JPS62271088A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To obtain a device which can recognize a continuous character, by considering enough a writing order variation at the time of generating a pattern, and thereafter, resolving the pattern into a fundamental stroke, and comparing it with a stroke of a pattern which is registered in advance. CONSTITUTION:When a writer writes a character to a tablet 1 with an input pen 2, a coordinate of the pen and up-down information of the pen are sent to a feature extracting part 3, and in the feature extracting part 3, from an inputted character pattern of one character portion, a feature is extracted and sent to a character discriminating part 4. In the character discriminating part 4, the feature which is extracted from the inputted character pattern is compared with a character pattern of each character which has been registered in advance in a dictionary part 5 and its similarity is derived, and the character having the highest similarity is outputted as a result of recognition. In such a case, the inputted pattern is resolved into a fundamental stroke, the resolved stroke information is compared with stroke information containing an end point coordinate of a stroke which is registered in advance, and based on a result of comparison, the pattern is recognized.

Description

[Detailed Description of the Invention] 3. Detailed Description of the Invention [Industrial Application] The present invention is a pattern recognition device, in particular, a pattern recognition device that recognizes characters, symbols, and shapes based on handwriting information such as handwritten characters, symbols, and figures. The present invention relates to a pattern recognition device for recognizing figures, etc.

[Conventional technology] Conventionally, online pattern recognition, which recognizes the pattern of characters written on the data tablet by inputting pen coordinates and pen up/down information from a data tablet, has strict stroke order restrictions and is prone to errors. Characters in stroke order were often misrecognized or rejected. Additionally, online pattern recognition devices that allow for free stroke order have been attempted, but the recognition time is longer compared to when the stroke order is restricted, making it difficult to expand the types of patterns to be recognized. This has the disadvantage that the rate of misrecognition increases.

In terms of recognizing incoming cover patterns, one of the major features of online pattern recognition, which is recognized by inputting pen coordinates and pen up-down information from a data tablet or the like, is that stroke order information can be used. The stroke order differs depending on the person inputting the pattern, and is not necessarily the same, but if by using stroke order information, character recognition can be achieved with a high recognition rate and a short recognition time, then stroke order information can also be used to a great extent. It's Bemi.

[Problems to be Solved by the Invention] The present invention fully takes into account stroke order variations during pattern creation, and then decomposes the pattern into basic strokes and compares them with strokes of patterns registered in advance. The present invention provides a pattern recognition device that has a high recognition rate for written characters and a short recognition time.

[Means for Solving the Problem] As a means for solving this problem, the pattern recognition device of the present invention decomposes an input pattern into basic strokes and uses the stroke information obtained by decomposition and the The apparatus includes a comparison means for comparing stroke information including stroke end point coordinates registered in advance, and a recognition means for recognizing the input pattern based on a comparison result of the comparison means.

[Operation] In this configuration, the input pattern is decomposed into basic strokes, and the pattern is recognized by comparison with stroke information registered in advance.

(The following is a margin) [Example] Hereinafter, an example of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1(a) is a block diagram of the pattern recognition device of the first embodiment, in which 1 is a tablet, 2 is an input pen, 3 is a feature extraction section, 4 is a character discrimination section, and 5 is a dictionary section. When a scribe writes a character on the tablet 1 with the input pen 2, the coordinates of the pen and the up-down information of the pen are sent to the feature extraction unit 3, which extracts features from the input character pattern for one character. The zero character discriminator 4 extracts the features extracted from the input character pattern and sends them to the character discriminator 4.
The degree of similarity is determined by comparing the character patterns of each character registered in advance with the character pattern, and the character with the highest degree of similarity (or the lowest degree of difference) is output as a recognition result. Note that the character pattern also includes symbols, figures, and the like. FIG. 1(b) is a block diagram of the character discrimination unit 4, in which 41 is a CPU for processing, 42
is a ROM that stores the program of the processing procedure of the CPU 41
, 43 are auxiliary RAMs.

FIG. 2 is a diagram showing the configuration of the dictionary section 5 of FIG. 1(a), and in this example, the dictionary section 5 is registered in the order of strokes. 51 is a diagram showing the structure of a character block of nine stroke characters, and 52 is a diagram in which one character block has a character code 53, division information 54,
5 is a diagram showing that it is composed of stroke information 55. FIG.

