JPH0793466A - Device for discriminating character kind and method therefor - Google Patents

Abstract

PURPOSE:To accurately discriminate whether a character to read is handwritten or typed in a recognition device used for reading characters. CONSTITUTION:For example, the white picture elements of one dot are added and a pattern with white frame is formed around each binarized character information. 16 kinds of 2X2 patterns which respectively consist of four picture elements of 2X2 dot and are made of the different combination of white and black picture elements are individually coordinated to the pattern with the white frame. Then a generation frequency of each 2X2 pattern at the pattern with the white frame is counted so as to discriminate the kind of a character from the rate of the non-linear/linear components of each 2X2 pattern. Thus, the caracter is always recognized through the use of a dictionary proper to the kinds of the character by discriminating the kinds of the character by utilizing that the outline of a typed character is linear than that of a handwritten character.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character type discriminating apparatus and a method for discriminating characters used for reading characters, such as address recognition in a mail automatic reading / sorting machine for reading and classifying addresses on mail.

[0002]

2. Description of the Related Art In recent years, as an automatic device using a character reading technique, for example, a mail automatic reading / sorting device has been developed which reads mailing addresses written on mails and sorts the mails. Has been introduced to the station.

In this automatic mail reading / sorting machine, it is very important to correctly detect the address information area, which is the address description position, and it is important to correctly identify the character type of the address. It's starting.

That is, the address of a postal item was generally written by hand, but recently, the number of postal items in which the address is described by printing type (printing) using a word processor or the like is increasing. .

Normally, in the character reading technique, it is essential to recognize the character accurately, and the accuracy of the recognition affects the reading accuracy.
Therefore, it is possible to read more accurately by determining the character type at the time of recognition and using a dictionary suitable for the character type.

Therefore, in the case of a mail automatic reading / sorting machine, it is determined whether or not the address is a handwritten stamp when the address is read, and the address recognition process is performed using a dictionary according to the result of the determination. There is.

In this way, the address recognition process is speeded up and the accuracy is improved, so that the address reading accuracy is improved and a high classification rate is secured. By the way, to use different dictionaries according to the character type, that is, in the case of handwriting, to use the dictionary for handwriting, and in the case of In printing, to use the dictionary for indicia, the character type of the character to be recognized is correct. Must be determined.

If this discrimination is erroneous, normal recognition processing cannot be performed, and conversely, erroneous reading or non-reading is caused, and the accuracy of address reading is deteriorated. However, in the past, handwriting was used to discriminate the character type by utilizing the fact that the character size and the distance (pitch) between the characters are different in handwriting than in printing. For this reason, when the number of characters is relatively large, it can be accurately determined, but when the number of characters is small, accurate determination cannot be performed.

That is, in the conventional automatic mail sorting machine, since it is determined whether the handwriting stamp is used or not based on the size of each character cut out in character units from the address information area of the mail and the pitch between characters, There is a problem in that the discrimination tends to be inaccurate depending on the length of the address, which hinders securing a high classification rate.

As described above, in the field of character reading, such as a sorting machine that reads the address and sorts the mail, it is necessary to correctly determine the character type. Was causing the decrease.

[0011]

As described above, conventionally, it is necessary to correctly determine the type of character to be read, and if it cannot be correctly determined, it reduces the accuracy of reading and, in turn, it leads to a decrease in the accuracy of reading. There was a problem that the performance of the automated equipment deteriorates.

Therefore, the present invention is capable of more accurately discriminating the type of a character to be read, and is suitable for use in a character reading device or the like for performing recognition processing by an automated device such as an automatic mail reading / sorting machine. It is an object of the present invention to provide a device and a method for determining the device.

[0013]

In order to achieve the above object, in a character type discriminating device of the present invention, in a device for discriminating whether a character to be read is a handwritten print type,
It is composed of detection means for detecting a linear component and a non-linear component of the character to be read, and discrimination means for discriminating the character type of the character by the ratio of the linear component and the non-linear component detected by the detection means. ing.

