JPH0795336B2 - Character recognition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a recognition method for characters such as handwritten Chinese characters, and more particularly to a character recognition method using a feature vector based on a multi-direction histogram method.

[Prior art]

The applicant has already proposed a character recognition method based on the multilayer directional histogram method. The present invention relates to the improvement of such a character recognition method.

In the character recognition method using the multi-layered directional histogram method, a direction code is attached to the character line contour pixel of the character pattern, and the character pattern is scanned from each side of the character toward the opposite side and next to the white pixel (background). The directional code appearing in is detected, and the directional code is classified into a plurality of layers according to the number of the detected directional code on the scan line. Then, for each divided area in the character frame, a histogram for each direction code is obtained for each layer up to a certain layer, and a vector having each histogram (feature amount) as a component is used as a feature vector of the character.

In general, each component (feature amount) of the feature vector is represented by a dimension. For example, attach 8 kinds of codes as direction codes,
When the character frame is divided into 4 × 4 mesh regions and the direction codes of the first layer and the second layer are targeted, the dimension number of the feature vector is 256 (= 4 × 4 × 2 × 8).

Regarding the dictionary, similar feature vectors are extracted from a plurality of patterns for each character type, and the average thereof is registered as the feature vector of the dictionary pattern (standard pattern).

Further, the applicant has already proposed a similar character recognition method in which the direction code is stratified in more detail in consideration of the scanning direction for stratification. The present invention can be similarly applied to this character recognition system.

In the character recognition method based on the multi-layered histogram method, the distance between the feature vector extracted from the input character and the feature vector of the dictionary is calculated to identify the character type of the input character.

However, as described above, when the dimension number of the feature vector is large, there is a problem that the distance calculation is large and the matching processing time is long.

〔Purpose〕

Therefore, an object of the present invention is to improve the efficiency of the matching process in the character recognition method based on the multi-direction histogram method.

Another object of the present invention is to speed up the matching process and improve the recognition rate of "blurred" characters in the character recognition method based on the multi-layered histogram method.

〔Constitution〕

The feature vector based on the multi-direction histogram method has a component of a dimension having a large character recognition effect and a component of a dimension in which the effect is not so remarkable. In this regard, consider a two-dimensional feature vector for ease of explanation.

When a dictionary is created by the multi-directional histogram method with the number of dimensions being 2, the feature vectors of “sentence”, “letter”, “recognition”, and “knowledge” are respectively g ₁ , g ₂ , g in FIG. It becomes like ₃ and g ₄ . In this example, as is clear from the figure, the component (feature amount) A of each feature vector is the component (feature amount)
The variance (or standard deviation) is larger than B. In other words, the component A has a higher identification ability for unknown characters.

Focusing on the property of such a feature vector, by preferentially calculating the distance from a component having a high character recognition ability, it cannot be a candidate when the distance calculation is executed up to a part of the dimensions having a high recognition ability. It will be possible to eliminate character types and narrow down possible character types at an early stage.

In addition, the feature vector by the multi-direction histogram method is
Even if the order of the respective dimensional components is exchanged, the characteristic of the pattern is preserved.

Focusing on the above points, in the present invention, the feature vector of the temporary dictionary pattern of all the character types is obtained by the multilayer directional histogram method, and the component of the feature vector has the standard deviation (or deviation) or large variance. A vector rearranged in the order of dimensions is registered in the dictionary as a feature vector of the dictionary pattern.

Then, character recognition is performed as follows. A feature vector is extracted from the input character by the multi-direction histogram method, and the distance calculation between the vector in which the components are rearranged in order of the standard deviation or the order of the variance and the feature vector of the dictionary pattern of each pattern type has a higher rank. The distance is compared with the determination threshold value corresponding to the character type, and it is determined whether the distance calculation for the character type is interrupted or whether detailed matching is performed up to the lower dimension according to the comparison result.

Further, the determination threshold value is determined for each character type as follows, for example. Feature vectors are extracted from multiple patterns for each character type using the multi-direction histogram method,
The standard deviation (deviation) or the variance of the distance to a certain upper dimension of the vector in which the components are arranged in the order of the larger dimension of the standard deviation (deviation) or the variance and the feature vector of the dictionary pattern of the same character type, The determination threshold is determined based on the value.

