JP3730073B2 - Template creation method, apparatus, and recording medium recording template creation program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a pattern recognition method for calculating a similarity or distance value between an input pattern and a template created in advance using a training pattern in a learning pattern and recognizing the category of the input pattern.
[0002]
[Prior art]
Hereinafter, character recognition will be described as an example of pattern recognition.
[0003]
2. Description of the Related Art Conventionally, in a character recognition device such as Kanji OCR (Optical Character Reader), a method of recognizing an input document or image pattern by pattern matching between an input pattern and a template is known. In general, this method is used to compare design-processed characters that are made up of unique textures found in advertisements and newspaper headlines, and images that are severely blurred, crushed, and background noise is significant. Has been reported to be robust.
[0004]
However, if the character shape changes due to image observation conditions such as lighting conditions, or if the character shape changes due to printing conditions or aging, etc., the similarity between the input pattern and the template decreases and the recognition rate decreases. There was a point.
[0005]
In order to solve such problems, conventionally, mainly, (1) a method of performing recognition using a template after removing other information in the input pattern as noise or correcting variations in the shape of the character Two approaches have been taken: (2) a method of absorbing the influence of deformation of individual characters by using a word template that uses the character string as a template pattern as it is.
[0006]
However, it is difficult to remove only other information in the input pattern. This is because character strokes and other information often occur very similar. As a result, other information often remains or the information of interest is lost. As a result, the degree of similarity at the time of matching with the template is lowered, and the recognition rate is lowered. In particular, small patterns such as “l” and “r” are often affected by other information in the surrounding area and are difficult to recognize.
[0007]
On the other hand, in the method using a word template, since multiple characters are collated, even if characters such as small characters fluctuate and partially misrecognized characters are partially included, the whole word can be recognized. Can be expected. However, it is necessary to create a template according to the number of words, and the number of combinations increases as the number of words increases. Along with this, the verification time also increases. Furthermore, when an unknown word is included in the input pattern, it cannot be recognized because there is no corresponding template.
[0008]
Further, another problem of the conventional method is that it cannot be followed when character variation or character shape variation occurs as usage proceeds. Any of the above-described methods proposed in the past cannot follow the difference in fluctuation for each input pattern, and cannot recognize patterns other than a specific fluctuation.
[0009]
[Problems to be solved by the invention]
As described above, the pattern recognition technology including the conventional OCR has been greatly deteriorated and deformed, and means for correctly recognizing a pattern that changes depending on the situation has not been sufficiently established.
[0010]
An object of the present invention is to provide a template creation method and apparatus for creating a template capable of obtaining a high recognition rate without manual work, and a recording medium on which a template creation program is recorded.
[0011]
[Means for Solving the Problems]
The template creation method of the present invention is a template creation method performed by a template creation device that creates a template to be matched with an input pattern using a training pattern in a learning pattern,
The training pattern extraction unit reads the learning pattern from the learning pattern storage unit that stores the learning pattern, reads the template from the template storage unit that stores a template created in advance for each category, Predetermined preprocessing is performed on the read learning pattern, and the preprocessed pattern is compared with the template to obtain the image position of the portion showing high similarity to the template and the category of the template. , A training pattern extraction stage for extracting a training pattern in the learning pattern from these,
The template selection unit obtains one or a plurality of training patterns similar to the extracted training pattern from the learning pattern storage unit, and when there is a category that occupies a majority of the categories of the similar training pattern, the extraction A template selection step of selecting the trained training pattern as a template and newly registering the template in the template storage means;
Have
[0012]
Since a template corresponding to the state of the input pattern is automatically and easily created, it is possible to automatically follow a change in the situation that has conventionally been difficult to recognize.
[0014]
According to another embodiment of the present invention, one or a plurality of similar patterns are obtained from the extracted set of training patterns, and whether or not to register a training pattern as a template is determined from the category distribution of the similar patterns. For example, if three similar training patterns are obtained for a certain training pattern (category 'e'), assuming that the categories become 'e', 'o', and 'c', this training pattern is classified into category 'e'. Do not select as' unreliable. On the other hand, if all of the similar training patterns are category “e”, it is determined that the reliability of the training pattern is high, and is selected as a template.
[0015]
According to another embodiment of the present invention, when selecting whether or not a certain training pattern is appropriate as a template of a given category, a training pattern similar to the training pattern is obtained, and according to the category distribution. Determine the category name. For example, if three similar training patterns are obtained for a certain training pattern (category 'e'), assuming that the categories are 'o', 'o', and 'o', this training pattern is classified into category 'e'. It is determined that the reliability of “is low and the possibility of being in the category“ o ”is high, and is selected as a template for the category“ o ”.
[0016]
The template creation device of the present invention is a template creation device that creates a template to be matched with an input pattern using a training pattern in a learning pattern,
Learning pattern storage means for storing a learning pattern;
Template storage means for storing templates created in advance for each category ;
Reads learning patterns from the learning pattern storage means, reads out the template from the template storage unit, performs a predetermined pre-treatment against the learning pattern read out, to the preprocessed pattern by performing matching with the template, the A training pattern extracting means for obtaining an image position of a portion showing a high similarity to the template and a category of the template, and extracting a training pattern in the learning pattern from them,
One or a plurality of training patterns similar to the extracted training pattern are obtained from the learning pattern storage means, and if there is a category that occupies a majority of the categories of the similar training pattern, the extracted training pattern is used as a template. And a template selection means for newly registering the template in the template storage means,
Having.
