JPH0242589A - Character pattern segmenting device - Google Patents

Abstract

PURPOSE:To improve character recognizing accuracy by segmenting a character pattern by means of the character shape evaluating value of the basic pattern or a coupled pattern of an objective character when the value is larger than a prescribed value, or with adding a character recognition evaluating value when it is smaller than the prescribed value. CONSTITUTION:The character string pattern obtained by a scanning means 1 is stored into a pattern memory means 2, the basic pattern and the coupled pattern are detected by a pattern area detecting means 3, and from this, a character shape evaluating value calculating means 4 calculates the character shape evaluating value. On the other hand, a character recognizing means 6 recognizes the basic pattern or the coupled pattern as the character pattern by a recognizing dictionary 5, and calculates the character recognition evaluating value. An evaluating value deciding/segmenting means 7 segments the character pattern when the character shape evaluating value is larger than a prescribed threshold. When it is smaller when the threshold, the character recognition evaluating value is obtained by the means 6, it is added to the character shape evaluating value, and the character pattern is segmented based on the value.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a character pattern cutting device that cuts out character patterns from character strings written in documents, and is particularly applicable to free-pitch character strings for which no character frame is specified. The present invention relates to a character pattern cutting device for cutting out a character pattern.

[Conventional technology]

Figure 8 shows, for example, the Japanese Patent Laid-Open No. 61-17 filed earlier by our company.
5878 (Japanese Patent Application No. 60-17265) is a block diagram showing the configuration of a conventional character pattern cutting device. In the figure, 1 is a scanning means for optically scanning and photoelectrically converting a character string written on a document, 2 is a character string pattern storage means for storing the pattern of the photoelectrically converted character string, and 3 is a character string for the above-mentioned characters. The basic pattern area is calculated from the coordinates of the left and right ends and top and bottom ends of the basic pattern obtained by dividing the character string pattern based on the peripheral distribution value of black points obtained by scanning the above character string pattern in a direction perpendicular to the column direction. 4 is a character shape evaluation value calculated from the coordinates of the left and right ends and the top and bottom ends of a combination of a character shape evaluation value of a single basic pattern region and a combination of a plurality of consecutive basic pattern regions; Calculation means, 9 is character shape evaluation value calculation means 4
Character cutting means 8 cuts out individual character patterns from the character string pattern based on the character shape evaluation value of the basic pattern area and the character shape evaluation value of the combined pattern area obtained in . This is an output means for outputting character patterns.

Next, the operation will be explained. A character string pattern in a document is scanned by a scanning means 1 and the obtained character string pattern is stored in a character string pattern storage means 2. The stored character string pattern is passed to basic pattern area detection means 3 to detect the basic pattern. Specifically, the character string pattern is scanned perpendicularly to the character string direction, the number of black dots is counted, and the marginal distribution value is determined. A basic pattern is obtained by finding a region whose marginal distribution value exceeds a predetermined threshold. FIG. 5 is an example of a character string pattern and marginal distribution values, and FIG. 6 is a basic pattern obtained from it. Next, by inputting the coordinates of the upper and lower ends and left and right ends of each basic pattern obtained by the basic pattern area detection means 3, the character shape evaluation value calculation means 4 calculates the shape of each basic pattern or the combination of a plurality of consecutive basic patterns. A character shape evaluation value is calculated, which is a value quantitatively indicating how much a pattern (hereinafter referred to as a combined pattern) resembles a single character. This is determined, for example, by a function that takes on a larger value as the rectangle circumscribing the pattern is closer to a square and the blank areas at both ends of the pattern are larger.

Next, the character cutting means 9 uses the character shape evaluation value calculating means 4.
Individual character patterns are cut out from the character string pattern based on the character shape evaluation values of the basic pattern and combined pattern determined by. More specifically, the character shape evaluation value of each pattern is multiplied by the number of basic patterns included in that pattern (hereinafter referred to as the weighted shape evaluation value), and each pattern is considered to be one character and combinations that can be extracted are calculated. Among them, the one with the largest sum of weighted shape evaluation values is selected and used as the cutout result. FIG. 7 shows examples of combinations that can be cut out for the basic pattern shown in FIG. 6. The pattern cut out by the character cutting means 9 is outputted from the output means 8 as one character pattern.