According to such a character recognition method, the matching process for a character type that cannot be a candidate for an input character is interrupted early, so that the efficiency of the matching process is improved and the recognition speed is increased. In addition, since the threshold for determining the interruption of the matching process is determined for the character type as described above, it is possible to reliably prevent an error in which the matching process of the character type that should not be excluded from the candidates is interrupted by mistake. A high recognition rate can be achieved without losing the essential advantages of the recognition method.

〔Example〕

 Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic block diagram showing a simplified functional configuration of an embodiment of the present invention.

In the figure, 10 is a reading unit for reading a character pattern from a document and inputting the character pattern information to the preprocessing unit 12. The preprocessing unit 12 is a unit that performs preprocessing such as clipping and normalization of an input character pattern, and inputs the processed character pattern to the feature extraction unit 14 for each character.

The feature extraction unit 14 is a unit that extracts a feature vector from the input character pattern by the multilayer directional histogram method.

Here, in this embodiment, there are a dictionary creation mode and a character recognition mode as operation modes. First, the case of the dictionary creation mode will be described below. A schematic flow chart of the dictionary creating process in this mode is shown in FIG. 2, and corresponding step numbers are shown in parentheses in the following description.

For one character type, M character patterns are sequentially input from the reading unit 10 (step 50). The input character pattern is preprocessed by the preprocessing unit 12 (step 52),
The feature vector is input to the feature extraction unit 14 and is a feature vector obtained by the multi-directional histogram method (in this case, a 256-dimensional vector)
Are extracted (step 54). The extracted feature vector is input to the dictionary creating unit 20 via the rearrangement unit 16. The dictionary creation unit 20 obtains an average vector of the feature vectors extracted from the M input character patterns (step 56), and the average vector is temporarily stored in the dictionary 22 as the feature vector of the temporary dictionary pattern of the character type. It is registered (step 58).

The dictionary creating unit 20 checks whether the processing has been completed up to the last character type (step 60), and if there is an unprocessed character type, the process returns to step 51 and the same processing is performed for the next character type.

When the processing is completed for all the character types, the rearrangement table creation unit 26 calculates the standard deviation (or deviation) or variance for each dimension of the feature vectors of all the temporary dictionary patterns temporarily registered in the dictionary 22 ( Step 62). Then, a table of information (sorting table) for sorting the original feature vectors in order of the standard deviation (deviation) or the dimension having the largest variance is created and registered in the sorting table unit 18 (step 64).

Up to this point, this is the processing stage for creating the rearrangement table, after which the actual dictionary creation processing begins.

For one character type k, M character patterns are sequentially input from the reading unit 10 (step 66). The input character pattern is preprocessed by the preprocessing unit 12 (step 6
8) Then, the feature vector is input to the feature extraction unit 14 and the feature vector is extracted by the multi-direction histogram method (step 70).

The extracted feature vector is rearranged by the rearrangement unit 16 according to the information of the rearrangement table registered in the rearrangement table unit 18, in which the components are rearranged in the order of the standard deviation (deviation) or the larger variance (step 72). ). In this way, the M feature vectors after the rearrangement of the components are sent to the dictionary creating unit 20, and their average vector is main-registered in the dictionary 20 as the feature vector of the dictionary pattern of the character type k (step 74). .

The dictionary creation unit 20 checks whether or not the dictionary creation processing has been completed for all the character types (K types) (step 76), and if there are unprocessed character types, the same processing from step 66 is performed for the unprocessed character types. Is done.

When the processing is completed for all the character types, the dictionary itself is finished, but the processing for determining the judgment threshold value is performed next.

In this process, M character patterns similar to those used to create the dictionary for each character type are input (step 7).
8) After pre-processing (step 80), feature vectors by the multi-direction histogram method are extracted (step 82). The feature vectors of the M character patterns are respectively sorted by the sorting unit 16 in the sorting table. The components are rearranged according to the information (step 84), and the matching unit
Entered in 24.

Next, by starting the matching unit 24 and the threshold value determining unit 28,
A threshold decision is made. That is, the matching unit 24 calculates the distance between each feature vector and the feature vector of the dictionary pattern of the character type k up to the upper N dimensions, and the threshold value determination unit 28 calculates the standard deviation of the distance from the calculation result to the upper N dimensions. Is required (step 86). Then, the standard value is used as the determination threshold of the character value k in the threshold table unit
Is registered (step 88).

Similarly, determination thresholds for other character types are determined and registered in the threshold table unit 30.

