JP4805485B2 - Word recognition method and word recognition device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a word recognition method and a word recognition device that perform word recognition based on a character recognition result from an input character image, for example, in a document image processing device.
[0002]
[Prior art]
In character recognition, a general method is to prepare a character dictionary and perform matching processing with a dictionary pattern in order to determine which category a written character belongs to. Similarly, when word recognition is performed, a database of words that can be candidates is prepared in advance, and the word that is closest to the character is compared with the character recognition result of each character constituting the word. It is a common technique to measure the degree of. A well-known method is a so-called DP-matching method.
[0003]
By the way, if the recognition result of each character is uniformly reliable, word recognition is not so difficult, but if the result of character extraction in the word to be read is indeterminate, When characters change due to the influence of adjacent characters such as continuation characters, and when there are many similar characters, the result of character recognition ranked by the similarity value becomes unstable. Since correct character candidates are difficult to obtain correctly, there is a problem that a technique such as the DP-matching method described above is not effective.
[0004]
[Problems to be solved by the invention]
A prominent example in the case where correct character candidates cannot be obtained correctly is a case where there are many similar characters (assuming that the result of character segmentation is reliable to some extent). For example, take a Chinese character typified by a 2-byte character as an example. Compared with alphabetic characters, the number of categories to be recognized is thousands, and there are many similar characters. For example, as seen in the relationship between the characters “萩” and “荻”, the conventional word recognition method does not necessarily store the structural information of these characters, even though they are very close when measured by a similarity scale. The problem is not reflecting.
[0005]
Here, for example, in the case of recognizing a name described in a name list, when the words “Iwata” or “Iwata” must be strictly distinguished from the reading result, “萩” and “荻” It is necessary to compare the similarity values, which is equivalent to the conventional word recognition method. However, when this is a distinction between “Iwata” and “Akita”, and also “Iwata” and “Kida”, the general information of the character structure is extracted from the candidate sequence of character recognition results, and the former Can be recognized by reflecting the character structure by incorporating a process to check for the presence of "Kusakumuri", and in the latter, a process to more simply check the number of component characters that make up the character. Will improve.
[0006]
Therefore, an object of the present invention is to provide a word recognition method and a word recognition device that are more flexible and versatile and can improve the performance of word recognition.
[0007]
[Means for Solving the Problems]
The word recognition method of the present invention is a word recognition method for performing word recognition based on a character recognition result. Based on the character recognition result, for each character appearing in a word candidate, the separable component characters constituting the character A first step of determining a character type consisting of a spatial layout position and the number of characters of its component parts and acquiring character type information, and word recognition processing is performed using the character type information acquired in the first step. And a second step.
[0008]
According to another aspect of the present invention, there is provided a word recognition method for performing word recognition based on a character recognition result, a first step for determining whether or not to perform character structure recognition based on the character recognition result; If it is determined that character structure recognition is to be performed in one step, the spatial layout position and the configuration of separable component characters constituting the character for each character appearing in the word candidate based on the character recognition result A second step of determining a character type consisting of the number of component characters and acquiring character type information; a third step of performing word recognition processing using the character type information acquired in the second step; A fourth step of outputting the character recognition result as it is as a word recognition result when it is determined in step 1 that character structure recognition is not performed.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0011]
FIG. 1 schematically shows a configuration of a word recognition apparatus to which a word recognition method according to an embodiment of the present invention is applied. This word recognition device is roughly divided into a character recognition unit 1 that performs character recognition on an input character image including a word to be recognized, a character structure recognition determination unit 2 that determines whether to perform character structure recognition, and character structure recognition. The character structure recognition unit 3, the detailed character recognition unit 4 that performs detailed character recognition, the character structure database 5, the word database 6, and the intra-word character analysis unit 7 are configured.
[0012]
The character structure recognition unit 3 includes a character type determination unit 31 that determines a character type for each character that appears in a word candidate, and a recognition determination unit 32 that determines whether or not to perform detailed character recognition. Yes.
[0013]
The detailed character recognition unit 4 includes a component character position detection unit 41, a component character image cutout unit 42, and a component character recognition unit 43.
[0014]
The character structure database 5 includes a character type storage unit 51, a similar component character storage unit 52, a similar component character number storage unit 53, a similar component character position storage unit 54, and a component character storage unit 55.
[0015]
The in-word character analysis unit 7 includes a word candidate extraction unit 71, a character type extraction unit 72, a detailed character recognition position determination unit 73, and a similar component character position determination unit 74.
[0016]
Next, the word recognition method according to this embodiment in such a configuration will be described in detail with reference to the flowchart shown in FIG.