[Problem to be solved by the invention]

As mentioned above, conventional character pattern extraction devices determine characters only by the position and shape of the rectangle circumscribing the character pattern, so if adjacent characters are close together or in the case of kanji, it may be difficult to identify characters. There is a problem in that character patterns may be cut out incorrectly due to factors such as the distance between the characters and the characters, which reduces the accuracy of character recognition.

This invention was made to solve the above-mentioned problems, and the position and shape of each character is unstable due to the characters being close together or the kanji's bias and labor being far apart. It is possible to correctly cut out character patterns even from character strings written in free pitch, and the letter L! An object of the present invention is to obtain a character pattern cutting device capable of improving the recognition accuracy.

[Means to solve the problem]

The character pattern cutting device according to the present invention includes a recognition dictionary 5 storing standard patterns of characters to be recognized, and a character recognition dictionary 5 that recognizes a basic pattern or a combination pattern as a character pattern and outputs a character recognition evaluation value. When the character shape evaluation value is larger than a predetermined threshold value, the recognition means 6 cuts out a character pattern based on the character shape evaluation value, and when it is less than the threshold value, the character pattern is passed to the character recognition means 6. The present invention is characterized by comprising an evaluation value determination/cutting means 7 for cutting out a character pattern based on the value obtained by adding the recognized and outputted character recognition evaluation value to the character shape evaluation value.

[Effect]

The character recognition means 6 uses the recognition dictionary 5 to recognize the basic slam or combination pattern as a character pattern and outputs a character recognition evaluation value. Evaluation value determination/cutout means 7 cuts out a character pattern based on the character shape evaluation value when the character shape evaluation value is larger than a predetermined threshold. Furthermore, when the character shape evaluation value is less than the threshold value, the evaluation value determination/cutting means 7 passes the character pattern to the character recognition means 6 for recognition, and converts the resulting character recognition evaluation value into the character shape evaluation value. Cut out a character pattern based on the value obtained in addition to .

[Embodiments of the invention]

FIG. 1 is a block diagram showing the configuration of a character pattern cutting device according to an embodiment of the present invention.

In FIG. 1, components corresponding to those shown in FIG. 8 are given the same reference numerals, and their explanations will be omitted. In FIG. 1, 5 is a recognition dictionary storing standard patterns of characters to be recognized; 6 is character recognition means for recognizing basic patterns or combination patterns as character patterns using the recognition dictionary 5 and outputting character recognition evaluation values; 7; When the character shape evaluation value is larger than a predetermined threshold, a character pattern is cut out based on the character shape evaluation value, and when it is less than the threshold, the character pattern is passed to the character recognition means 6 to be recognized and output. This evaluation value judgment/cutting means cuts out a character pattern based on the value obtained by adding the character recognition evaluation value obtained by adding the character recognition evaluation value to the character shape evaluation value.

FIG. 2 shows a character string pattern and a basic pattern detected from the character pattern, and 10 in the figure shows an example of the character string pattern, and 11 shows an example of the detected basic pattern. FIG. 3 shows an example of the basic pattern 11 in FIG. 2, the character shape evaluation value of the combined pattern created by combining the basic patterns 11, and the character recognition evaluation value. In the figure, 12 is a character shape evaluation value, and 13 is a character recognition evaluation value. In this example, the combined pattern is a combination of three or less consecutive basic patterns.

Figures 4 (al) and (b) are examples of calculating the total evaluation value in this example. In the figure, 14 indicates the character string pattern 10 in Figure 2, consisting of 10 "sentence" + "character" + "word". “Ninja words”
In this example, 15 is the weighted shape evaluation value of each divided pattern, 16 is the sum of the shape evaluation values 15, and 17 is the character recognition evaluation value multiplied by the number of basic patterns. value (hereinafter referred to as weighted recognition evaluation value),
18 is the sum of the recognition evaluation values 17, and 19 is the total of the weighted shape evaluation value 15 and the weighted recognition evaluation value 17.

