JP2002183666A - Pattern-recognizing device, image reader, pattern- recognizing method and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a new technique for recognizing a pattern, and a technology for realizing this technique. SOLUTION: A shape of an X-direction projection histogram for making a standard pattern of a character 'Large' (a Chinese character) by counting the number of black dots existing in the X direction in accordance with a position in Y direction becomes a unique shape proper to the standard pattern. The other projection histogram made by a similar technique also becomes a unique shape. The number of respective black dots counted by dividing the standard pattern into four partial areas, so as to be respectively equal in the lengths in the XY direction is different by the standard pattern. Thus, the number of black dots for these histograms and the respective partial areas is used as information for expressing a characteristic of the pattern, and a pattern belonging category is recognized.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technique for recognizing a category to which an input pattern read by being decomposed into pixels belongs.

[0002]

2. Description of the Related Art Pattern recognition for recognizing input patterns, that is, categories to which images such as letters, numbers, symbols, and figures belong, is a basic technical field applicable to various fields. Yes, research and development are actively conducted.

In the pattern recognition, a plurality of standard patterns prepared in advance are compared with an input pattern to identify the most similar standard pattern, and a category name represented by the specified standard pattern is identified by the input pattern. It is basically output as a name. For this reason, the processing time required for the recognition tends to increase as the number of patterns to be recognized increases. For this reason, in the pattern recognition, in addition to improving the accuracy of recognizing an input pattern, shortening of a processing time has become a major technical problem.

Various techniques for pattern recognition have been developed. For example, from the point of view, there are three methods: pattern matching, statistical identification, and parsing.
It is roughly divided into two. A combination of a plurality of them has recently been available. However, even now, a new method is being developed with a focus on achieving the demand for higher accuracy or higher speed.

An object of the present invention is to provide a new technique for performing pattern recognition and a technique for realizing the technique.

[0006]

The pattern recognition apparatuses according to the first to fourth aspects of the present invention are based on the premise that they recognize the category to which the input pattern read out after being decomposed into pixels. The following means are provided.

A pattern recognition apparatus according to a first aspect divides an input pattern into a plurality of regions, and for each of the regions, counts the number of pixels having a pixel value within a predetermined range. A candidate specifying means for specifying a candidate of a category to which the input pattern belongs from the distribution state of the number of pixels for each area counted by the pixel number counting means; and a recognition of a category to which the input pattern belongs for the candidates specified by the candidate specifying means. And category recognition means for performing

[0008] A pattern recognition device according to a second aspect is a pattern storage means for storing, for each category, a blank pattern created based on an area in which an input pattern belonging to a category does not exist, and a pattern storage means. And a category recognizing unit that compares the blank pattern with the input pattern and recognizes a category to which the input pattern belongs.

According to a third aspect of the present invention, there is provided a pattern recognition apparatus for generating a histogram by focusing on the positions of pixels constituting an input pattern, and a category to which the input pattern belongs based on the histogram generated by the histogram generation means. Category recognizing means for recognizing.

[0010] In the pattern recognition apparatus of the third aspect, the histogram creating means counts the number of pixels having a pixel value within a predetermined range for each position in a predetermined direction to thereby obtain a histogram. It is desirable to create. It is preferable that the predetermined direction includes at least one of the vertical direction and the horizontal direction of the input pattern.

A pattern recognition apparatus according to a fourth aspect, in addition to the configuration according to the first to third aspects, further comprises, when category recognition means cannot recognize the category to which the input pattern belongs, the density of the input pattern is determined. The apparatus further includes a density adjusting unit for adjusting.

The image reading apparatus of the present invention reads an image by decomposing the image into pixels, and causes the pattern recognition apparatus to recognize a category to which a pattern present in the read image belongs, and reads the image. Image reading means;
A recognition requesting unit for requesting the pattern recognition device to recognize a category to which a pattern present in an image read by the image reading unit belongs; and an image reading unit for notifying the category to which the pattern belongs if the pattern recognition device cannot recognize the category to which the pattern belongs. Density adjusting means for adjusting the density when reading the image.

