is a property that the partial or low patterns that make up a separate significant character always combine with other adjacent patterns to form a single character pattern. , the difference similarity calculation means quantitatively determines how much the pattern of interest is like a character based on the difference between the similarity value for the pattern of interest and the similarity value of the pattern that includes the pattern of interest. Calculate the value of the differential similarity expressed in . Therefore, although the value of the differential similarity with respect to the pattern of the separated significant characters calculated by the above-mentioned differential similarity calculation means does not decrease, the value of the differential similarity with respect to the partial or vertical patterns forming the separated significant characters becomes small. Therefore, by using the cutout evaluation value calculated based on the value of the difference similarity, it is possible to correctly cut out the significant separable characters such as "bright".

[Embodiments of the invention]

Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 1 is a diagram showing the configuration of an embodiment of the present invention. In the figure, numerals 1 to 6 and 9 to 11 are the same as the conventional device shown in FIG. In the figure, 7 is a differential similarity calculating means for calculating a differential similarity based on the similarity value stored in the similarity table, and 8 is a differential similarity calculating means for storing the differential similarity calculated by the differential similarity calculating means 7. This is a similarity table.

FIG. 6 is a diagram for explaining the operation of the present invention. In the figure, 17 is a character string pattern consisting of separable significant characters that are difficult to extract correctly with conventional devices, 18 is a marginal distribution value, and 19 is a This is a basic pattern.

FIG. 7 is a configuration example of the similarity table 6 shown in FIG. 1, in which 20 is an example of similarity;
21 is an example of a basic pattern or a combination pattern corresponding to the degree of similarity shown to make the explanation easier to understand. The symbol "*" in the figure means that, as in the case of FIG. 4, there is no pattern corresponding to the predetermined area.

FIG. 8 is a diagram showing an example of the configuration of the differential similarity table 8 shown in FIG. This is an example of a basic pattern or a combination pattern corresponding to degrees. In addition, the symbol "*" in the figure
means that there is no pattern corresponding to the predetermined area.

FIG. 9 is a diagram showing an example of the configuration of the differential similarity calculation means 7 shown in FIG. 1, and in the figure, 6 and 8 are the same as those shown in FIG. 1. In the figure,
24 is a pattern similarity detector, 25 is an inclusive pattern similarity detector, 26 is a maximum element detector, 27 is a subtractor, and 28 to 34 are signal lines.

The operation of the embodiment shown in FIG. 1 will be explained below using FIGS. 6 to 9.

First, the character string written on the paper 1 is photoelectrically converted by the scanning means 2, and the resulting character string pattern "Meiji" 17 for one line is stored in the character string pattern storage means 3. Next, the character string pattern "Meiji" 17 is passed to the basic pattern area detection means 4. The basic pattern area detection means 4 calculates the marginal distribution value 18 of the character string pattern "Meiji" 17,
Next, the basic patterns "day", "month", "〓", and "stand" 19 are extracted based on this peripheral distribution value, and the coordinates of the left and right ends corresponding to each basic pattern are used as the position information of the basic pattern area as described above. It is transferred to the character recognition means 5.

In addition, the character recognition means 5 performs pattern recognition processing on all patterns, including a single basic pattern and a combined pattern that combines a plurality of consecutive basic patterns, as in the conventional device, and The degree of similarity corresponding to the pattern is stored in the degree of similarity table 6 shown in FIG.

Next, the operation of the differential similarity calculating means 7 will be explained. First, differential similarity will be briefly explained.

In general, a pattern P _A created by combining a pattern P ₀ of one character existing in a meaningful string of Japanese sentences with a pattern P ₁ of adjacent other characters or a part of other characters is one It is known that the probability of a character being a character is only close to 0. Therefore, the value of similarity S (P _{0 ) obtained by recognizing the pattern P 0 of} one character and the value of similarity S (P _A ) obtained by recognizing the pattern P _A including pattern P ₀ are large. It makes a difference. On the other hand, both the partial or vertical pattern P ₂ constituting the separate significant character and the pattern P ₃ of the separate significant character become one character. Therefore, the value of similarity S (P ₂ ) obtained by recognizing the partial or partial pattern P ₂ that constitutes the separated significant character and the pattern of the separated significant character that includes the pattern P ₂
There is no significant difference in the value of similarity S(P ₃ ) obtained by recognizing P ₃ . Therefore, it is possible to accurately express how likely the pattern P is to be a single character, depending on the value of the differential similarity ΔS(P) of the pattern P defined by equation (3).

ΔS (P) = S (P) - maixS (Pi) ... (3) Here, Pi is a pattern that includes pattern P. Figure 8 shows the similarity values stored in the similarity table shown in Figure 7. This is an example of a differential similarity table in which the differential similarity calculation means 7 stores differential similarities corresponding to each of the basic pattern and the combined pattern based on the above-mentioned differential similarity calculation means 7. For example, the patterns "Moon", "Ming",
The similarity to “Moon〓” is 0.89, 0.90, respectively.
Since it is 0.71, the differential similarity ΔS (month) corresponding to the pattern “month” = 0.89 − max (0.90, 0.71) = −0.01
becomes.

