JPH0740287B2 - Pattern recognition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a method for recognizing a pattern such as a character, and more particularly, to a pattern recognition method using a multi-directional histogram method.

[Prior art]

The applicants of the present invention are Japanese Patent Application No. 59-202822 and Japanese Patent Application No. 58-2028.
No. 25 and others have already proposed a pattern recognition method using the multi-direction histogram method. The present invention relates to an improvement of such a pattern recognition method.

In this pattern recognition method using the multilayer direction histogram method, a direction code is attached to the contour pixels of a pattern such as a character, and the pattern is scanned from each side of the frame of the pattern toward the opposite side, and white pixels (background) are formed. The next appearing directional code is detected and the directional code is stratified into multiple layers depending on how many times it was detected on the scan line.
Then, for each of the divided areas within the frame of the pattern, a histogram of direction codes of layers up to a certain layer is obtained, and a vector that is each histogram component (feature amount) is used as a feature vector of the pattern.

For example, if 8 kinds of codes are added as direction codes, the mesh is divided into 4 × 4 mesh areas in the pattern frame, and the direction codes of the first layer and the second layer are also 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 as the same pattern, and the average thereof is registered as the feature vector of the dictionary pattern (standard pattern).

Further, the present applicant has already proposed a similar pattern recognition method in which the direction code is further stratified in consideration of the scanning direction for stratification. The present invention can be applied to this pattern recognition method as well.

Further, the method of dividing the area within the pattern frame is not limited to the method disclosed in the specification and drawings of the above-mentioned prior application. For example,
Similar to the pattern recognition method of the prior application, the pattern frame is divided into meshes so that the direction codes are evenly distributed,
A method may be adopted in which the mesh areas are partially overlapped according to preset parameters and integrated into a small area. The pattern recognition method by the multilayer direction histogram method adopting such a region division method is
The present invention has been proposed by the present applicant, but the present invention can be similarly applied to such a pattern recognition method.

Now, in such a pattern recognition method by the multi-direction histogram method, the unknown pattern and the dictionary pattern are calculated by calculating the distance or similarity of the corresponding dimension component between the feature vector extracted from the unknown pattern and the feature vector of the dictionary pattern. Is performed, and the dictionary pattern having the smallest sum of distances or the dictionary pattern having the largest sum of similarities is set as the recognition result.

However, as described above, when the dimension number of the feature vector becomes large, there is a problem that the calculation amount of the distance or the similarity becomes large, the matching time becomes long, and the dictionary capacity becomes large.

〔Purpose〕

Therefore, an object of the present invention is to improve the efficiency of matching and delete the dictionary capacity in the pattern recognition method by the multi-layered histogram method.

〔Constitution〕

The feature vector obtained by the multi-layered histogram method has a dimensional component having a large pattern identification effect and a dimensional component having a less significant effect. 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 the dictionary patterns of “sentence”, “letter”, “recognition”, and “knowledge” are g ₁ and g in FIG. 5, respectively.
_It looks like ₂ , g ₃ , g ₄ . In this example, as is clear from the figure, the component (feature amount) A of each feature vector has a larger variance (or standard deviation) than the component (feature amount) B. In other words, the component A has a higher ability to identify an unknown pattern.

The matching between the unknown pattern and the dictionary pattern is basically a dictionary in which the distance or similarity between the corresponding dimension components of the feature vectors of the unknown pattern and dictionary pattern is calculated, and the sum of the distances is the minimum or the similarity is the maximum. The pattern is used as the recognition result. Here, paying attention to the characteristics of the feature vector as described above, by preferentially calculating the distance or the similarity from the portion having a high pattern identification ability,
It may be possible to exclude dictionary patterns that cannot be candidates and narrow down dictionary patterns that can be candidates at an early stage when the distance or the similarity is calculated for some components having high pattern identification ability.

Also, the feature vector by the multi-direction histogram method is
The characteristics of the pattern are preserved even if the order of the respective dimensional components is changed.