FIG. 3(a) is an example of one character block 52, and "
It shows the structure of the character block for `` Thoughts''. r3b57J
is the character code 53 for "shi", and "5" is the division information 54 indicating that "shi" is divided into the first five strokes and the last four strokes, and is processed in the next nine blocks. The stroke information 55 for "thought" is registered according to the stroke order of the measure.

FIG. 3(b) shows a character block having division information 54 which additionally has "1" as the number of divisions.

The operation of this embodiment will be explained below.

Figure 4 is a diagram showing an example of how to write ``r thought'', and ``residence'' is written as ``residence''.
There is a sudden change in the stroke order of the part ``田'', and although the stroke order shown in 40 is the basic stroke order, it is often written in different stroke orders such as the stroke order shown in 41, 42, and 43. However, when writing ``heart''``Ten'' is written down after ``Ten'', or ``Kokoro'' is written in the middle of ``Ten'', and the rest of ``Ten'' is not written in the middle.
In most cases, ``kokoro'' (heart) is written after writing down ``田''. In order to recognize the ``thoughts'' of various stroke orders in this way,
There are two possible ways to do this: register ``shi'' in various stroke orders in the dictionary section 5 while restricting the stroke order, or set the stroke order freely and compare the entered cover patterns with all patterns of ``shi''. However, in the former case, the capacity of the dictionary section 5 becomes extremely large, which is not practical, and in the latter case, unnecessary comparisons are made, which increases the recognition time and may even lead to incorrect responses. .

FIG. 5(a) is a diagram showing a comparison result between stroke information of input characters and stroke information of characters stored in the dictionary section 5 according to this embodiment. FIG. 5(b) is a diagram showing weighting for comparison results. The input character pattern is divided into the first 5 strokes and the latter 4 strokes according to the division information 54,
Compare the first 5 strokes with the first 5 strokes of dictionary section 5, and compare the last 4 strokes.
This figure is compared with the last four strokes of dictionary section 5. The comparison results are weighted, and in this example, if they are less than a tolerance value, they are determined to be the same character. For example, if the allowable value is 100, the fifth
In the example of FIG. 4A, the input character pattern is written in the incorrect stroke order of 42 in FIG. 4, but comparison shows that the correct correspondence can be achieved. Note that the weighting and tolerance values are not limited to those in this example, and if there are multiple character codes determined to be the same, the one with the smallest value is determined to be the same character. Hereinafter, the output value of the comparison result will be referred to as a dissimilarity value. Reversing the weighting results in a similarity value.

FIG. 6 is a flowchart showing the processing procedure for dividing into two parts according to this embodiment. In step S60, the feature extraction unit 3
After receiving the input character pattern, it is determined in step S61 whether there is any character block for which comparison has not been completed, and if there is no character block, the process is terminated. If there is a character block that has not been compared yet, step S6
2, input the character block 52 from the dictionary section 5, and in particular, in step S63, divide information 54 is input from the character block 52.
Take out. In step S64, the input character pattern is divided into the first half and the second half according to the division information 54, and the strokes of the first half are compared. In step 365, the strokes in the latter half are compared, and in step S66, it is determined whether the input character pattern and the character pattern of the character block 52 are the same based on the results of step S64 and step S65, and it is determined that they are the same pattern. If possible, the character code 53 is output in step S67, and then the process returns to step S61. If it cannot be determined that the patterns are the same, the process immediately returns to step S61. In step S67, when the character code is output and the difference value at that time is set as the allowable value for subsequent determination, one character can finally be found. Also, if the allowable value is greater than or equal to the predetermined value,
You may also consider communicating input errors to the scribe.

In this embodiment, writing is performed using an input ben on a tablet, but writing may be performed using a finger using a tablet that senses pressure. Further, the method for determining the comparison result is not limited to weighting. Furthermore, depending on the number of strokes of the input character, in this example, if the comparison starts from 9 strokes, the processing time will be shorter.

However, in addition to errors in stroke order, it is also possible that the continuation is written. Therefore, the response of the pattern recognition device of the embodiment when a continuation of characters is written next will be explained.