[0017]
First, for example, when a character image in a document image acquired by a scanner (not shown) is given, the character recognition unit 1 performs character recognition on the input character image, and the recognition result list is a character code. It is given as a pair with similarity (step S1). In a character candidate string (word candidates that are character recognition results), a plurality of pairs are given.
[0018]
The character structure recognition determination unit 2 determines whether or not to perform character structure recognition from a plurality of character candidate lists given in the character candidate string based on the magnitude of the similarity value (step S2). If the character structure recognition is not performed as a result of this determination, the similarity output (character recognition result) is directly output as the word recognition result (step S3). This is equivalent to a conventional word matching algorithm.
[0019]
As a result of the determination in step S2, when the character structure is recognized as a similar character candidate based on the similarity value, the process proceeds to the process of the character structure recognition unit 3. In the character structure recognition unit 3, the character type determination unit 31 first determines the character type of each character that appears in the word candidate, and acquires character type information (steps S4 and S5).
[0020]
That is, the character type determination unit 31 makes a request to the character structure database 5 in order to check the character types of the upper character candidates in the given character candidate list. In the character structure database 5, component character layout information is stored in advance for each character code. For example, when an inquiry is made regarding the character “萩” (“upper”, “lower left”, “lower right”), etc. Information about the global structure of characters can be obtained. This is called a character type.
[0021]
FIG. 3 shows the classification of these character types. The characters “萩” and “荻” described above are of type 1, “autumn” and “field” having the structure of “(left” “right”). This indicates that a character type is defined for each component character layout, such as type 2. Although it is expected that the same character types are arranged in the character candidate list, if different character types are mixed, the representative character type is selected while considering the upper character candidate list.
[0022]
Since the character type is global information, it is used for the purpose of reducing word candidates to be collated. When the character type determining unit 31 of the character structure recognizing unit 3 requests the word database 6 for the character type of the word candidate, the intra-word character analyzing unit 7 can obtain the character type information of the corresponding word from the character structure database 5. It can be done.
[0023]
The recognition determination unit 32 of the character structure recognition unit 3 compares the character type information obtained as described above with the character type of the input word candidate to determine whether word recognition can be performed with the character type information. Is determined (step S6). If the result of this determination is that word recognition can be performed using the character type information, the word recognition processing is performed using the obtained character type information (step S7), and the recognition result is output (step S3).
[0024]
If the character type is significantly different and it is determined that detailed character recognition is not necessary, reject output is selected as the recognition result.
[0025]
If the result of determination in step S6 is that word recognition cannot be performed using the character type information, the process proceeds to the detailed character recognition unit 4 (step S8). The detailed character recognition unit 4 makes a request to the word database 6 and obtains information related to the detailed character recognition of the corresponding character. In the word database 6, a plurality of similar word candidates that are not rejected by the character type check remain in the list. If the character types are equal, a request is made to the character structure database 5 for information on which part of the part character is different. By obtaining information about the correspondence between the character type stored in the character structure database 5 and its component part characters, it is first determined which part to focus on when recognizing the character.
[0026]
FIG. 4 takes up an example of type 1 in FIG. Many characters belonging to type 1 in FIG. However, when only “Iwata” or “Iwata” is discriminated from the list of similar word candidates, the part characters (“upper” and “lower right”) in the character type are common. The information that the component character located at (“lower left”) should be noted for distinction is obtained by calculation from the correspondence between the character type from the character structure database 5, the component character, and its position information.
[0027]
In addition, since component characters that have similar shapes are clustered in advance, there is no need to strictly distinguish between “禾” and “tree” included in similar component categories as shown in FIG. Both are treated as the same part character. The number of elements included in the same similar component category held in the character structure database 5 is a measure indicating the degree of ambiguity. If the number of elements is large, the shape becomes indefinite. If the number of elements is examined for each component character, it indicates a selection criterion as to which component character should be examined when determining the character.
[0028]
In the detailed character recognizing unit 4, since the information of the component character position to be distinguished is obtained from the output of the recognition determining unit 32, after normalizing the input character image, the component character image is obtained from the image near the corresponding region. Extract. Generally, in a print character, the component character is clearly separated, and it is easy to obtain a component character image by labeling processing. If the component characters are not separated, the component character image is obtained by cutting at an appropriate position using projection or peripheral features. By performing recognition processing on the component character image obtained in this way, it is possible to determine whether or not the component character image belongs to the desired component character category.
[0029]
FIG. 5 shows the correspondence between characters and their structure information. The character structure database 5 includes, for each character code, a character type a representing global information, information b indicating the shape of the component character, and information on similar component character categories to which the component characters belong ( Shape and number) c, and can provide such information upon request.
[0030]
FIG. 6 schematically shows how characters appearing in word candidates are recognized from the input character image through character type determination and detailed character recognition.