Each of the pattern recognition methods according to the first to third aspects of the present invention is a method for recognizing a category to which an input pattern read out after being decomposed into pixels. Make that recognition.

In the pattern recognition method according to the first aspect, the input pattern is divided into a plurality of regions, and the number of pixels having a pixel value within a predetermined range is counted for each of the regions. A category candidate to which the input pattern belongs is specified from the distribution state of the counted number of pixels, and a category to which the input pattern belongs is recognized from the candidates with respect to the specified candidate.

In the pattern recognition method according to the second aspect, a blank pattern is created for each category based on an area where an input pattern belonging to the category does not exist, and the blank pattern created for each category is compared with the input pattern. , The category to which the input pattern belongs is recognized.

In the pattern recognition method of the third aspect, a histogram is created by focusing on the positions of the pixels constituting the input pattern, and the category to which the input pattern belongs is recognized based on the created histogram.

The recording medium according to the first to third aspects of the present invention comprises:
A recording medium readable by an apparatus for recognizing a category to which an input pattern read after being decomposed into pixels, and recording a program for realizing any of the pattern recognition apparatuses according to the first to third aspects. are doing.

[0018]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a schematic sectional view of a form processing apparatus equipped with an image reading device and a pattern recognition device according to the present embodiment.

In this apparatus, as shown in FIG. 1, when a worker inserts a form (medium) on which a barcode or a number is printed from an insertion port 101 provided in a casing, the worker inserts the form into a transport path 102. Along to the reject box 103. During the transportation, an image (barcode, numeral, etc.) on the form is read by an image sensor (for example, a contact type optical sensor) 104, and a predetermined process including printing is performed on the form whose contents have been recognized. Then, it is transported (stored) in the reject box 103. The stepping motor 107 is a power source for conveying a form. Printing is on the form and ribbon 1
05 and a platen (not shown) and the thermal head 10
6 is pressed between them so that heat is supplied from the thermal head 106 which is in contact with the ribbon 105. Ribbon winding motor 108
Is for winding the ribbon 105.
The image reading device according to the present embodiment is mounted for reading an image on a form to be conveyed, and another pattern recognition device is mounted for recognizing a digit present in the image read by the image reading device. Have been.

On the transport path 102, a plurality of sensors 110 to 114 are arranged for detecting a form. The sensors 110 to 114 are optical sensors including, for example, a light emitting element and a light receiving element. More specifically, for example, the sensors 110 to 113 are transmission-type optical sensors in which those elements are opposed to each other with the transport path 102 interposed therebetween, and the sensor 114 has the same elements as viewed from the transport path 102. It is a reflection type optical sensor arranged on the side. In addition, various sensors are provided for detecting the opening and closing of the cover, detecting the ribbon 105, and detecting the empty state of the reject box 103. Also, various solenoids and the like are provided. A solenoid is used to press the platen toward the thermal head 106.

Controller (control printed board) 109
Controls the entire apparatus, and is operated by electric power supplied from the outside. External device (for example, host,
(Computer) and exchanges signals through a LAN.

FIG. 2 is a circuit block diagram of the controller 109. The MPU 201 has ROMs 202 and 20
3 to control the entire apparatus. The control is performed in accordance with the detection results of various sensors and the instructions of the worker while using the RAM 204. The program stored in the ROM 202 is mainly for initializing the apparatus, and the program stored in the other ROM 203 is mainly for performing control operations after the initial setting. Various data including standard patterns (standard characters) for recognizing numbers are stored in the ROM 203, for example.

The network interface 205 is
It exchanges signals with external devices via a LAN (line). Information on the form on which the barcode or the number is read and the content is recognized and processed is transmitted to an external device (for example, a host computer) via the interface 205.

The built-in I / O 207 is used when the MPU 201 sends data to the barcode decoder 208. The built-in I / O 207 is used for transmitting the image recognition result of the barcode read by the image sensor 104, for example, a code string indicating the type of each bar constituting the barcode. The barcode decoder 208 recognizes the content represented by the barcode from the recognition result (code string) and outputs the recognition result to the MPU 201.