Further, in the character type discriminating apparatus of the present invention, in the case of discriminating whether the character to be read is a handwritten print character, one dot of white is provided around the binary dot pattern of the character to be read. Forming means for forming a pattern to be collated with a frame of pixels, and 4 pixels of 2 × 2 dots for the pattern to be collated formed by this forming means, white pixels and black pixels Mask means for applying 16 kinds of mask patterns, which are combinations of
By the counting means for counting the occurrence rate in the pattern to be collated for each mask pattern applied by the mask means, and the occurrence rate for each mask pattern counted by the counting means, the character is manually written and printed. It is composed of a judging means for judging whether it is a printed character.

Further, according to the character type discriminating method of the present invention, when discriminating whether the character to be read is a handwritten print character, the linear component and the non-linear component of the character to be read are detected, and this is detected. The character type of the character is determined based on the ratio between the linear component and the non-linear component.

Further, in the character type discriminating method of the present invention, in the case of discriminating whether the character to be read is a handwritten print character, one dot of white is provided around the binary dot pattern of the character to be read. A pattern to be collated is formed with a frame made up of pixels, and 16 types of 4 pixels of 2 × 2 dots are formed for the formed pattern to be collated, which is a combination of white pixels and black pixels. Apply the mask pattern of, for each of the mask pattern thus applied, counting the occurrence rate in the pattern to be collated, by the occurrence rate of each of the counted mask pattern,
It is designed to determine whether the character is a handwritten print type.

[0017]

According to the present invention, since it is possible to discriminate whether the character is a handwritten print type by the means described above, regardless of the number of characters, it is possible to use an appropriate dictionary according to the character type in the subsequent recognition processing. It will be.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 schematically shows the configuration of a mail piece automatic reading / sorting machine according to the present invention.

That is, this automatic postal mail reading and sorting machine is
A supply unit 11 in which postal items (subjects to be read) P, which are letters such as postcards and sealed letters, are set collectively and in an upright position, and one postal unit P set in the supply unit 11 is sequentially arranged from the front end. A take-out section 12 for taking out, a take-in carrying path 13 for carrying the postal matter P taken out by the taking-out section 12, a reading section 14 for reading address information of the postal matter P carried through the carrying path 13.
The postal unit P whose address information has been read by the reading unit 14 is configured by a sorting unit 15 that sorts based on the reading result (sorting designation data).

The sorting unit 15 is a letter transporting route 16 for transporting the postal matter P that has passed through the reading unit 14, and a plurality of stages (here, 7 stages A to G, for example) sorting transporting routes provided in the vertical direction. 17a to 17g, and a large number of pockets (collection boxes) 18 to be provided along each of the divided conveying paths 17a to 17g.

An operator panel 19 as an operation panel operated by an operator (post office worker) is provided above the supply unit 11. In addition, the transport path 1
3, 16, 17a to 17g are provided with transport detectors (not shown) configured to detect the transport of the postal matter P on the transport path, for example, configured by photosensors.

Then, the postal matter P set in the supply section 11 is sequentially taken out by the take-out section 12, conveyed through the take-in conveyance path 13 and sent to the reading section 14. Then, the address information described on the postal matter P is read by the reading unit 14.

After this, the postal matter P is sent to the sorting section 15,
Based on the sorting designation data corresponding to the address information read by the reading unit 14, any one of the letter transport path 16 and the sorting transport paths 17a to 17g is selectively transported to a predetermined pocket, that is, the sorting designation data. It is divided and accumulated in the corresponding pocket 18.

FIG. 2 shows a schematic structure of the reading unit 14 described above. That is, the reading unit 14 obtains an image of the entire surface of the postal matter P and performs photoelectric conversion on the photoelectric conversion unit 14a, and recognizes the address information by recognizing the character pattern according to the output of the photoelectric conversion unit 14a. It is composed of a part 14b.

The photoelectric conversion unit 14a obtains a pattern signal (reading signal) by optically scanning the surface of the mail P on which the address information is written and photoelectrically converting it. And a self-scanning CCD image sensor that receives reflected light from the mail P and converts it into an electric signal.