20 is the case where the pattern is divided into “sentence” + 6 characters” + “recognition” + “knowledge”, 21 is the weighted shape evaluation value of each pattern, 22 is the sum of the shape evaluation values 21, and 2
3 is the weighted recognition evaluation value, 24 is the sum of the recognition evaluation values 23, and 25 is the sum of the weighted shape evaluation value 21 and the weighted recognition evaluation value 23.

Next, the operation will be explained. A character string pattern 10 is obtained by scanning a character string "character recognition" as shown in FIG. 2, for example, by the scanning means 1. The character string pattern 10 is stored in the character string pattern storage means 2, and the basic pattern region S detection means 3
detects the basic pattern 11. Based on the coordinates of the upper and lower ends and left and right ends of each basic pattern 11 determined by the basic pattern area detection means 3, the character shape evaluation value calculation means 4 calculates the character shape evaluation value 12 of each basic pattern 11 and combined pattern. The evaluation value judgment/cutting means 7 calculates the weighted shape evaluation value 15.21 of each pattern 14.20 from the character shape evaluation value 12, examines the one with the maximum weighted shape evaluation value, and calculates the weighted shape evaluation value 15.21 as shown in FIG. Sentence “+1 character”+
Obtain the combination 14: “Koto + “Ninja” + “Ya”.

Next, the character shape evaluation value 12 of each pattern and the threshold value T
Compare with H. For example, the threshold value TH=0.65.

If the character shape evaluation values 12 of each pattern are all larger than 0.65, the patterns included in this combination 14 are cut out as character patterns, but in the example of FIG.
Since the character shape evaluation value 12 of the pattern "" is smaller than 0.65, the evaluation value judgment/cutting means 7 passes the basic pattern and the combined pattern to the character recognition means 6.The character recognition means 6 recognizes the passed pattern. A character recognition evaluation value ■3 is calculated from the standard pattern in the dictionary 5. Here, the character recognition evaluation value 13 is a value that quantitatively indicates how much a certain pattern is like a character. The extracted feature vector is expressed as P = (p I+ p
z + ... pfi), the standard feature vector of the character c1 stored in the recognition dictionary 5 is F(ci) =
(feil + 'Ci2+ H+ Hfci7
), it is given by the maximum value of S shown below.

(P, F (p= )) S= IPII - IP (c5) ■ Here, (P, F(Ci)) are vector P and vector F
(Ct), and [lPI is (P, P”
) is shown.

The evaluation value determination/cutting means 7 obtains a weighted recognition evaluation value from the character recognition evaluation value 13 calculated by the character recognition means 6, and adds it to the weighted shape evaluation value.

Then, find the combination that maximizes the sum, “sentence” +
Obtain 20 combinations of "character" + "recognition" + "knowledge".

The output means 8 extracts this combination 20 and outputs each character pattern as a result.

In fact, looking at Figure 4, the sum of weighted shape evaluation values is 16.2.
2 is 4.23 in combination 14, which is “bun” + “character” + “word” + “ningo” + “ya”, and “bun” + “character” +
In the combination 20 of “recognition” + “knowledge”, it is 3.9
8, and the sum of the weighted shape evaluation values, 16, is larger than the sum of the weighted shape evaluation values, 22. However, when weighted recognition evaluation values are added to these, the value of 7.32 for the former combination 14 becomes The value of the latter combination 20 is larger, 8.70, and is reversed.

As mentioned above, if the character shape evaluation value is only 12, the combination of “sentence” + “character” + “word” + “ninja word” + “fa” is 14.
Even if the character recognition evaluation value 13 is used in conjunction with the character recognition evaluation value 13, the correct combination 2o of "sentence" + "1 character" + "recognition" + "sense" will be selected.

According to the above embodiment, the evaluation value when cutting out a character pattern is not only the character shape evaluation value that indicates the position and shape of the pattern (by using the character recognition evaluation value that is pattern recognition information in conjunction with the character recognition evaluation value that indicates the position and shape of the pattern), Character patterns can be correctly extracted even from character strings that include many characters with uneven shapes.Also, when the character shape evaluation value takes a relatively large value, that is, the position and shape of each pattern is stable, and the character pattern If there is a high possibility that the patterns extracted based on the shape evaluation value are all one character, recognition is not performed and they are extracted as character patterns as they are, thereby minimizing the increase in processing time due to the addition of recognition processing. can be suppressed to

In the above embodiment, the combined pattern is a combination of up to three consecutive basic patterns as shown in FIG. 3, but it may be a combination of four or more basic patterns.