Figure 7 is a diagram showing an example of writing ``shi'', and 70 shows ``shi'' written in the correct stroke order and in a Buddhist temple, while 71
is written in the wrong order of strokes, and part of it shows the ``thoughts'' written.

Figure 8(a) shows the continuation in this example written as “thought”.
It is a figure which shows the comparison result of the character Batano and the character pattern of a dictionary part. FIG. 8(b) is a diagram showing weighting of comparison results. Compare the first five strokes of the input character pattern with the first five strokes of the character pattern in the dictionary section, and compare the last four strokes of the input character pattern with the last four strokes of the character pattern in the dictionary section. Indicates that a comparison is to be made.

The example in FIG. 8(a) shows that even if the order of strokes is incorrect and some parts are continued, a correct correspondence can be taken by comparison.

It may be more advantageous to perform not only two divisions as described above but three or more divisions.

For example, if the character ``number'' is divided into ``rice'', ``女'', and ``久'', the recognition rate will be higher and the recognition time will be shorter. Next, a pattern recognition device that divides characters into arbitrary numbers will be described.

FIG. 10(a) is an example of a "number" character block 52, and r3F74J is a number character code 53, "6" and "
3" is the division information 54 indicating that the image is divided into the first six strokes of "rice," the next three strokes of "1 woman," and the remaining strokes of "missing." The next 13 blocks have “number” stroke information 5
5 are stored in the correct stroke order. Figure 10 (b
) is the division information 54 that separately has “2” as the division number.
It shows a block of characters that contains .

Figure 9 is a diagram showing an example of writing the word ``number''. 90 is an example of writing in the correct stroke order, while 91 is written in the wrong stroke order, and the part for ``ku'' is partially written in the correct stroke order. What follows is a written example. In this way, "rice" constitutes "r number",
``Onna'' and ``miss'' are each written in various stroke orders. However, in most cases, it is written in the order of ``rice'', ``woman'', and ``missing''.

FIG. 11 is a diagram showing a comparison result between the character pattern of the dictionary section for "number" and the input character pattern. Since the division information 54 for "r number" is "6" and "3", the character pattern is divided into "rice" (6 strokes), "r woman" (3 strokes), and "missing".
Each of these is compared with the first six strokes, six to nine strokes, and the last four strokes of the input character pattern to obtain correct pattern correspondence. In this way, in this embodiment, the incorrect stroke order characters and the continuation characters are made to correspond to the correct pattern with a small amount of processing.
In practical terms, free stroke order and continuous writing are possible.

FIG. 12 is a flowchart showing the processing procedure in the case of arbitrary division in this embodiment. In step 5120, the character pattern inputted from the feature extraction unit 3 is received, and the number of strokes of the character to be compared is determined from the number of strokes. Each character block 5
In the second comparison, it is determined in step 5121 whether or not it is the end, and if it is the end, the process ends. If it is not the end, in step 5122, the character block with the number of strokes determined in step 5120 is read, and in step 5123, particularly the division information 54 is extracted from it. In step 5124, the character pattern is divided according to the division information 54, and in step 5125, the strokes of the input character pattern and the previously registered character pattern are compared. In step 3126, it is determined whether the comparison of all strokes has been completed, and if not, steps 3125 to 126 are repeated. In step 5127, as a result of the comparison, it is determined whether the input character pattern and the character block are the same character. If they are determined to be the same character, the character code is output in step 5128, and the process returns to step 5121. to move on to comparing the next block of characters. Otherwise, the process returns directly to step 5121. Note that, as described above, if there are multiple character codes determined to be the same, the one with the smallest value is determined to be the same character.

In addition, in step 3128, when the character code is output, and at the same time, the value of the degree of difference at that time is set as the tolerance value for subsequent determination, one character can finally be found, and finally the tolerance value is greater than or equal to the predetermined value. If so, you may consider communicating the input error to the scribe.

In this example, an example in which the image is divided into three parts has been explained, but the number of divisions may be any number. Also, as for the division method, it is sufficient to consider the best method for each character.

Furthermore, if the determination is stopped when the degree of difference exceeds a predetermined value without waiting for the comparison of all strokes to be completed, the next recognition can be started quickly, and the recognition time can be shortened. Therefore, next, a pattern recognition lid making method will be described in which the determination is stopped when the degree of difference exceeds a predetermined value.