The identification section 14b is the address area detection section 4
2, a character recognition unit 43, a town name / block recognition unit 44, an address dictionary 45, and an address recognition unit 46.
The address area detection unit 42 detects an area (reading area) in which the address information is described from all the information described in the mail P based on the signal from the photoelectric conversion unit 14a, The data indicating the position of the address information area is output.

The details of the detection method in the address area detecting unit 42 are described in, for example, Japanese Patent Application No. 5-6236.
Since it is described in No. 5, the explanation here is omitted.
The character recognition unit 43 detects the signal supplied from the address area detection unit 42, that is, each piece of character information corresponding to the address information in the address information area, on a line-by-line basis and further on a character-by-character basis. The output from the detection / cutout circuit 52, that is, the normalization circuit 53 for normalizing and sampling the cut-out character information, and the character information processed by the normalization circuit 53 are recognized using, for example, a dictionary 55. It is composed of a recognition circuit 54.

The details of the recognition process using the dictionary 55 in the recognition circuit 54 will be described later. The town name / block recognition unit 44 recognizes the town name / block based on the address registered in the address dictionary 45 for the recognition character supplied from the character recognition unit 43.

The address recognition section 46 recognizes the address by the town name / block supplied from the town name / block recognition section 44,
The classification designation data corresponding to this address is output.

That is, the position of the pocket 18 in the sorting section 15 is indicated by the sorting designation data, and the mail items P corresponding to the sorting designation data are sorted and collected in the pocket 18.

Here, the recognition circuit 54 recognizes a character, for example, by calculating the similarity between the character information to be recognized and the reference pattern corresponding to the character in the dictionary 55 by the composite similarity method or the like. At that time, the character type of the character information to be recognized is determined, and the dictionary 55 is selectively used according to the result of the determination.

That is, the dictionary 55 has a handwritten character dictionary 55a having a reference pattern corresponding to handwritten characters in advance, and a printing pattern having a reference pattern corresponding to print characters (printing). A character dictionary 55b is prepared, and when the character information to be recognized is determined to be handwritten, the handwritten character dictionary 55a is used. Is used for each and the above recognition processing is performed.

A method of discriminating the character type of the character information to be recognized in the recognition circuit 54 will be described below.
FIG. 3 shows an example of character types used to describe the address information of the postal matter P.

That is, as the address information of the postal matter P,
There are cases of printed characters written by a word processor or the like (FIG. 11A) and cases of handwritten characters written by handwriting (FIG. 9B).

Since the printed characters are usually mechanically constructed, the contours of the characters are often linear compared to handwritten characters. In this embodiment, by utilizing the feature that the contour portion of the printed character is more linear than the contour portion of the handwritten character, it is checked whether or not the contour portion of the character information to be recognized is linear. At the same time, the character type is determined by detecting the size and pitch of each character information corresponding to the address information.

FIG. 4 shows a flow of processing relating to the above-mentioned character type discrimination operation. For example, the recognition circuit 5
In 4, the character information (binarized pattern), which is supplied through the normalization circuit 53 and is inspected by the detection / extraction circuit 52, is surrounded by one dot for each of the upper, lower, left, and right. Is added to form a pattern with a white frame as a pattern to be collated (step ST1).

FIG. 5 shows an example of a white framed pattern in which a white frame is added around the binarized pattern "T" cut out from the address information area on the mail P. Next, in the recognition circuit 54, 16 kinds of 2 × 2 which are composed of 4 pixels of 2 × 2 dots and which are different combinations of white pixels and black pixels
Create a pattern (mask pattern) (step ST
2).

Then, each of the 16 types of 2 × 2 patterns created is made to correspond to the above-mentioned white framed pattern, and the frequency (occurrence rate) of occurrence of each 2 × 2 pattern in this white framed pattern is counted ( Step ST3).

Here, the 16 kinds of 2 × 2 patterns to be created are related to a component that can form a straight line (straight line component), a component that cannot form a straight line (non-linear component), and a straight line configuration, respectively. It can be classified into three groups of components that cannot be obtained (ineffective components).

