JPS61136181A - Production of dictionary for character recognizer - Google Patents

Abstract

PURPOSE:To shorten the collation processing time by supplying plural times a character which is used frequently to a specific application, detecting the standard patterns having coincidence of geometric features and then carrying out the dictionary rearrangement according to the difference between count values corresponding to said standard patterns. CONSTITUTION:A character which is used frequently to a specific application is recorded on the paper, etc. and then read by a reading head 1. The patterns of the character undergo the rough sorting process in terms of the extracted character features. Here the standard pattern data having the highest priority is loaded to a buffer register 34 among those pattern data stored in a dictionary memory 31 for rough sorting. At the same time, the count value stored in the counter area corresponding to the relevant standard pattern is loaded to a buffer register 33. Then the count values are compared with each other by a value comparator 38 for decision of the order of standard patterns.

Description

[Detailed Description of the Invention] <Technical Field of the Invention> This invention reads unknown characters, symbols, figures, etc. (hereinafter collectively referred to as "unknown characters") by optical methods, etc., and determines the geometry of the unknown characters. The present invention relates to a character recognition device for identifying and recognizing unknown characters by extracting a scientific feature and then comparing the feature with a standard pattern stored in a dictionary means. The present invention relates to a method for creating a new dictionary used in a type character recognition device.

<Summary of the Invention> In the dictionary creation method for a character recognition device of the present invention, a counter means is provided in advance in the dictionary means corresponding to each of the standard patterns, and characters frequently used for a specific purpose are stored in the character recognition device. By inputting the input multiple times to the reading means, the count amount of the counter means corresponding to the standard pattern that matches the long sword character pattern is incremented by 1, for example, according to the number of matches, and the count amount in the dictionary means is incremented by 1 according to the number of matches. It is configured to change the arrangement of standard patterns, thereby making it possible to easily and quickly create a dictionary for a character recognition device in which frequently cited standard patterns can be collated more quickly.

<Background of the Invention> In conventional character recognition devices, as shown in FIG.
The unknown characters recorded thereon are read by, for example, an optical reading head 1, and the optical signal is sent to an A/D converter 2.
The pattern is converted into a digital electrical signal and converted into black and white binary data, and the pattern is temporarily stored in the image memory 3. Next, the preprocessing circuit 4 performs so-called preprocessing such as smoothing, noise removal, and thinning processing on this pattern to improve the precision and facilitate the subsequent feature extraction processing. Thereafter, the feature extraction circuit 5 detects and extracts the geometric features of the unknown character necessary for character recognition, such as the presence or absence and number of intersections, branch points, loops, etc., from the preprocessed pattern. The dictionary matching circuit 6 sequentially matches the extracted character features with standard patterns for each character stored in a predetermined order in the dictionary 7, identifies unknown characters based on the degree of matching, and outputs the results. For example, display means (not shown)
Send to.

FIG. 7 specifically shows the memory contents of the dictionary 7. In this dictionary 7, standard patterns for each character, including multiple subcategories, are arranged by hand or by a computer method using a learning method. has been done.

By the way, the matching process in the dictionary matching circuit 6 is performed by sequentially comparing unknown character features with each standard pattern. In this case, the time required for verification becomes extremely long, and as a result, the processing speed of character recognition is extremely reduced. Therefore, in current character recognition devices, as shown in Figure 8, first, only general characteristics such as the presence or absence of loops and the number of connected characters are detected, and candidate characters are then grouped in advance according to these characteristics. It is classified into one of the following. After that, the details of the features are sequentially taken in, medium classification is performed based on character intersections, branches, etc., and small classification and detailed classification are performed based on more detailed features, and candidate characters are sequentially narrowed down to obtain a single or multiple candidate characters. Then, detailed comparison and judgment are performed only on these candidate characters, and unknown characters are recognized according to the degree of matching.