Reference numeral 130 in FIG. 13 is a character pattern for "shi" registered in the dictionary section 5, which is stroke information that follows the correct stroke order. 131 is a handwriting example of "to" and stroke information according to the correct stroke order.

In comparing these two characters, ``田'' is the first five strokes of ``し''.
Compare the first five strokes of the writing example ``ke'' with ``芦J''. Figure 14 shows the stroke correspondence between ``shi'' and ``ke'', and each stroke of ``田''. exists in the upper half of the character frame, whereas each stroke of "A" exists in the left half of the character frame, and the degree of difference between the first five strokes is large.
becomes larger than the allowable value. Since the degree of difference has become larger than the allowable value, the remaining "heart" and "ko↑" are not compared.

FIG. 15 is a flowchart showing the processing procedure when processing is interrupted due to the degree of difference. In step 5150, the character pattern inputted from the feature extraction unit 3 is received, and the number of strokes of the character to be compared is determined from the number of strokes. In comparing each character block 52, it is determined in step 5151 whether it is the end or not, and if it is the end, the process ends. If it is not the end, in step 5152, the character block with the number of strokes determined in step 5150 is read, and in step 5153, particularly the division information 54 is extracted from it. Step 515
In Step 4, the character pattern is divided according to the division information 54, and in Step 5155, the strokes of the input character pattern and the previously registered character pattern are compared. In Step 5156, the degree of difference of the comparison result in Step 5155 is added. do. In step 5157, step 51
It is determined whether the addition result in step 5156 is less than or equal to the allowable value, and if it exceeds the allowable value, the comparison for that character block is interrupted, and the process returns to step 5151 to start comparison with the next character block. In step 5158, it is determined whether the comparison of all strokes has been completed, and if not, step 515
Repeat steps 5 to 158. In step 5159, as a result of the comparison, it is determined whether the input character pattern and the character block are the same character. If they are determined to be the same character, the character code is output in step 3160, and the process returns to step 5151. to start comparing the next block of characters. Otherwise, the process returns directly to step 5151. still,
As described above, if there are multiple character codes determined to be the same, the one with the smallest value is determined to be the same character.

Also, in step 5160, if the character code is output and the difference value at that time is set as the tolerance value for subsequent determination, one character can be finally found, and finally the tolerance value is greater than or equal to the predetermined value. If so, you may consider communicating the input error to the scribe.

Furthermore, considering the order of characters in the dictionary section, step 5157
If the degree of dissimilarity exceeds the allowable value, it is possible to skip the comparison between the partially compared partial pattern and the character with the blank pattern.

The allowable value of the degree of difference may be determined by the number of strokes of the input character pattern, the number of strokes of the character pattern in the dictionary, or the number of strokes of each character or part. In addition, when comparing the characteristics of the input character pattern with character patterns registered in advance, the smallest difference between the characteristics of the input character pattern and the previously compared character pattern is selected. It may be a permissible value. Also, in this example, “
Although the comparison was made starting with ``田'', even when comparing starting with ``心'', depending on the character, the recognition time may be shorter or the recognition rate may be higher if the pattern that shows the most character features is compared first.

As explained above, when comparing the characteristics of the input character pattern and the character pattern in the dictionary, one or both of them are divided into several partial patterns, and as a result of comparing the corresponding partial patterns, the degree of dissimilarity is determined. If the similarity is greater than or equal to a certain value, or conversely, if the similarity is less than or equal to a certain value, the ongoing comparison between the input character pattern and the character pattern in the dictionary section is terminated and the process proceeds to the next step, or if the similarity is By skipping the comparison of characters in the pattern, recognition time can be shortened.

Next, an example of stroke information and a comparative example using stroke information will be described.

FIG. 16 is a diagram showing an example of basic strokes used to create stroke information, in which typical stroke shapes and their numbers are registered. Each stroke of the input character pattern is one of these basic strokes 16. It is classified as one of the following.