When it is determined that the thresholds have been determined and registered for all character types (step 90), the entire processing in the dictionary creation mode ends.

Next, the case of the character recognition mode will be described below. Further, a schematic flowchart of the processing in this mode is shown in FIG. 3, and the corresponding step numbers are shown in parentheses in the following description.

The unknown character pattern to be recognized is input from the reading unit 10 (step 100), pre-processed by the pre-processing unit 12 (step 102), and then the feature vector ( A 256-dimensional vector) is extracted (step 104).

The rearrangement unit 16 rearranges the components of the feature vector according to the information in the rearrangement table, and the rearranged feature vector Yn is input to the matching unit 24 (step 106).

In the matching unit 24, the matching distance d between the feature vector Fkn of the dictionary pattern of the character type k and the feature vector Yn of the input character to the upper N dimensions is calculated (step
108). Then, the distance d is compared with the determination threshold Thk of the corresponding character type k registered in the threshold table unit 30 (step 110).

If d> Thk, there is no possibility that the character type of the input character is the current character type k, so the matching process with the dictionary pattern is interrupted and an end determination is made (step 11).
6).

If d ≦ Thk, the current character type is highly likely to be the character type of the input character, and therefore detailed matching is performed. That is, the matching distance D is calculated for all dimensions (step 112). Then, the distance D is compared with the distances of the candidates so far, and the smaller the distance, the character type is left as a new candidate (step 114). The process proceeds to step 116 of end determination.

When it is determined in step 116 that the character type to be matched remains, the step 10 for the next character type is performed.
Matching processing after 8 is performed.

When matching processing is performed for all character types,
In step 116, it is determined that the process has ended, and the character code of the finally remaining candidate character type is output as the recognition result. This completes the recognition process for one input character.

Now, the deformation, tendency, and degree of characters such as handwritten Chinese characters vary depending on the writer (user of the recognition system). In order to deal with this, character patterns considering various character transformations are prepared and used for dictionary creation.

However, even in such a case, the matching between the handwritten character pattern and the dictionary pattern is good or bad depending on the user. Also, even the same user tends to improve the matching property with the dictionary and increase the recognition rate as the user gets used to the recognition system.

Focusing on such a point, it is convenient that the threshold for determining whether or not to perform the detailed matching can be changed. That is, in the case of a user who has a good matching with the dictionary or a user who is accustomed to it, by lowering the determination threshold value and reducing the frequency of detailed matching, without lowering the recognition rate,
The recognition speed can be increased.

Next, the second aspect of the present invention that enables such a threshold adjustment
Examples will be described. Since the functional configuration of this embodiment is the same as that of the previous embodiment, its block diagram is omitted, and reference will be made to FIG. 1 as necessary in the following description.

In this embodiment, in the dictionary creation mode, the sorting table and the dictionary are created as in the previous embodiments. However, a plurality of thresholds for determining whether or not to perform detailed matching are determined for each character type.

This point will be described with reference to the flowchart of FIG. 2 related to the first embodiment. If the threshold value determined in step 88 is Thk, in this embodiment, Thk / 1, The values of Thk / 2 to Thk / L (L is a positive integer) are determined by the threshold determination unit 28 as the thresholds Thk (1) and Thk (2) to Thk (L) of the character type k, respectively, and the threshold table unit 30. Be registered with.

FIG. 4 is a schematic flow chart of the pattern recognition mode in this embodiment. The pattern recognition mode in this embodiment will be described with reference to this drawing.

In this embodiment, the user level UL (= 1, 2, 3, ..., L), which is a parameter for selecting the threshold value, is set in the first step 200. The user level UL can be specified by the user of the pattern recognition system according to this embodiment.

Steps 202 to 210 are the steps shown in FIG.
The contents are the same as 100 to step 108. Step 21
Steps 4 to 220 have the same contents as steps 112 to 118 in FIG.

Step 212 is a step for determining whether or not to perform detailed matching from the matching distance to the upper N dimensions, and corresponds to step 110 in FIG. In the case of this embodiment, one threshold value Th (UL) corresponding to the user level UL is selected from a plurality of threshold values Th (1) to Th (L) of the character type k.
Is used as a threshold for the determination.

That is, in the case of a user who is familiar with the pattern recognition system according to this embodiment, if the user level UL is designated as a high level (the highest is L), a smaller threshold value (the lowest is Th.
(L)) is selected and used for the determination, the frequency of proceeding to the detailed matching (step 214) is reduced, the matching efficiency is improved and the character recognition speed is increased accordingly.