FIG. 6 shows an example in which 16 kinds of 2 × 2 patterns are classified into three groups. In this case, (a) in the figure consists of two types of patterns in which all four pixels are white pixels or black pixels, and is an invalid component that cannot be related to the straight line configuration.

In FIG. 4B, one of the four pixels has four types of patterns each of which is a black pixel (the other three are white pixels), which is a non-linear component that cannot form a straight line. . In the same figure (c), two diagonal lines out of four pixels are composed of two types of patterns each consisting of a white pixel or a black pixel, and are non-linear components that cannot form a straight line.

In the same figure (d), four of the four pixels consist of four types of patterns consisting of black pixels (the other one is a white pixel), which is a non-linear component that cannot form a straight line. . In the same figure (e), four types of patterns in which the upper, lower, left, and right of the four pixels are white pixels or black pixels, respectively, are straight line components that can form a straight line.

That is, 16 types of 2 × 2 patterns that can be considered by combining four pixels of white and black are the invalid components (first group) shown in FIG. ) And (d), respectively, and the linear component (third group) shown in (e) of the figure.

Thereafter, the recognition circuit 54 counts the frequency of occurrence of each 2 × 2 pattern in the white framed pattern for each component based on the above classification, and the ratio of the linear component to the non-linear component ( Based on the ratio of the straight line component), it is determined whether the character is a handwritten character or a printed character (step ST4).

For example, when it is judged that the linear component is more than a certain reference value (threshold value) as compared with the non-linear component, it is judged as a print character, and when it is judged that it is less than a certain reference value as a handwritten character. To be done.

Here, a specific example of discriminating the character type will be described by taking the "mouth" of the printing and the "b" of the handwriting as examples.
FIG. 7 is a pattern with a white frame formed from the “mouth” of the Inko and the “b” of the handwriting respectively, and FIG. 8 is 2 × 2 for the “mouth” of the Inou and the “b” of the handwriting. It shows the frequency of occurrence (linear component and non-linear component) for each pattern.

For example, when the frequency of occurrence for each 2 × 2 pattern for the linear component and the non-linear component in the respective white framed patterns for the “mouth” of the printing and the “b” of the handwriting is examined, In this case, the ratio of the linear component to the non-linear component is 40: 8 (=
5: 1).

On the other hand, regarding "B" in the handwriting, the ratio of the linear component and the non-linear component is 30:28 (= 15: 14). From this, it can be seen that the "mouth" of In print has more linear components than non-linear components (that is, less non-linear components) than "B" of handwriting. .

As described above, when the linear component is greater than a certain reference value compared to the non-linear component, it is possible to determine printing and when the linear component is less than a certain reference value, it is possible to determine handwriting. Although it is possible to determine the character type based only on the result of this determination, in this embodiment, in order to ensure the accuracy of the determination, for example, the detection cutout circuit 52 supplied via the normalization circuit 53. The size of each piece of character information and the pitch between the pieces of character information that have been cut off are detected, and the detection result is taken into consideration.

For example, when the size and pitch of each character information is constant, it is judged that the character is a print character, and when it is not constant, it is a handwritten character. Also, if you want to determine the character type by comparing with a reference value with a linear component, this process,
For example, by performing a plurality of pieces of character information that have been cut out by the detection / cutout circuit 52, or all of the character information, it is possible to ensure the level of reliability necessary for satisfying the accuracy of reading.

In this way, when the character type of the character information corresponding to the address information written on the mail P is determined, an appropriate dictionary is selected from the dictionary 55 according to the result of the determination.

That is, when it is determined that the character is a handwritten character, the handwritten character dictionary 55a is selected (step ST5).
a), if it is determined to be a print character, the print character dictionary 5
5b is selected (step ST5b).

Then, the selected dictionaries 55a, 5
Characters are recognized by the above-described composite similarity method using 5b for each (step ST6).
As a result, it becomes possible to more accurately discriminate the character type of the handwritten seal stamp, so that the character information corresponding to the address information can be correctly recognized.