[0031]
7 and 8 show specific examples in which similar words (names) are recognized. First, in FIG. 7, name candidates consisting of three characters are listed, and each character type is shown in a block diagram. The example shows that it can be determined by comparing the character types of the first two characters.
[0032]
Similarly, in FIG. 8, similar word candidates having the same character type are listed, and it is possible to obtain information about which position should be recognized by making an inquiry to the word database 6. Show. In the example, recognition of names in the same household is given.
[0033]
As described above, according to the above-described embodiment, character structure and character type are defined, and characters included in word candidates to be collated are classified in advance. In particular, when there are many similar categories such as kanji, the similarity is obtained by appropriately utilizing the global information of characters and the information on the shape of component characters according to the word candidates to be matched. In addition, the classification based on the classification of the character structure becomes effective, and it is more flexible and versatile, and the performance at the time of word recognition is remarkably improved.
[0034]
This word recognition method is also effective for continuous characters, broken characters, and the like found in handwritten characters. Even if characters that are not recorded in the JIS code are expanded or added in the future, the same method can be used.
[0035]
【The invention's effect】
As described above in detail, according to the present invention, it is possible to provide a word recognition method and a word recognition device that are more flexible and versatile and can improve the performance of word recognition .
[Brief description of the drawings]
FIG. 1 is a block diagram schematically showing a configuration of a word recognition device to which a word recognition method according to an embodiment of the present invention is applied.
FIG. 2 is a flowchart for explaining the flow of a word recognition method according to the embodiment of the present invention.
FIG. 3 is a diagram for explaining character type classification;
FIG. 4 is a diagram illustrating grouping of character components.
FIG. 5 is a diagram for explaining the correspondence between characters and their structure information.
FIG. 6 is a diagram schematically illustrating a state in which characters appearing in word candidates are recognized from an input character image through character type determination and detailed character recognition.
FIG. 7 is a diagram showing a specific example of similar word recognition when it can be recognized in a character type.
FIG. 8 is a diagram showing a specific example of similar word recognition when detailed character recognition is required.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Character recognition part, 2 ... Character structure recognition determination part, 3 ... Character structure recognition part, 4 ... Detailed character recognition part, 5 ... Character structure database, 6 ... Word database, 7 ... In-word character analysis part, 31 ... Character Type determination unit, 32 ... recognition determination unit, 41 ... component character position detection unit, 42 ... component character image cutout unit, 43 ... component character recognition unit.

Claims (6)

In a word recognition method for performing word recognition based on a character recognition result,
Based on the character recognition result, for each character appearing in the word candidate, the character type information is obtained by determining the character type consisting of the spatial layout position of the separable component characters constituting the character and the number of the component characters. A first step to:
A second step of performing a word recognition process using the character type information acquired in the first step;
A word recognition method comprising:   In the first step, character type information is obtained by referring to a character structure database in which component character layout information is stored in advance for each character code for each character appearing in a word candidate. The word recognition method according to claim 1. In a word recognition method for performing word recognition based on a character recognition result,
A first step of determining whether to perform character structure recognition based on the character recognition result;
When it is determined that the character structure recognition is performed in the first step, based on the character recognition result, for each character appearing in the word candidate, a spatial layout position of separable component characters constituting the character and A second step of determining a character type consisting of the number of characters of the component parts and acquiring character type information;
A third step of performing word recognition processing using the character type information acquired in the second step;
A fourth step of outputting the character recognition result as it is as a word recognition result when it is determined not to perform character structure recognition in the first step;
A word recognition method comprising: In a word recognition device that performs word recognition based on a character recognition result,
  Based on the character recognition result, for each character appearing in the word candidate, the character type information is obtained by determining the character type consisting of the spatial layout position of the separable component characters constituting the character and the number of the component characters. First means to:
  Second means for performing word recognition processing using the character type information acquired by the first means;
  A word recognition device characterized by comprising: The first means obtains character type information for each character appearing in a word candidate by referring to a character structure database in which component character layout information is stored in advance for each character code. The word recognition device according to claim 4. In a word recognition device that performs word recognition based on a character recognition result,
  First means for determining whether to perform character structure recognition based on the character recognition result;
  When it is determined that the character structure recognition is performed by the first means, based on the character recognition result, for each character appearing in the word candidate, a spatial layout position of separable component characters constituting the character and A second means for determining a character type consisting of the number of characters of the component parts and obtaining character type information;
  Third means for performing word recognition processing using the character type information acquired by the second means;
  Fourth means for outputting the character recognition result as it is as a word recognition result when it is determined by the first means that character structure recognition is not performed;
  A word recognition device characterized by comprising:

Images