FIG. 17 shows a character block 52 registered in the dictionary section 5.
is a diagram showing an example of the character pattern of "What".

r323FJ is the character code 53 for "What". "6"
, r36, 92J, r8, 47J are the basic stroke number, the coordinates of the starting point, and the coordinates of the ending point of "rノ" which is the first stroke of "What". The coordinates of the starting point and the ending point are the x and y coordinate values of each end point after normalizing the position and size of the "what" standard pattern. Furthermore, "7", "22°60", r24
, 8J are the basic stroke number, starting point coordinates, and ending point coordinates of "1", which is the second stroke of "what". Stroke information 55 is similarly registered below. When dividing as described above, division information 54 is registered between character code 53 and stroke information 55.

FIG. 18 is a diagram showing an example of writing "What".

180 is an example of "what" written as a measure and in the correct stroke order, while 181 and 182 are examples of "what" written as a measure but in the wrong stroke order. 183 is a part of the word ``what'' written in the wrong stroke order.
This is an example. In conventional online character recognition using basic strokes, it is difficult to recognize such incorrect stroke order characters and continuation characters.

FIGS. 19 and 20 are flowcharts of an example of the recognition method according to this embodiment, and FIG. 21 is a diagram showing a part of the basic stroke similarity surface used for recognition.

FIG. 19 is a flowchart showing a procedure for comparing the input character pattern with the character pattern of the dictionary section based on stroke information. In step 5191, each stroke of the character pattern of the dictionary section and each stroke of the input character pattern are compared. are compared to find corresponding strokes, and then in step 5192 the sum of the distances between the corresponding strokes is found, which is used as the degree of difference between the input character pattern and the character pattern. In this way, the degree of difference between each character pattern registered in the dictionary section and the input character pattern is determined, and the character with the smallest degree of difference is output as the recognition result.

FIG. 20 is a flowchart showing the procedure for determining the correspondence between each stroke of the character pattern in the dictionary section and each stroke of the input character pattern. In step 5201, the distance between the starting points of both strokes is determined, and in step 5202
In step 5203, the distance between the end points is added, and in step 5203, the degree of difference between basic strokes is added according to the degree of difference between basic strokes table shown in FIG. 21, and this value is defined as the distance between strokes. , the stroke with the shortest distance among the strokes of the input character pattern is set as the corresponding stroke.

FIG. 22 is an example of writing "What", which is an example of writing 183 in FIG. FIG. 23 shows the basic stroke number, starting point coordinates, and ending point coordinates of each stroke extracted from the writing example shown in FIG. 22 according to the stroke order. Note that for decomposition into basic strokes, it is possible to set the starting point of each stroke and the basic stroke to the same coordinates and select the basic stroke with the shortest distance between the end points.

FIG. 24 is a diagram showing the correspondence between the input handwriting example of FIG. 22 and each stroke of the character pattern of FIG. 17. According to the input character pattern, the first stroke of the character pattern in the dictionary section is compared with each stroke of the input character pattern, and the first stroke of the character pattern in the dictionary section is compared with the first stroke of the input character pattern. This indicates that the strokes corresponded. The same applies below. The fourth stroke of the input character pattern corresponds to the fifth and sixth strokes of the character pattern in the dictionary section. Since the stroke information corresponds correctly, the degree of difference between the input character pattern "what" and the dictionary part character pattern "what" becomes a relatively small value and is correctly recognized.

As described above, by using the coordinates of the stroke end points of the basic strokes, correct stroke correspondence is often obtained even for incorrect stroke order characters and continuation characters, resulting in a high recognition rate. In this example, the basic strokes are broken down and compared, but a method of comparing the character strokes themselves may also be considered.

Now, in addition to dividing characters into groups, the recognition time can be shortened by comparing and finding specific occurrence rates, such as radicals, in advance. A pattern recognition device that realizes the above pattern recognition will be described below.