Focusing on the above points, in the present invention, when creating a dictionary, the standard deviation or variance for each dimension of the feature vector of the dictionary pattern is obtained, and the component of the feature vector of each dictionary pattern is set to the standard value or variance. Are sorted in descending order, and the feature vector after such sorting is registered in the dictionary.

For example, it is assumed that the feature vector of a certain dictionary pattern created by the multi-direction histogram method is as shown in FIG. 6 (a). Then, the order of large standard deviation or variance calculated for all standard patterns is X ₄ , X ₁ , X ₃ , X ₇ ,
X ₅ , X ₂ , X ₈ , X ₆ , ... Then, the feature vector is registered in the dictionary by rearranging the components of the respective dimensions X ₁ , X ₂ , X ₃ , ... As shown in FIG. 6 (b).
That is rearranged as component dimension Y ₁ of the uppermost feature vectors reordered components other dimensions X ₄ of the original feature vector, the components of the dimension Y ₂ components of dimension X ₁ is next order, that .

Then, the feature vector extracted from the unknown pattern is
After rearranging the components according to the rearrangement order of the components of the feature vector of the dictionary pattern, by calculating the distance or similarity with the corresponding dimension component of the feature vector of the dictionary pattern, the unknown pattern and the dictionary pattern Match.

Here, in the case of recognizing many kinds of patterns such as handwritten Chinese characters, it is necessary to perform detailed matching using a feature vector having a large number of dimensions. However, when recognizing a pattern such as a small number of characters such as ANSK characters, the dictionary pattern and unknown pattern components are sorted in descending order of standard pattern deviation or variance, that is, in descending order of pattern identification ability, as described above. If rearranged, the number of dimensions in the upper rank is relatively small (for ANSK characters, for example, 20 dimensions, 24
It was confirmed that a sufficient recognition rate can be achieved by matching only the dimensions.

Focusing on this point, in the present invention, as described above, only the upper N dimensions of the feature vector of the dictionary pattern after the dimension rearrangement are left, and the vector having the lower dimensions is finally converted into the dictionary. By registering the pattern feature vector in the dictionary, the dictionary capacity is deleted and the matching efficiency is further improved.

〔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. It should be noted that in this embodiment, a relatively small type of character pattern such as ANSK characters is assumed as the pattern to be recognized.

In the figure, 10 is a pattern 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 character cutting and normalization of the 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 multi-layered direction histogram method.

Here, in this embodiment, the operation modes include a dictionary creation mode and a pattern recognition mode. 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.

To create the dictionary, M character patterns for each character type are sequentially input from the pattern reading unit 10 (step 102). The input character pattern is pre-processed by the pre-processing unit 12 (step 102) and is input to the feature extraction unit 14 to receive a feature vector (eg
A 256-dimensional vector) is extracted (step 104). The extracted feature vector is input to the rearrangement unit 16.

Reference numeral 18 denotes a rearrangement table unit referred to by the rearrangement unit 16, in which a rearrangement table created in advance is registered.

This rearrangement table is created as follows. For each character type, a feature vector (for example, a 256-dimensional vector) is extracted from a plurality of character patterns by the multi-direction histogram method, and the average thereof is used as the feature vector of the standard pattern of the character type. The standard deviation δ _kn for each dimension n of the feature vector of the standard pattern of the preceding character type (K type as a whole) thus obtained is Calculate by Here, k is the character type number and g
_kn is a component (feature amount) of dimension n of character type k, and
_kn is the average of all character types of the component of dimension n.

Instead of the standard deviation, the squared variance may be obtained.

The dimension numbers are arranged in descending order of the standard deviation or variance calculated in this way, and a correspondence table of the dimension numbers and the corresponding original dimension numbers is created up to the upper N dimensions. This correspondence table is the rearrangement table.