[0017]
DETAILED DESCRIPTION OF THE INVENTION
Next, embodiments of the present invention will be described with reference to the drawings.
[0018]
Referring to FIG. 1, a pattern recognition apparatus according to an embodiment of the present invention includes a learning pattern storage circuit 11, a training pattern extraction circuit 12, a template selection circuit 13, a template storage circuit 14, an input pattern storage circuit 15, and a recognition circuit 16. It is configured.
[0019]
The learning pattern storage circuit 11 stores a learning image which is a learning pattern used for creating a template. The learning image is composed of n (n ≧ 2) pixels, and the pixel values are represented by, for example, “0” for white pixels and “1” for black pixels.
[0020]
The training pattern extraction circuit 12 reads a learning pattern from the learning pattern storage circuit 11 and extracts a training pattern from the learning pattern. A category name is assigned to the training pattern.
[0021]
The template selection circuit 13 inputs a training pattern from the training pattern extraction circuit 12, selects an appropriate template, and newly registers it in the template storage circuit 14.
[0022]
The template storage circuit 14 stores one or a plurality (generally a plurality) of templates for each category used for recognition recognition of the input image pattern for each category.
[0023]
The input pattern storage circuit 15 stores one or more recognized image patterns as input patterns. These image patterns are composed of M (M ≧ 2) pixels. For example, in the case of a binary image pattern, each pixel is indicated by “0” or “1”.
[0024]
The recognition circuit 16 reads a template from the template storage circuit 14, reads an input pattern from the pattern storage circuit 15, performs pattern recognition by collating the input pattern with the template, and outputs the recognition result.
[0025]
A feature of this embodiment is that a recognition template is automatically created (added) using a learning pattern and an initial template. Hereinafter, the process of automatically creating a template from the learning pattern will be described in detail with reference to the flowchart of FIG.
[0026]
In step 21, the learning pattern is read from the learning pattern storage circuit 11 and input to the training pattern extraction circuit 12. It is assumed that the learning pattern includes one or a plurality of training patterns. An example of the learning pattern is shown in FIG.
[0027]
Next, in step 22, the training pattern extraction circuit 12 reads a template created in advance from the template storage circuit 14. For example, it is assumed that the “template created in advance” is created by manually cutting out one template for each category from the learning image.
[0028]
In step 23, the training pattern extraction circuit 12 recognizes the learning pattern using the read template. As a recognition method, for example, a preprocessed pattern is created by performing preprocessing such as tilt correction on the learning pattern, and a method based on shift matching using complementary similarity is performed on the preprocessed pattern (for example: Japanese Patent Application No. 9-343292) can be used. As a result of the recognition, an image position of a portion showing a high similarity and a category of the template are obtained. FIG. 3B shows a preprocessed pattern obtained by performing inclination correction on the learning pattern (FIG. 3A). Further, in FIG. 3 (3), when the upper row of FIG. 3 (2) is subjected to the shift collation using the template created in advance, the category showing the high similarity is the position showing the high similarity. Shown in
[0029]
Next, a training pattern is extracted using the position and category that showed a high degree of similarity in step 24. As an example of the training pattern extraction method, a rectangular region having a predetermined size is extracted with the recognized position as the center. The extracted training pattern is input to the template selection circuit 13.
[0030]
In step 25, the template selection circuit 13 selects a template using the input training pattern. As a template selection method, for example, a similar pattern is obtained from a plurality of training patterns, and when the majority of the category distribution of the similar patterns is the same, the training pattern is selected as a reliable template, and the majority is the same. If it is not, there is a method of dismissing. FIG. 4 shows an example of a rejected template. FIG. 4A shows a pattern that has been deleted by mistake, and FIG. 4B shows a pattern that has been deleted correctly. Moreover, the category name of the training pattern is determined using the similar pattern. For example, a category name that is a majority of the category distribution of the similar pattern is set as the category name of the training pattern.
[0031]
In step 26, the selected template is stored in the template storage circuit 14.
[0032]
Further, by repeating the steps 22 to 26 using the newly saved template and updating the template, it is possible to obtain a more accurate template and further follow changes in the situation.
[0033]
FIG. 5 shows the results when characters are recognized using the template thus obtained. (A) is an initial template, (b) is all training patterns (t = 1), (c) is a selected template (t = 1), (d) is all training patterns (t = 2), (E) shows the case of the selected dictionary (t = 2). t is the number of repetitions of steps 22 to 26 in FIG. It can be seen that there is almost no difference in the recognition rate when the learning pattern is recognized by the template (◯ mark) and when the other pattern is recognized by the template (□ mark).
[0034]
In the above description, characters are mainly shown, but the same processing can be performed for patterns other than characters.