FIG. 25 is a block diagram of the pattern recognition device of the second embodiment, in which 251 is a tablet, 252 is an input bezel, 253 is a feature extraction section, 254 is a partial pattern discrimination section, 255 is a character discrimination section, and 256 is a dictionary section. , 257 is a partial pattern dictionary section of the dictionary section 257, and 258 is a character dictionary section of the dictionary section 257. When a scribe writes characters on the tablet 251 using the input pen 252, the coordinates of the pen and the up-down information of the pen are sent to the feature extraction unit 253.
53 normalizes the position and size of the input character pattern for one character, extracts features such as the number of strokes, the coordinates of the end points of each stroke, and the outline of each stroke, and discriminates partial patterns. 254. The partial pattern discrimination unit 254 compares the features sent from the feature extraction unit 253 with the features of each partial pattern registered in the partial pattern dictionary unit 257 to obtain partial pattern candidates, and compares the features sent from the feature extraction unit 253 with the features of each partial pattern registered in the partial pattern dictionary unit 257. The candidate partial pattern is sent to the character discrimination unit 255 along with the characteristics of the input character pattern. The character discrimination section 255 selects the character dictionary section 2 according to the candidate partial patterns according to the characteristics of the input character pattern and the candidate partial patterns sent from the partial pattern discrimination section 254.
Select character patterns to be compared from among the character patterns registered in 58, compare the characteristics of the input character pattern with the character pattern, and select the characters with the smallest degree of difference or highest degree of similarity as the recognition result. Output. Note that the partial pattern discrimination unit 254 and the character discrimination unit 255 are
As in FIG. 25, it is composed of a CPU, ROM, and RAM, and processing programs are stored in the ROMs 254a and 255a shown in FIG.

Figure 26 is a diagram showing an example of partial patterns.
Numbers correspond to radicals such as crowns and partial patterns that appear in many kanji. FIG. 27 is a diagram showing an example of strokes that appear in many characters, and a number is associated with each stroke. Since these strokes are the basic strokes that make up characters, they will be referred to as basic strokes hereinafter.

FIG. 28 is a diagram showing the structure of the partial pattern dictionary section 257 in FIG. The 0 part consisting of the number of strokes, the partial pattern number of the measure body of the partial pattern, the part 257b, and the cutting information contains the coordinates of the end points of each stroke making up the partial pattern and the basic information shown in FIG. A number representing stroke 27 is registered.

FIG. 29 is an example of a partial pattern block 257a, which corresponds to "thread". "12" is the partial pattern number of "thread", and "5" is the partial pattern number of "thread".
" is the number of strokes of "thread", "11" is the partial pattern number of the measure body "thread" of "thread", and the next five pieces of information contain the coordinates of the starting point, coordinates of the ending point, and outline of each stroke. A portion 257b including the representing number is registered. The last "0" is cutting information, and indicates that when comparing and determining the input character pattern and this partial pattern, the "first" 5 strokes in the stroke order are cut out from the input character pattern and compared. . If the extraction information is “1”,
5 of "end" in the stroke order from the input character pattern
Cut out the images and compare them. Figure 30 is part 2 of Figure 29.
This diagram illustrates the partial pattern on which 57b is based, and all strokes are in the left half of the character frame.

FIG. 31 is a flowchart showing the processing procedure of the partial pattern discrimination section, and FIG. 32 is a diagram showing a part of the difference table between basic strokes used in discrimination of partial patterns. The operation will be explained below according to FIG. 31. When the partial pattern discrimination section 254 receives the characteristics of the character pattern input from the feature extraction section 253 in step 5310, it accesses the partial pattern dictionary section 257 of the dictionary section 256, and proceeds to step 5311. In step 5311, it is determined whether there are any partial patterns that have not been compared, and if there are, the process advances to step 5312. Step 53
In step 12, each piece of information on the partial pattern to be compared is input, in step 5313 the corresponding information is cut out from the input character pattern using the cutting information, and in step 5314, the input character pattern and the end point of each stroke of the part are The sum of the distances between the strokes and the sum of the degrees of dissimilarity between strokes are calculated, and the sum of these two values divided by the number of partial strokes is determined as the degree of dissimilarity of this partial pattern with respect to the input character pattern. In this way, once the degree of difference of one partial pattern is determined, the process returns to step 5310. If it is determined in step 5311 that the comparison between the input character pattern and all partial patterns has been completed, the process proceeds to step 5314, and the three partial patterns with the smallest degree of difference are output as candidate partial patterns.

FIG. 33 is a diagram showing the configuration of an example of the character dictionary section 258 of this embodiment, and the character dictionary section 258 is divided by the number of strokes,
Furthermore, each stroke number dictionary section 258a is organized for each partial pattern. For example, as shown in FIG. 33, characters including "shi" form one group, and characters including "thread" form another group. At the beginning of each group, its partial pattern number is stored.