On the other hand, if the user is unfamiliar or has poor matching with the dictionary, if a low user level UL (lowest is 1) is specified, a large threshold value (highest is Thk (1)) is selected and judged. Since the frequency of detailed matching increases, the recognition speed decreases but the recognition rate can be increased.

As described above, the recognition rate varies depending on the user, but also varies depending on the character type. This is remarkable when a pattern with a certain habit is used as a pattern for dictionary creation, but such a point is difficult to check at the dictionary creation stage when there are many character types such as Kanji. is there.

It is possible to absorb the variation in the recognition rate for each such character type. The third embodiment of the present invention will be described below.
The functional configuration of the pattern recognition system according to this embodiment is the same as that of the first embodiment. The processing in the dictionary creation mode is similar to that in the second embodiment, and a plurality of types of determination thresholds Th (1) to Th (L) are determined for each character type.

FIG. 5 is a schematic flowchart of the processing in the pattern recognition mode in this embodiment. In this figure, steps 300 to 308 have the same contents as steps 202 to 210 in FIG. 4, and steps 314 to 320 have the same contents as steps 214 to 220 in FIG.

In step 310, the level LV (k) for each character type is set as the threshold selection parameter V. This level LV
(K) is given as a value obtained by storing the reject / recognition rate Rk of the character type k during operation of the pattern recognition system and dividing the Pk value by a certain constant.

Then, step 312 is a step corresponding to step 212 in FIG. 4, and is a plurality of thresholds Thk (1) to Thk of the character type k.
One threshold Th corresponding to the parameter V from (L)
It is determined whether or not to proceed to the detailed matching (step 314) by selecting k (V) and comparing it with the distance d.

A fourth embodiment, which is a combination of this embodiment and the second embodiment, will be described with reference to the flowchart of FIG. 5 for convenience.

In this embodiment, a user level UL setting step (corresponding to step 200 in FIG. 4) is added before step 300. In step 310, the sum of the level LV (k) of the character type k and the user level UL is the parameter V.
Is set as. Then, in step 312, the threshold Thk (V) corresponding to the parameter V is used for the determination.

Other than this, it is the same as each of the above-mentioned embodiments.

According to this embodiment, since it is possible to absorb the variation factors depending on the user and the character type and determine whether or not to perform the detailed matching using the optimum threshold value, the second and third embodiments described above. Both advantages are obtained.

By the way, complicated characters such as kanji are apt to be "crushed" regardless of whether they are printed by handprint or handwritten. According to the inventor's research, the pixels of the first layer are not affected by the "blurred" character, and the feature amount of the first layer is stable even in the case of the "blurred" character, It was found to be effective in recognizing.

Based on this point, another embodiment in which a high recognition rate can be achieved even for "blurred" characters and efficient recognition processing is possible will be described below.

Since the functional structure of this embodiment is almost the same as that of each of the above-mentioned embodiments, it will be described with reference to the block diagram of FIG.

In this embodiment, a collapse determination unit 32 is added as shown by the broken line in FIG. The crush determination unit 32 measures the ratio Rc between the perimeter length Sc of the input character and the number of black pixels Tb, and calculates the ratio.
If the Rc (degradation degree) is greater than or equal to a predetermined threshold value, the input character is judged to be a normal character, and if it falls below that threshold value,
Determine the input character as a "blank" character. The determination result is given to the rearrangement unit 16 and the matching unit 24.

It should be noted that the algorithm for determining the collapse may be changed as appropriate.

Further, in this embodiment, the first rearrangement table for normal characters and the second rearrangement table for collapsed characters are created at the time of dictionary creation and are registered in the rearrangement table unit 18, respectively.

In the first rearrangement table for normal characters, the feature vectors extracted from the input characters are matched with the component order after rearrangement of the feature vectors of the dictionary, and the standard deviations of the feature amounts of the first layer and the second layer are set. Alternatively, information for sorting in the order of larger deviation or variance is stored. Information for rearranging the feature vectors extracted from the input characters in the second rearrangement table for the collapsed characters with respect to the feature amount of the first layer Is stored.

In the character recognition mode, when the input character is determined to be a normal character, the feature vector extracted from the input character is
The rearrangement unit 16 rearranges the components with respect to the characteristic amounts of the first layer and the second layer according to the information in the first rearrangement table, and then sends the components to the matching unit 24. On the other hand, when the input character is determined to be a “blurred” character, the extracted feature vector is rearranged in the component amount only with respect to the feature amount of the first layer according to the information of the second rearrangement table.