Return to the description of the dictionary creation mode. In the sorting section 16,
The feature vector g _knm extracted from the input character pattern is rearranged according to the rearrangement table of the rearrangement table unit 18 in each dimensional component, and the feature vector g _kn consisting of rearranged upper N-dimensional components is the dictionary creation unit 20. (Step 106).

In this way, when the N-dimensional feature vector after the dimension rearrangement for the M character patterns of the same character weave is accumulated in the dictionary creating unit 20, the dictionary creating unit 20 averages the M feature vectors. Is _calculated , and is added to the feature vector of the dictionary pattern of the character type k (N-dimensional vector _kn) , the character code is added, and registered in the dictionary 22 (step 108).

Next, the character type number k is updated (step 110), the process returns to step 100, and a similar dictionary creating process is executed for the next character type.

When the process is executed up to the last character type (k = k), it is determined in step 112 that the process is completed, and the process is completed.

Next, the case of the pattern 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 character pattern to be recognized (unknown character pattern) is input from the pattern reading unit 10 (step 200) and subjected to preprocessing by the preprocessing unit 12 (step 202), and the feature extraction unit 1
4 and the feature vector X _n (for example, a 256-dimensional vector) is extracted by the multilayer histogram method (step 204).

The feature vector X _n is input to the rearrangement unit 16, the components are rearranged in the descending order of standard deviation or variance according to the rearrangement table, and converted into the feature vector Y _n including the upper N-dimensional components, and then the matching is performed. It is sent to section 4 (step 206).

In the matching section 24, matching between the unknown input pattern and the dictionary pattern is performed as follows (step 208).

Feature vector _kn of dictionary pattern of each character type k
And the distance Dk between the unknown character pattern feature vector Y _n and Calculate by Then sort that distance,
The dictionary pattern with the smallest distance is determined as a candidate character.

The character code of this candidate character is output as a recognition result for the unknown character pattern (step 210).

As described above, in this embodiment, the feature vector of the dictionary pattern has the components rearranged in the order of large standard deviation or variance, that is, in the descending order of pattern discriminating ability, and discards the components having low pattern discriminating ability other than the upper N dimensions. Are registered in the dictionary in the form of a list, and the components of the feature vector of the unknown input pattern are rearranged in the same order as the dictionary pattern, and the matching between the dictionary pattern and the unknown character pattern is the distance or similarity between N-dimensional vectors. Is performed by calculation.

Therefore, it is possible to achieve a high matching efficiency with a small amount of calculation of the distance or the degree of similarity, and moreover, it is possible to achieve a sufficient classification rate because a dimension having a high pattern recognition capability is used. Furthermore, due to the reduction in the number of dimensions, a significant reduction in dictionary capacity can be achieved.

Next, another embodiment of the present invention will be described. The overall functional configuration of this embodiment is the same as that of the previous embodiment, and the matching in the matching unit 24 is only partially different. Therefore, only the processing of the matching unit 24 will be described with reference to the flowchart of FIG.

When the N-dimensional feature vector of the unknown character pattern after the rearrangement of the components is input to the matching, the character type, that is, the number of the dictionary pattern is set to 1 (step 300), and the matching between the dictionary pattern and the unknown character pattern is performed. Be seen.

First, the sum d ₁ of the distances between the feature vector f _{n of the} dictionary pattern and the feature vector Y _n of the unknown character pattern is calculated for the upper N ₁ dimensions (N ₁ <N) (step 30).
2). Then, a comparison judgment is made between the sum d _{1 of} the distances and the threshold value Th ₁ (step 304).

If d ₁ > Th ₁ , the dictionary pattern is too large in distance to be a candidate pattern (candidate character), so the matching is divided at this stage and the dictionary pattern number k is incremented (step 306). , Return to step 306.

That is, the unknown character patterns are largely classified (candidate patterns are narrowed down) by calculating the distance to the upper N _1- dimensional dimension, and the matching of the dictionary patterns excluded here is completed at this stage.