FIG. 34 is a flowchart showing the processing procedure of the character discriminator. In step 5340, the character discriminator 255 receives the characteristics of the character pattern and candidate partial patterns input from the partial pattern discriminator 254. In step 5341, it is determined whether there is a group of partial patterns that have not yet been determined, and if there is, the process proceeds to step 5342.In step 5342, the character dictionary section 258 is accessed, the partial pattern is input, and in step 5343, the process proceeds to step 5342. It is determined whether the partial pattern is included in the candidate partial patterns, and if so, the process proceeds to step 5344, where the character pattern belonging to the group of this partial pattern is compared with the input character pattern, and step 534
Return to 1.

If the partial pattern is not included in the candidate partial patterns in step 5343, the process advances to step 5345, where character patterns included in the group of partial patterns are skipped, and the process returns to step 5341. The next group of subpatterns is reached by performing a comparison in step 5344 or by skipping character patterns in step 5345. When the end of the stroke number dictionary section 258a is reached, the end of the stroke number dictionary section 258a is determined in step 5341, and the process ends. When comparing other stroke number dictionary sections 258a,
The process starts again at step 5340.

In this embodiment, the partial pattern discriminator outputs three candidate partial patterns, but the number does not necessarily have to be three, and may be as few as one or five. , it may be possible to output a certain number of candidate partial patterns, such as at most three candidate partial patterns.Furthermore, in comparing the characteristics of the input character pattern with each partial pattern dictionary section, the input If the character pattern is significantly different from all subpatterns, i.e.
If the degree of difference is very large, it may be output that there is no candidate partial pattern, and in such a case, the character discrimination unit may compare the input character pattern with all character patterns. It is also conceivable to combine the first embodiment and the second embodiment and select the one with a higher recognition rate and shorter recognition time depending on each character. In this case, those whose dissimilarity is smaller than a predetermined threshold are returned as candidate partial patterns, and if there is no smaller dissimilarity, the first partial pattern is returned.
Combinations such as comparing examples are possible.

As explained above, the dictionary section is provided with a partial pattern dictionary section together with a character dictionary section, and the partial pattern dictionary section includes at least the number of strokes of each partial pattern, the coordinates of the end points of each stroke, and information representing the shape of the stroke. Register information for cutting out a partial pattern from one character's information, use the partial pattern dictionary section to identify partial patterns for the input character pattern, and select character patterns to be compared based on the results. However, by comparing the characteristics of the input character pattern with the character pattern, the number of comparisons between the characteristics of the input character pattern and the character pattern is reduced, and a pattern recognition device with short recognition time can be realized. As a result, it becomes possible to expand the number of recognized characters, register many character patterns, and eliminate the processing time required for comparison with single character patterns, which eases restrictions such as the number of strokes and stroke order. Information processing equipment becomes possible.

In this example, the explanation was given for a character pattern that includes one symbol, one figure, but in pattern recognition such as audio and one image, division into cohesive subpatterns and backup of subpatterns with a high occurrence rate are also possible. Technical ideas such as judgment can be widely used.

[Effects of the Invention] According to the present invention, the pattern can be broken down into basic strokes, and the strokes can be compared with the strokes of a pattern registered in advance, with full consideration given to variations in stroke order during pattern creation. It is possible to provide a pattern recognition device that can be recognized and has a short recognition time.

[Brief explanation of the drawing]