As described above, regardless of the number of characters, it is possible to determine whether it is a handwritten stamp. That is, when determining whether a handwritten character is used, the fact that the contour of the printed character has more straight line components than the handwritten character is used.
As a result, it becomes possible to make a more accurate determination as compared with the case where the determination is simply made by the size and pitch of the character, so that it becomes possible to use an appropriate dictionary according to the character type in the subsequent recognition processing. Therefore, it is possible to reduce the problem that the character cannot be correctly recognized because the character type is erroneously identified, and it is possible to improve the accuracy and speed of the recognition process, and finally, It is possible to improve the accuracy of reading and also improve the performance of the automatic mail sorting machine.

In the above embodiment, the case where the composite similarity method is used for character recognition has been described, but the present invention is not limited to this, and various readers that use, for example, the pattern matching method or the simple similarity method. The same applies. Of course, various modifications can be made without departing from the scope of the invention.

[0056]

As described above in detail, according to the present invention, the type of character to be read can be more accurately discriminated, and the character reading is performed by the automatic device such as the automatic mail sorter or the like for recognition processing. It is possible to provide a character type discriminating apparatus and a discriminating method suitable for use in an apparatus or the like.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an outline of a mail piece automatic reading / sorting machine according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a schematic configuration of a reading unit of the automatic mail reading / sorting machine.

FIG. 3 is a diagram similarly showing an example of a character type used for describing address information of a mail item.

FIG. 4 is a flow chart similarly shown for explaining the flow of processing relating to determination of a character type.

FIG. 5 is a diagram similarly showing an example of a pattern with a white frame.

FIG. 6 is a diagram showing classification of 2 × 2 patterns.

FIG. 7 is a pattern diagram similarly shown for explaining a process for determining a character type by giving a specific example.

FIG. 8 is a frequency distribution diagram similarly shown for explaining a process for determining a character type with a specific example.

[Explanation of symbols]

14 ... Reading unit, 14a ... Photoelectric conversion unit, 14b ... Identification unit,
42 ... Address area detection unit, 43 ... Character recognition unit, 54 ... Recognition circuit, 55 ... Dictionary, 55a ... Handwritten character dictionary, 55b ...
Ink letter dictionary, P ... Mail.

Claims (6)

[Claims] 1. An apparatus for discriminating whether a character to be read is a hand-printed print type, a detecting means for detecting a linear component and a non-linear component of a character to be read, and the linear component detected by this detecting means A character type discriminating apparatus comprising: a discriminating means for discriminating a character type of the character based on a ratio with a straight line component. 2. A device for discriminating whether a character to be read is a hand-printed print type, in which a frame consisting of one dot of white pixels is attached around a binary dot pattern of the character to be read for collation. 16 kinds of mask patterns consisting of 4 pixels of 2 × 2 dots and consisting of a combination of white pixels and black pixels are applied to the forming means for forming a pattern and the pattern to be collated formed by this forming means. The mask means, the counting means for counting the occurrence rate of the masked pattern for each mask pattern applied by the mask means, and the occurrence rate for each mask pattern counted by the counting means A character type discriminating apparatus comprising a discriminating means for discriminating whether or not is a handwritten print type. 3. The character type discriminating apparatus according to claim 2, wherein the discriminating unit makes a determination in consideration of the size and pitch of each character to be read. 4. A method for discriminating whether a character to be read is a handwritten print type character, wherein a linear component and a non-linear component of the character to be read are detected, and a ratio between the detected linear component and non-linear component is detected. A character type discriminating method characterized in that the character type of the character is discriminated. 5. A method for determining whether a character to be read is a hand-printed print type, in which the character to be read is provided with a frame consisting of one dot of white pixels around the binary dot pattern for verification. A pattern is formed and 2 × 2 is formed for the formed pattern to be collated.
16 kinds of mask patterns consisting of 4 pixels of dots and consisting of a combination of white pixels and black pixels are applied, and the occurrence rate in the pattern to be collated for each of the applied mask patterns is counted. A character type discriminating method, characterized in that the character is discriminated whether it is a hand-printed print type based on the occurrence rate of each mask pattern. 6. The character type determination method according to claim 5, wherein the determination is performed in consideration of the size and pitch of each character to be read.