FIG. 9 shows the details of this narrowing down process, and in step 11 the general characteristics are A1, A2゜..., A
Broadly classified according to which group of n it belongs to,
For example, if the characteristics indicated by A apply, step 1
Proceed to step 2. In step 12, intermediate classification is performed and the BI
+ B2 r...-, Bn is determined, and the process similarly proceeds to step 13 of minor classification and step 14 of detailed classification. Step 1 of detailed classification
When step 4 is performed, only candidate characters belonging to that detailed classification are compared and determined. Note that although FIG. 9 depicts only the flow shown in steps 11, 12, 13, and 14 in detail, and omits the flow to other classifications, the same applies to these. In addition, if there is no subgroup corresponding to the characteristics of the unknown character in each stage of classification, it is possible that the unknown character is not prepared in the standard pattern or is not in the form of a character due to a typographical error. Therefore, in step 15, reject (
(unrecognized) and proceed to re-input or other processing.

According to the above method, standard patterns having features that are significantly different from those of unknown characters are excluded from matching targets, so that the matching speed is considerably improved. However, the classification itself is based on grouping based on character feature numbers, and the order of judgment in classification remains fixed to the order arranged when the dictionary was created. For this reason, with conventional dictionary creation methods, even if a great deal of effort is put into creating a dictionary for character recognition devices, frequently used classifications are not always determined and collated quickly.
There are many cases in which a frequently used classification is compared after a less frequently used classification is compared, and there is a certain limit to the improvement of the matching speed and therefore the character recognition speed.

Particularly, when there are many types of input characters, such as in the case of inputting Chinese characters, this phenomenon is particularly noticeable.

<Object of the invention> This invention is intended to eliminate the above-mentioned drawbacks,
It is an object of the present invention to provide a method for creating a dictionary for a character recognition device, which shortens collation processing time and thereby speeds up character recognition processing.

<Structure and Effects of the Invention> In order to achieve the above object, in the dictionary creation method for a character recognition device of the present invention, a counter means is provided in advance in the dictionary means corresponding to each of the standard patterns, and a counter means is provided in advance for each standard pattern. By inputting frequently used characters into the reading means of the character recognition device multiple times, a standard pattern that matches the geometrical features of the input character pattern is detected in the dictionary means, and a counter corresponding to the standard pattern is detected. Varying the count amount of the means by a predetermined amount according to the number of matches, and comparing the count amounts of the counter means with each other,
The arrangement is such that the arrangement order of the standard patterns in the dictionary means is changed and rearranged depending on the size of the count amount.

According to this invention, the count amount of each counter means is an index of the frequency with which the corresponding standard pattern is cited, and based on this, the standard patterns are automatically rearranged to determine the frequency of use. Since it is possible to create a dictionary in which standard patterns of high-quality characters are arranged in a higher priority order, in a character recognition device using this dictionary, the order of matching is performed with priority given to the classification with higher frequency. , it is possible to shorten the collation time and therefore speed up character recognition.

<Description of Embodiments> FIG. 1 shows the main parts of a character recognition device used for carrying out the present invention. This device example and the conventional device shown in Fig. 6 are as follows:
A dictionary rearrangement circuit 21 is newly added, a counter area as a counter means is provided in the dictionary 23 corresponding to each standard pattern, and an external storage device 24 is capable of storing dictionary contents. CPU22
They differ in that they are programmed to also execute control and dictionary creation operations, which will be described later.

An example of the configuration of the dictionary rearrangement circuit 21 and its peripheral circuits is shown in FIG. In FIG. 2, the dictionary memory 31 is
The memory address register 32 is a memory that forms the main part of the dictionary 23 described above, and is used to hold an address when performing a read/write (R-/W) operation to the dictionary memory 31. be. This dictionary memory 31
The arrangement of storage areas in is shown in FIG.

In the illustrated example, an area for storing standard pattern character feature data and a counter area for storing count amounts (to be described later) are paired and arranged in a predetermined order. It is set corresponding to each classification, minor classification, and detailed classification. At this point, this arrangement is the arrangement when a dictionary arrangement created by hand design or learning method is loaded into the dictionary memory 31 from the external storage device 24, such as a floppy disk or magnetic tape.