The image sensor 104 includes a light-emitting element, a device (for example, a CCD) in which a large number of light-receiving elements are arranged in one or more rows, and a drive circuit. The light emitted from the light-emitting elements is reflected on the form being transported, and the reflected light is received by each light-receiving element of the device, so that the image of the form is read out for each row or for every plurality of rows and the image data is output. I do. The image data is obtained by binarizing the value of a pixel (dot) according to a preset threshold value. By outputting the interrupt signal INT5 to the interrupt control register 206 each time the image data is output, the MPU 20
Notify 1 to that effect.

The drive circuit is provided with a sensitivity adjustment function for appropriately recognizing bar codes and numerals. The sensitivity adjustment function is, for example, a function of changing a setting of a threshold value referred to when binarizing a pixel value. The sensitivity adjustment function allows the density of an image to be read to be adjusted as needed. In addition,
The density of the image becomes higher as the threshold value is set lower, and appears in the binarized image in such a manner that black portions and pixels to be read increase.

CIS data transfer circuit (DMA) 209
Receives the image data output by the image sensor 104 reading the form, and
Transfer to an area of the RAM 204 designated in advance. MP
The U201 performs recognition by extracting (cutting out) bar code and numeric image data from the image data. In the barcode, the type of each bar constituting the barcode is recognized, and the recognition result (code sequence) is sent to the barcode decoder 208. For numbers, the category to which each number belongs is recognized, and a predetermined process is performed using the recognition result. If there is a number that could not be recognized, the sensitivity of the image sensor 104 is adjusted, the form is returned to the position where the image sensor 104 reads the image, and the image is read again. On the other hand, a predetermined mark that is formed on a form and indicates a position where printing should be performed,
For example, the position of the frame is detected, and the print start position is set based on the detection result.

The CIS I / F 210 mediates the exchange of signals between the MPU 201 and the image sensor 104. The MPU 201 instructs reading of a form via the I / F 210.

The MPU 210 prepares image data of information (for example, a date) to be printed on a form in the RAM 204. By sending the initial (head) address and transfer size of the area of the RAM 204 storing the image data to the thermal head data transfer circuit 211,
The image data is transferred to the thermal head drive circuit 213.

The thermal head 106 is provided with, for example, heating elements arranged in a line. The thermal head drive circuit 213 drives the thermal head 106 in accordance with the transferred image data to perform printing on a form in units of one row, and controls the interrupt signal INT6 every time printing of one row is completed. Register 20
6 to notify the MPU 201 of the fact.

The built-in I / O 212 is used when the MPU 201 sends a command to the thermal head drive circuit 213. The MPU 201 basically determines the timing of transmitting a command for instructing printing in accordance with the position of a frame on which printing information (for example, a date) detected from image data of a form is to be printed, and at the determined timing. The command is sent to the built-in I / O 212. Further, if necessary, a command for designating the print density is transmitted to adjust the density of information printed on the form.

The four input / output registers 214 to 217 are
It is used for exchanging data between the stepping motor drive circuit 219, the DC motor drive circuit 220, and the other drive circuit group 221 and the MPU 201, or transmitting instruction contents from the MPU 201. Stepping motor drive circuit 2
19 is for driving the stepping motor 107, and the DC motor drive circuit 220 is for driving the ribbon winding motor 108. Other drive circuit groups 221 include drive circuits for various sensors,
This diagram collectively shows a drive circuit for various solenoids, an operation detection circuit for various buttons, and the like. Among the circuits, the signal output from the sensor is converted into a digital value according to the level by the built-in ADC 218,
It is stored in any of the input / output registers 214 to 217.

The MPU 201 has these registers 214 to
Reference is made to the digital value of the sensor output signal stored in any one of 217 and is reflected in the control of the device. Also,
With reference to the contents of the interrupt control register 206, control of the conveyance and printing of the form is performed.