〔Effect of the invention〕

As described above, according to the present invention, there is provided a recognition dictionary storing standard patterns of characters to be recognized, a character recognition means for recognizing basic patterns and combination patterns as character patterns, and a character shape evaluation value that satisfies a predetermined threshold value. When it exceeds the threshold, it is cut out based on the value, and when it is below the threshold, the character recognition means is made to recognize the character pattern, and the character recognition evaluation value output as a result is added to the character shape evaluation value. Since the structure is equipped with an evaluation value judgment/cutting means for cutting out a character pattern based on the obtained value, the character recognition evaluation value can be used in cases where an incorrect character pattern is cut out using only the character shape evaluation value. This is used in conjunction with this method, allowing for correct character patterns even from character strings written with free pitch, where the positions and shapes of each character are unstable, such as when the characters are close together or the kanji are far apart from each other. Therefore, the effect of improving character recognition accuracy can be obtained.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing the configuration of a character pattern cutting device according to an embodiment of the present invention, Fig. 2 is an example of a character string pattern and a basic pattern, and Fig. 3 is a character shape of the basic pattern and a combined pattern. An example diagram of evaluation values and character recognition evaluation values, FIG. 4(a). (b) is an example diagram of the final evaluation value obtained for a specific combination of patterns, Figure 5 is an example diagram of character example patterns and marginal distribution values, Figure 6 is an example diagram of basic patterns,
FIG. 7 is an example of a combination of combined patterns created by combining the basic patterns shown in FIG. 6, and FIG. 8 is a block diagram showing the configuration of a conventional character pattern cutting device. 1... Scanning means, 2... Character string pattern storage means,
3... Basic pattern area detection means, 4... Character shape evaluation value calculation means, 5... Recognition dictionary, 6... Character recognition means, 7... Evaluation value determination/cutting means. Agent: Masuo Oiwa (and 2 others) 1st issue, 3rd item, 4th statement (spontaneous) 1. Indication of the case Patent Application No. 1983-193975 2. Name of the invention 3. Representative of the person making the amendment 4, 88 Figure 85 Figure 5 Detailed description of the invention subject to amendment, drawing column. α Contents of the amendment (1) On page 8, line 5 of the specification, the phrase "in Figure 1" is amended to read "in Figure 1." (Correct the phrase ``of each pattern 14.20'' in lines 15 and 16 of page 10 of the 21st edition to ``of each pattern.'' (3) Correct the drawings and Figure 3 as shown in the attached sheet. . that's all

Claims (1)

[Claims] a scanning means for optically scanning and photoelectrically converting a character string written on a document; a character string pattern storage means for storing a pattern of the photoelectrically converted character string; Basic pattern area detection means for detecting a basic pattern area from the coordinates of the left and right ends and top and bottom ends of the basic pattern obtained by dividing the character string pattern based on the peripheral distribution value of black points obtained by scanning the character string pattern. and the character shape evaluation value of a single basic pattern area and the character shape evaluation value of a combined pattern area that combines a plurality of consecutive basic pattern areas, at the left and right ends of the basic pattern area detected by the basic pattern area detection means. and character shape evaluation value calculation means for calculating from the coordinates of the upper and lower ends, the character shape evaluation value calculating means calculates each character from the character string pattern based on the character shape evaluation value of the basic pattern area and the character shape evaluation value of the combined pattern area. A character pattern cutting device for cutting out a pattern includes a recognition dictionary that stores standard patterns of characters to be recognized, and a character recognition means that uses the recognition dictionary to recognize a basic pattern or a combined pattern as a character pattern and outputs a character recognition evaluation value. When the character shape evaluation value is greater than a predetermined threshold, a character pattern is cut out based on the character shape evaluation value, and when it is less than the threshold, the character pattern is passed to the character recognition means to be recognized and output. 1. A character pattern cutting device comprising: evaluation value determination/cutting means for cutting out a character pattern based on the obtained character recognition evaluation value and character shape evaluation value.