Next, the buffer registers 34 and 36 are stored in the dictionary memory 31.
The other buffer registers 33 and 35 are registers that load and temporarily hold data related to the character features of standard patterns in the buffer register 34.
．． This is for holding the contents of the counter area, that is, the count amount, corresponding to each of the standard patterns buffered in 36. The memo IJR/W control circuit 37 uses the memory address register 32 to control the timing of read/write operations to the dictionary memory 31. Further, the magnitude comparison circuit 38 is
This circuit compares the amounts buffered in the buffer registers 33 and 35, that is, the counted amounts, and generates an output according to the magnitude relationship. Furthermore, the match detection circuit 39 extracts the character feature data buffered in the buffer register 34 and the feature extraction circuit 5 (FIG. 2).
A gate circuit 40
and is applied to the buffer register 33. This detection output is configured to be activated only when there is a match, and thereby the count amount in the buffer register 33 is incremented by 1, and the gate circuit 4
Gate 0 is now open. When the gate of this gate circuit 40 opens, the comparison output from the magnitude comparison circuit 38 is output to the memory R/W control circuit 37, and requests interrupt processing to the C, PU 22, thereby The rearrangement in the dictionary memory 31, which will be described later, is performed.

Next, the present invention will be described in detail with reference to these figures. First, a sheet of paper with characters relatively often used for a specific purpose, such as commercial texts, is prepared, and read quickly by the reading head 1 (FIG. 1). The character pattern read in this way is subjected to A1 day conversion, black and white binarization, preprocessing and feature extraction processing in the same order as explained in FIG. The major classification process explained and illustrated in FIG. 8 is performed for the features. In this case, first, among the standard pattern data for major classification stored in the dictionary memory 31, the first one, that is, the one with the highest priority, is loaded into the buffer register 34, and the counter area corresponding to the standard pattern is loaded. The stored count amount is loaded into the buffer register 33. The other buffer registers 35 and 36 shall be cleared to O before this, and the dictionary memory 3
Assuming that the counter area in 1 is also initially 0, at this point the buffer register 33°35.3
The contents of 6 are all O. An example of the arrangement of the dictionary memory 31 at this stage is shown in FIG. 4(a).

Next, the counts of each of the buffer registers 33 and 35 are compared by the magnitude comparison circuit 38, and the -number construction circuit 39 checks the consistency between the characteristics of the standard pattern in the buffer register 34 and the characteristics of the input character. To detect. If it is determined that they match, the output from the -i detection circuit 39 is activated and the buffer register 33
The count value in is incremented by 1, and the gate of gate circuit 40 is opened. However, if the magnitude comparison circuit 38 is configured so that its output is activated only when the count amount of the buffer register 33 is larger than the count amount of the buffer register 35, in this example, Buffer register 33.35
Since the count amounts of each of are 0 and are equal (the comparison operation of the magnitude comparison circuit 38 is performed before the above-mentioned increment operation), its output is in an inactive state and is not rearranged. There will be no such requirement. Therefore, the CPU 22 re-stores the incremented contents of the buffer register 33 and the standard pattern data in the buffer register 34 at the original address in the dictionary memory 31. On the other hand, when a match of features is not found, the output of the minus number construction circuit 39 is inactive, and the increment and rearrangement described above are not active.

When the comparison with the first standard pattern is completed and the next standard pattern is to be compared, the buffer register 33.
The contents of 34 are transferred to the buffer register 3536A-, and the character feature data of the standard pattern having the next rank in the dictionary memory 31 is transferred to the buffer register 34.
The count amount is loaded into the buffer register 33, respectively. Thereafter, the same match detection and size comparison as described above are performed, but here, the standard pattern stored in the buffer register 34 and the input character feature match, and the count amount of the buffer register 33 is Assume that the value is larger than the count amount. That is, at present, the standard pattern for the buffer register 34 having a lower priority is the buffer register 36.
This is a case where the count amount as a result of past character recognition shows that the word has a higher usage (citation) frequency than the standard pattern. At this time -
Due to the number, the count amount of the buffer register 33 is incremented by 1, and the magnitude comparison circuit 3 is incremented by 1.
8 becomes active, and requests rearrangement to the memory R/W control circuit 37 and CPU 22 via the gate circuit 40. As a result, the CPU 22 controls the buffer register 33 through setting the address of the memory address register 32 and setting the write timing control output of the memory R/W control circuit 37.
．． The contents of 34 and the contents of buffer registers 35 and 36 are stored in the dictionary memory 31 in the reverse order from when they were loaded, and the order of these two standard patterns is changed and rearranged.

When the major classification process and the reordering process executed in parallel are completed in this way, the process moves on to medium classification, small classification, and detailed classification, but the same reordering process is also performed in parallel for these. Let's do it.