The matching unit 24 calculates the distance between the rearranged input character feature vector and the dictionary feature vector, but in the case of normal characters, up to a maximum of 256 dimensions regarding the feature amounts of the first and second layers. Calculates the distance, while
In the case of "blank" characters, it is the highest regarding the feature amount of the first layer
Calculates distance up to 128 dimensions. That is, for the “blurred” character, the distance calculation is performed only for the feature amount of the first layer that is not easily affected by the “blurred”, and the effective dimension is limited.

In any case, as in the above-described embodiment, the distance to a certain higher-order dimension is compared with the determination threshold, and it is determined whether the distance calculation should be stopped or continued.

According to this embodiment, a high recognition rate can be achieved even for "blurred" characters, and the distance calculation amount can be reduced.
It is possible to further improve the efficiency of the matching process.

Although the present invention has been described above with reference to the embodiments, the present invention is not limited thereto and various modifications are allowed.

〔effect〕

As is clear from the above description, according to the present invention, the efficiency of the matching process is significantly improved, high-speed character recognition is possible, and the collapsed characters are A high recognition rate can also be achieved.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram relating to the first embodiment and other embodiments of the present invention, FIG. 2 is a schematic flowchart of a dictionary creating process in the first embodiment, and FIG. FIG. 4 is a schematic flowchart of character recognition processing in the first embodiment, FIG. 4 is a schematic flowchart of character recognition processing in the second embodiment of the present invention, and FIG. 5 is a schematic flowchart of character recognition processing in the third embodiment of the present invention. FIG. 6 is a vector diagram for explaining the property of the feature vector of the multi-layered histogram method. 10 ... reading unit, 14 ... feature extracting unit, 16 ... sorting unit, 18
… Sort table section, 22… Dictionary, 24… Matching section,
30 ... Threshold table section.

Claims (2)

[Claims] 1. A character recognition method using a feature vector according to a multi-directional histogram method, comprising a dictionary for storing feature vectors of various characters in which components are rearranged in order of large standard deviation or deviation or variance, and the rearrangement. Table for storing information for a character type, a threshold value table for storing a judgment threshold value corresponding to a character type, a means for extracting a feature vector from an input character, and a rearrangement table for the components of the extracted feature vector. A means for rearranging the standard deviation or the deviation or the variance in descending order according to the stored information, and the rearranged feature vector and the feature vector stored in the dictionary. Distance calculation is performed only on the characteristic component, and the distance is stored in the threshold table. Only when it is less than or equal to the character type determination threshold value, a means for recognizing an input character by performing a distance calculation only on a part of characteristic components added with a characteristic component having a smaller standard deviation or deviation or variance is provided. Character recognition method. 2. A character recognition method using a feature vector by a multi-layered directional histogram method, which stores a feature vector of various characters in which components are rearranged in order of large standard deviation or deviation or variance, and a first layer. And the second
A first rearrangement table that stores information for rearranging the components of the feature vector in descending order of standard deviation or deviation or variance with respect to the layer feature amount, and large standard deviation, deviation or variance with respect to the first layer feature amount. A second rearrangement table that stores information for rearranging the components of the feature vector in order, a threshold table that stores the determination threshold corresponding to the character type, a feature extraction unit that extracts the feature vector from the input character, and an input. Determination means for determining whether the character is a normal character or a collapsed character, and the components of the feature vector extracted from the input character determined to be a normal character by the determination are arranged according to the storage information of the first rearrangement table. In other words, the components of the feature vector extracted from the input character determined to be a collapsed character by the determination are subjected to the second rearrangement. Means for rearranging the input characters after the rearrangement, and the feature vector stored in the dictionary and the feature vector of the input characters after the sort, the first layer and the first layer With respect to the feature amounts of the two layers, and in the case of the input character determined to be a collapsed character, the distance calculation is performed only for the feature amount of the standard deviation or the deviation or the large variance for the feature amount of the first layer, Only when the distance is equal to or less than the judgment threshold value of the corresponding character type stored in the threshold value data, the distance calculation is performed only on a part of the characteristic components to which the characteristic components having the standard deviation or the smaller deviation or variance are added, and the input character is input. And a means for recognizing characters.