[0035]
Referring to FIG. 6, the pattern recognition apparatus according to another embodiment of the present invention includes an input device 31, storage devices 32 to 35, an output device 36, a recording medium 37, and a data processing device 38.
[0036]
The input device is for inputting an input pattern. The storage devices 32, 33, and 34 correspond to the learning pattern storage circuit 11, the template storage circuit 14, and the input pattern storage circuit 15 in FIG. The storage device 35 is a hard disk. The output device 36 is for outputting the recognition result. The recording medium 37 is a floppy disk, CD-ROM, magneto-optical disk, or the like on which a pattern recognition program consisting of the processes of the training pattern extraction circuit 12, the template selection circuit 13, and the recognition circuit 16 in FIG. 1 is recorded. It is a recording medium. The data processing device 38 includes a CPU, an interface, etc., and reads the pattern recognition program from the recording medium 37 into the storage device 35 and executes it.
[0037]
【The invention's effect】
As described above, the present invention has the following effects.
[0038]
According to the first , second, and third aspects of the present invention, the learning pattern is recognized by a template created in advance, the training pattern is extracted from the result, and it is determined whether to register the training pattern as a template from the category distribution of similar patterns. By doing so, a template with a high recognition rate can be automatically created without any manual work.
[0040]
The inventions of claims 2 , 4 and 6 can be changed to correct category names even when an incorrect category name is given at the time of training pattern extraction.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration example of a pattern recognition apparatus according to an embodiment of the present invention.
FIG. 2 is a flowchart showing template creation in the pattern recognition apparatus of FIG. 1;
FIG. 3 is a diagram illustrating an example of a learning pattern, a preprocessed pattern, and a recognition result.
FIG. 4 is a diagram showing an example of a rejected training pattern.
FIG. 5 is a diagram showing a recognition rate when a created template is used.
FIG. 6 is a configuration diagram of a pattern recognition apparatus according to another embodiment of the present invention.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 11 Learning pattern storage circuit 12 Training pattern extraction circuit 13 Template selection circuit 14 Template storage circuit 15 Pattern storage circuit 16 Recognition circuit 21-26 Step 31 Input device 32-35 Storage device 36 Output device 37 Recording medium 38 Data processing device

Claims (6)

The template to be matched with an input pattern, made using training patterns of the learning patterns in, a templating method performed by the template creation apparatus,
Exercise pattern extracting unit reads the learning patterns from the learning pattern storage means for storing a learning pattern, reads the template from the template storage means for storing the template that is created in advance for each category, the Predetermined preprocessing is performed on the read learning pattern, and the preprocessed pattern is compared with the template to obtain the image position of the portion showing high similarity to the template and the category of the template. , A training pattern extraction stage for extracting a training pattern in the learning pattern from these ,
The template selection means obtains one or more training patterns similar to the extracted training pattern from the learning pattern storage means , and when there is a category that occupies a majority of the categories of the similar training pattern, It screened the extracted training pattern as a template, and the template selecting step of registering a new the template to the template storage means,
Template creation how to have a. The template selection means, in the template selecting step, the category name similar pattern a majority, determines the category name when registering the exercise pattern as a template, the template creation process according to claim 1, wherein. A template creation device that creates a template to be matched with an input pattern using a training pattern in a learning pattern,
Learning pattern storage means for storing a learning pattern;
Template storage means for storing templates created in advance for each category ;
A learning pattern is read from the learning pattern storage means, a template is read from the template storage means, a predetermined preprocessing is performed on the read learning pattern, and a preprocessed pattern is compared with the template. A training pattern extracting means for obtaining an image position of a portion showing a high similarity to the template and a category of the template, and extracting a training pattern in the learning pattern therefrom,
One or a plurality of training patterns similar to the extracted training pattern are obtained from the learning pattern storage means , and when there is a category that occupies a majority of the categories of the similar training pattern, the extracted training pattern is It was selected as a template, and the template selecting means for registering a new the template to the template storage means,
Template creation apparatus to have a. The template selecting means, the category name similar pattern a majority, determines the category name when registering the exercise pattern as a template, the template creation apparatus according to claim 3, wherein. A recording medium recording a template creation program for causing a computer to execute a process of creating a template to be matched with an input pattern using a training pattern in a learning pattern,
The learning pattern is read from the learning pattern storage means storing the learning pattern, the template is read from the template storage means storing the template prepared in advance for each category, and the read learning pattern is used as the read learning pattern. Predetermined preprocessing is performed, the preprocessed pattern is compared with the template, and the image position of the portion showing a high similarity to the template and the category of the template are obtained. A training pattern extraction procedure for extracting a training pattern in the pattern ;
One or a plurality of training patterns similar to the extracted training pattern are obtained from the learning pattern storage means , and when there is a category that occupies a majority of the categories of the similar training pattern, the extracted training pattern is was selected as a template, and the template selection procedure for registering a new said template storage means the template,
A recording medium on which a template creation program for causing a computer to execute is recorded . The recording medium according to claim 5 , wherein in the template selection step, a category name of a similar pattern occupying a majority in the template selection step is determined as a category name when the training pattern is registered as a template.

Images