FIG. 1(a) is a block diagram of the pattern recognition device of the first embodiment, FIG. 1(b) is a block diagram of the character discrimination section, FIG. 2 is a diagram showing the structure of the dictionary section, and FIG. 3(a) is a block diagram of the pattern recognition device of the first embodiment. ), (b) is a diagram showing the structure of the character block for ``shi'', Figure 4 is a diagram showing ``shi'' in various stroke orders of measures, and Figure 5 (
Figure 5 (b) is a diagram showing the weighting that represents the degree of difference, Figure 6 is a flowchart showing the processing procedure in the case of two divisions, and Figure 7 is a diagram showing the comparison results of the "thoughts" of the measures. The following is a diagram showing a handwritten example of the written ``shi'', and Figure 8 (a) is a diagram showing the comparison result between the input character pattern of the written ``shi'' and the character pattern in the dictionary section. Figure 8 (b) is a diagram showing the weighting representing the degree of difference, Figure 9 is a diagram showing an example of writing "number", and Figures 10 (a) and (b) are configuration diagrams of the character block of "number". , Fig. 11 is a diagram showing the comparison result of the input character pattern of "r number" and the character pattern of the dictionary section, Fig. 12 is a flowchart of the processing procedure in case of arbitrary division, and Fig. 13 is a diagram showing the comparison result of the input character pattern of "r number" and the character pattern of the dictionary section. Figure 14 shows an example of the registered writing of "shi" and "ke", Figure 14 shows the stroke correspondence between "shi" and "ke", Figure 15 shows the input when interrupting the process due to the degree of difference. Flowchart showing the processing procedure of the character pattern and the character pattern of the dictionary part, FIG. 16 is a diagram showing an example of the basic stroke table, TS17
The figure shows an example of the character pattern for ``What''. Figure 18 shows an example of how the word ``What'' is written. Figures 19 and 20 are flowcharts showing the recognition procedure. Figure 21 shows the differences between basic strokes. Figure 22 shows an example of how to write "What"; Figure 23 shows the basic stroke number and coordinates of the start and end points of each stroke in the example of how to write "What" in Figure n. FIG. 24 is a diagram showing the correspondence between the strokes of the input character pattern and the character pattern in the dictionary section when it is decomposed into basic strokes; FIG. 25 is a configuration diagram of the pattern recognition device of the second embodiment; Figure 26 is the side opening of the partial pattern, Figure 27 is the side opening of the basic stroke, Figure 28 is a configuration diagram of the partial pattern dictionary, and Figure 29 is the "thread" part registered in the partial pattern dictionary. Figure 31 is a flowchart showing the processing procedure of the partial pattern discriminator, Figure 32 is part of the difference table between basic strokes. FIG. 33 is a diagram showing the configuration of the character dictionary section, and FIG. 34 is a flowchart showing the processing procedure of the character discrimination section. In the figure, 1...tablet, 2...input pen, 3...
・Feature extraction unit, 4...Character discrimination unit, 5...Dictionary unit,
41...CPU, 42...ROM, 43...RA
M, 251... Information input section, 252... Input pen,
253...Feature extraction unit, 254...Partial pattern discrimination unit, 255...Character discrimination unit, 256...Dictionary unit,
257...Partial pattern dictionary section, 258...Character dictionary section, 254a. 255a...ROM. Patent applicant: Canon Co., Ltd. Figure 1 (a) Figure 2 Figure 3 (a) Figure 3 (b) Figure 4 (1) Figure 5 (a) Figure 5 (b) Figure 8 ( Q) Figure 8(b) Figure 11 Figure 16 Figure 17 Figure 18 Figure 19 Figure 20 Figure 21 Figure 22 Figure 23 Figure 25 Figure 26 Figure 28 Figure 29 Figure 30 Figure 31 Figure 32 Figure 33

Claims (5)

[Claims] (1) In a pattern recognition device that recognizes the input pattern by comparing the input pattern with a pre-registered pattern, the input pattern is decomposed into basic strokes. It is characterized by comprising a comparison means for comparing stroke information and stroke information including stroke end point coordinates registered in advance, and a recognition means for recognizing the input pattern based on the comparison result of the comparison means. pattern recognition device. (2) The pattern recognition device according to claim 1, wherein the pattern is a character pattern including symbols and figures. (3) The comparison means compares the difference between the strokes based on the sum of the value representing the difference in the coordinates of the end points and the value representing the difference in the basic stroke, and a predetermined tolerance value. The pattern recognition device according to scope 1. (4) The recognition means recognizes the difference between the patterns based on the magnitude of a value representing the difference in stroke between the input pattern and the pre-registered pattern and a predetermined tolerance value. The pattern recognition device according to scope 1. (5) The pattern recognition device according to claim 4, wherein the pre-registered pattern includes a code for identifying the pattern and a stroke representing a stroke of the pattern.