If d ₁ ≦ Th ₁ in step 304, the sum d ₂ of the distances to all dimensions N is obtained (step 310). And
The distance d ₂ is compared with the distance between the candidate pattern and the unknown character pattern immediately before that distance, the smaller candidate pattern and the distance are stored (step 312), and the process returns to step 302 via step 206.

When the matching is completed up to the final dictionary pattern, it is determined to be completed in step 308, the character code of the finally remaining candidate pattern is output (step 314), and the unknown character pattern recognition processing is completed.

As described above, in this embodiment, the distance is preferentially calculated from the upper dimension of the feature vector, and the dictionary pattern that cannot be the candidate pattern is eliminated early according to the calculation result up to a certain dimension. Only the distance calculation of all dimensions is performed and detailed matching is performed. Therefore, useless distance calculation is reduced as compared with the above embodiment, and the matching efficiency is further improved.

Such stepwise indentation of the candidate pattern may be performed in two or more steps. For example, for each dictionary pattern, the unknown character patterns are roughly classified by calculating the distance to the upper N ₁ (that is, N ₁ <N) dimensions. If not excluded, then the top N for that dictionary pattern
₂ (That is, N ₁ <N ₂ <N) The distance to the dimension is calculated and the unknown character pattern is classified into the middle. If not excluded even by this middle classification, the unknown character patterns are subclassified by distance calculation up to the top _three (thus N ₂ <N ₃ <N) dimensions. If this small classification is not excluded, detailed matching of unknown character patterns is performed by distance calculation for all dimensions.

By doing so, the matching calculation with the dictionary pattern that cannot be a candidate can be stopped earlier, and the matching time can be further shortened.

In each of the above embodiments, the distance is used for matching the unknown pattern and the dictionary pattern, but it goes without saying that similar matching processing may be performed by obtaining the similarity.

Further, the present invention is not limited to character patterns, but can be similarly applied to recognition of general patterns such as voices.

[Effect]

As is clear from the above description, according to the present invention, it is possible to significantly improve the matching efficiency by the multi-direction histogram method and shorten the pattern recognition time, and at the same time, it is possible to significantly reduce the dictionary capacity. can get.

[Brief description of drawings]

FIG. 1 is a simplified block diagram showing a simplified functional configuration of an embodiment of the present invention, FIG. 2 is a schematic flowchart of a dictionary creating process in the same embodiment, and FIG. 3 is a pattern recognition process in the same embodiment. FIG. 4 is a schematic flowchart of the matching process in the embodiment of the method of the present invention, FIG. 5 is a vector diagram for explaining the characteristics of the feature vector in the multi-directional histogram method, and FIG. 6 is a component of the feature vector. It is an explanatory view of rearrangement. 10 …… Pattern reading section, 12 …… Preprocessing section, 14 …… Feature extraction section, 16 …… Sorting section, 18 …… Sorting table section, 20 …… Dictionary creating section, 22 …… Dictionary.

Claims (2)

[Claims] 1. A pattern recognition method using the multi-directional histogram method, wherein components of a feature vector of a standard pattern obtained by the multi-directional histogram method are rearranged in advance in descending order of standard deviation or variance, and then the upper N-dimensional components thereof are arranged. A vector consisting of only the dictionary is registered in the dictionary as a feature vector of the dictionary pattern, and the components of the feature vector extracted from the unknown pattern by the multilayer histogram method are rearranged according to the rearrangement order of the components of the feature vector of the dictionary pattern. A pattern recognition method characterized by performing matching between an unknown pattern and a dictionary pattern by calculating a distance or a similarity between corresponding dimension components of the feature vector after the component rearrangement and the feature vector of the dictionary pattern. 2. A feature vector distance or similarity calculation is preferentially performed from an upper dimensional component, and whether the distance or similarity calculation of a lower dimensional component is performed is determined based on the result of the calculation. The pattern recognition method according to claim 1, wherein