FIG. 9 shows an example of a configuration for calculating differential similarity. First, pattern index signals i, M specifying the pattern for calculating the differential similarity.
is connected to the pattern similarity detector 24 via a signal line 28.
The pattern similarity detector 24 loads the similarity corresponding to the pattern indicated by the pattern index signals i and M from the similarity table via the signal line 29, and then subtracts it via the signal line 30. transfer to the device 27. Next, the inclusive pattern similarity detector 25 detects pattern index signals i, M
are input through the signal line 28 to generate index signals i', M' of patterns including the patterns specified by the pattern index signals i, M. Then, based on the pattern index signals i' and M', the similarity is loaded from the similarity table 6 via the signal line 31 and transferred to the maximum element detector 26 via the signal line 32. Maximum element detector 2 above
6 detects the maximum value among the similarities sent from the inclusive pattern similarity detector 25 and transfers it to the subtracter 27 through the signal line 33. The subtracter 27 calculates the difference between the similarity transferred via the signal line 30 and the similarity transferred via the signal line 33, and sends the result to pattern index signals i, M via the signal line 34. The difference similarity table 8 is stored in the area indicated by .

Next, a process of cutting out characters from a character string pattern using the differential similarities stored in the differential similarity table 8 will be described.

The character cutout position determining means 9 uses the boundary points between the basic pattern regions as cutout candidate positions of the character string pattern, and finds all combinations of these cutout candidate positions. Next, for each combination of cropping candidate positions, a cropping evaluation value is calculated based on the value of the differential similarity stored in the differential similarity table 8 corresponding to the pattern cropped at the cropping candidate position, and the highest cropping evaluation value is determined. A combination of given cutting patterns is detected and this information is transferred to the character cutting means 10.

The cutout evaluation value can be specifically determined by calculating the differential similarity corresponding to each pattern, and using the results, for example, as an arithmetic average value, as in conventional devices. For example, for the character string pattern shown in Figure 6, if the cutout evaluation value V is calculated by the arithmetic average of the differential similarities stored in the differential similarity table shown in Figure 8, then + “〓” + “stand”) = (0.01−0.01
−0.02
+0.02)/4=0.000 V ("Sun" + "Month" + "Cure") = (0.01-0.01+0.24)
/3
= 0.080 V ("Sun" + "Month" + "Unit") = (0.01 + 0.01 + 0.02
)/
3 = 0.013 V (“bright” + “〓” + “base”) = (0.20−0.02+0.02)
/3
= 0.007 V ("Ming" + "Chi") = (0.20 + 0.24) / 2 = 0.220 V ("Ming" + "Tai") = (0.04 + 0.02) / 2 = 0.030 V ("Sun" + "month = unit") = (0.01 + 0.02) / 2 = 0.01
It becomes 5. As a result, the character cutout position determining means 9 determines that the cutout evaluation value V (“bright” + “ji”) of the combination of cutout patterns “bright” and “ji” =
Since 0.220 is the maximum value, the patterns "Ai" and "ji" are determined to be the optimal cutting results, and information on this combination is transferred to the character cutting means 10.

Finally, the character cutting means 10 cuts out the character pattern one character at a time from the character string pattern storage means 3 based on the information on the set of character cutting positions determined by the character cutting position determining means 9, and stores the character pattern in the cutting pattern buffer. Output to.

In addition, although the above embodiment describes the case where the degree of similarity is calculated by the character recognition means, the present invention is not limited to this, and may be used when the degree of dissimilarity is calculated by the character recognition means. This can be achieved by replacing the circuit that detects the maximum value with a circuit that detects the minimum value.

〔Effect of the invention〕

As described above, according to the present invention, characters are extracted one by one based on the value of the difference in similarity obtained by recognizing a pattern. The present invention has the effect of providing a device that can cut out individual characters with high accuracy from a character string written in free pitches including separable significant characters.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention, FIG. 2 is a block diagram of a conventional character pattern extraction device, FIG. 3 is an explanatory diagram of a processing example of the basic pattern area detection means, and FIG. 4 FIG. 5 is an explanatory diagram showing an example of a combination of patterns to be evaluated by the character extraction position determining means;
FIG. 6 is a diagram showing an example of a separation significant character string pattern,
FIG. 7 is a diagram showing an example of the configuration of the similarity table, FIG. 8 is a diagram showing an example of the configuration of the differential similarity table 8,
FIG. 9 is a diagram showing an example of the configuration of the differential similarity calculating means. In the figure, 1 is paper, 2 is a scanning means, 3 is a character string pattern storage means, 4 is a basic pattern area detection means, 5 is a character recognition means, 6 is a similarity table, 7
8 is a differential similarity calculation means, 8 is a differential similarity table, 9 is a character cutting position determining means, 10 is a character cutting means, and 11 is a cutting pattern buffer. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] 1. A character pattern cutting device that cuts out a character pattern from a character string written on a sheet of paper, etc., includes a scanning means that optically scans and photoelectrically converts the character string on the sheet, and stores the photoelectrically converted pattern of the character string. a string pattern storage means for
Basic pattern area detection means that divides the character string pattern based on the continuity of the marginal distribution values obtained by scanning the character string pattern, and determines a basic pattern area from the coordinates of the ends of the basic pattern obtained by dividing;
Character recognition means that recognizes the basic pattern and a combined pattern that combines a plurality of consecutive basic patterns and calculates the similarity of the basic pattern and the combined pattern, and the above-mentioned similarity of the focused basic pattern itself or the combined pattern itself. and a difference similarity calculating means for calculating the difference between the basic pattern to be focused on or the combined pattern including the combined pattern as the differential similarity of the basic pattern to be focused on or the combined pattern to be focused on, based on the position information of the basic pattern area. to find possible combinations of cropping candidate positions,
The average value of the above-mentioned differential similarity of each pattern cut out at the cutout candidate positions for the obtained combination of cutout candidate positions is determined as a cutout evaluation value, and the optimal combination of character cutout positions is determined based on this cutout evaluation value. A character characterized by comprising: a cutting position determining means; and a character cutting means for cutting out and outputting a character pattern from a character string pattern stored in the character string pattern storage means based on the result of the character cutting position determining means. Pattern cutting device.