In addition, after candidate characters are narrowed down by detailed classification,
A similar rearrangement can also be performed when comparing multiple candidate characters.

Once such processing is completed, the frequently used characters that were input last time are inputted into the reading head 1 again. This further increments the count corresponding to that character, and by repeating this procedure multiple times as necessary, standard patterns with greater usage (citation) frequency are given higher priority. Rearrangement will be possible. For example, "I", "Z"
, "3", and "0" are frequently used.
The arrangement of input characters as shown in FIG. 5 is sequentially input into this character recognition device to obtain a dictionary having a standard rearrangement pattern as shown in FIG. 4(b). In FIG. 4(b), "3" shown in area 43 is input three times as shown in counter area 44, and "0" shown in area 45 is also input three times as shown in counter area 45. It can be seen that it has come to have a high priority. Note that in this figure, the existence of multiple standard patterns for the same character is due to the existence of subcategories.

When such rearrangement is completed, the desired dictionary is obtained, but if it is stored only in the dictionary memory 31,
This dictionary disappears when the power is turned off after work. Therefore, it is effective to dump the created dictionary contents to the external storage device 24 at the end of the business.

In this way, when starting the next business, the created rearrangement dictionary can be loaded from the external storage device to the dictionary memory 31 and used again.

In addition, the above rearrangement can be performed not only when inputting frequently used characters, but also during subsequent character recognition operations, so dictionaries created by this method are not only useful in themselves, but also useful for subsequent character recognition. There is also the advantage that a more desirable arrangement can be achieved by repeating the recognition operation. Conversely, by adding the counter area used for dictionary creation at load time and deleting it at dump time, 1.
When a dictionary created by this method is installed in a device that will actually be used, it can be done so that it does not affect recognition processing.

In addition, in the above embodiment, if matching is assumed to be performed using classification, or if sequential matching is performed under circumstances such as the number of types of input characters is small, rearrangement of unclassified standard patterns may be performed. You can also do it. Further, it is of course possible to make modifications such as providing a down-counter type counter area as the counter means, decrementing the count every time a negative number is reached, and rearranging the count in descending order. Furthermore, there is no particular limitation on the input order of frequently used characters; after inputting the same character multiple times as shown in Figure 5, other frequently used characters may be input multiple times. You may enter the information multiple times by sequentially inputting the information each time and then inputting it again from the beginning.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the main parts of an embodiment of the present invention, FIG. 2 is a block diagram showing a dictionary reordering circuit and its peripheral circuits, and FIG. 3 is an example of the arrangement of the standard pattern data area and the counter area. 4 is a diagram showing an example of rearrangement in the embodiment, FIG. 5 is a diagram showing an example of input characters, FIG. 6 is a block diagram showing a conventional character recognition device, and FIG. 7 is a diagram showing a standard Bata recognition device. −
FIG. 8 is a flowchart showing classification matching, and FIG. 9 is a flowchart showing details of classification matching. 5... Feature extraction circuit 6... Dictionary matching circuit 21.
...Dictionary rearrangement circuit 24...External storage device 31...
・Dictionary memory 33 to 36...Buffer register 38...Size comparison circuit 39...-Several structure output circuit Patent applicant Tateishi Electric Co., Ltd.") r2U Mo'f Natsuyama product C+/1 kVs once manufactured by Obe Showing doo i diagram

Claims (1)

[Claims] The geometric features of the unknown character pattern read by the reading means are extracted, the geometric features are sequentially compared with standard patterns arranged in a predetermined order in the dictionary means, and based on the matching results, the unknown character pattern is extracted. In a character recognition device that recognizes characters, counter means is provided in advance in the dictionary means corresponding to each standard pattern, and characters frequently used for a specific purpose are inputted multiple times to the reading means of the character recognition device. A standard pattern that matches the geometrical feature of the input character pattern is detected in the dictionary means, and the count amount of the counter means corresponding to the standard pattern is changed by a predetermined amount according to the number of matches. and a character recognition device characterized in that the counts of the counter means are compared with each other, and the arrangement order of the standard patterns in the dictionary means is changed and rearranged according to the magnitude of the count amounts. How to create a dictionary.