The operation of the above configuration will be described. Here, the basic operation will be described first, focusing on the processing on the form for convenience. When a worker inserts a form into the apparatus from the insertion slot 101, the sensor 110 detects it and the output signal changes. The output signal is converted into a digital value by the built-in ADC 218 and then stored in one of the input / output registers 214 to 217. Thus, the input / output registers 214 to 217
By referring to the digital value of the output signal of the sensor 110 stored in any of the above, the MPU 201 recognizes that the form has been inserted into the apparatus, and starts transporting the form.
The conveyance is performed by instructing the stepping motor drive circuit 219 to drive the stepping motor 107 via any of the input / output registers 214 to 217. The conditions (pulse frequency, pulse number, etc.) for driving the stepping motor 107 are specified at that time.

The form that has started to be conveyed from the insertion slot 101 is
Next, it is detected by the sensor 111. MPU201
Waits for the sensor 111 to detect the form,
It instructs the image sensor 104 to read a form image via the ISI / F 210. The reading instruction is released after the sensor 112 stops detecting the form, for example. While the image sensor 104 is reading the image of the form, the stepping motor drive circuit 219 changes the conditions for driving the stepping motor 107 to those for image reading. Thus, each time the image sensor 104 outputs the interrupt signal INT5, the form is transported by the number of columns (lines) read by the image sensor 104.

The form image data read and output by the image sensor 104 is transferred to the CIS data transfer circuit 20.
9 to the RAM 204. MPU201
Performs recognition by cutting out bar code and numeric image data printed on a form from the image data. In the barcode, the type of each bar constituting the barcode is recognized, and the recognition result is transmitted to the barcode decoder 2.
08 to make it recognize the contents of the barcode. As described later, for each numeral, a category to which the numeral belongs is recognized, and a predetermined process is performed using the recognition result. If there are numbers that could not be recognized,
After a command to instruct sensitivity adjustment is sent to the image sensor 104 to change the sensitivity setting, the image sensor 104
The document is reversely conveyed (conveyed toward the insertion slot 101) to the position to be read in step, and the image is read again. In addition, the position of the frame indicating the printing position is detected, and the information to be printed inside the frame is printed so that the information to be printed can be printed inside the frame in the transport direction and the orthogonal direction thereof (hereinafter, referred to as the transport orthogonal direction). Determine the starting position.

All numbers can be recognized, and the decoder 208
Is able to recognize the contents of the barcode, the MPU 201
Transports the form toward the reject box 103. When the decoder 208 cannot recognize the contents of the barcode, the form is reversely conveyed (conveyed toward the insertion slot 101), and the form is conveyed to the insertion slot 101. This causes the worker to confirm the form.

The form conveyed toward the reject box 103 is next detected by the sensor 113, and the conveyance is temporarily stopped, for example, at or near that position (hereinafter referred to as a temporary stop position for convenience). The MPU 201
When the sensor 113 detects the form, for example, by driving a solenoid driving circuit in the other driving circuit group 221, the power of the stepping motor 107 is transmitted from the sensor 113 to the preceding conveyance path 102. , The conveyance to the destination is started at a predetermined timing. The transport is first performed to a position where the thermal head 106 can print at the determined print start position on the form (print start position in the transport direction). When the platen is conveyed to that position, the solenoid for pressing the platen is driven by the corresponding drive circuit in the other drive circuit group 221, and is sandwiched between the platen and the thermal head 106 to press the form and the ribbon 105 into pressure. As a result, the printing is made ready.

Thereafter, the thermal head data transfer circuit 211 is caused to transfer the image data of the first column,
An instruction to drive the thermal head 106 is issued to the thermal head drive circuit 213 via / O212. Thereby, the information is printed from the print start position.

When the thermal head drive circuit 213 drives the thermal head 106 and outputs the interrupt signal INT6, the thermal head 106 drives the thermal head 106 to convey the form by the number of columns on which printing has been performed, and causes the data transfer circuit 211 to transfer the image data of the next column.
In this way, until printing using the thermal head 106 is completed, the MPU 201 repeatedly performs the operation of the cycle of transferring the image data → driving the thermal head 106 → transporting the form → transferring the image data. . It should be noted that the conveyance of the form for the row on which printing was performed by one drive of the thermal head 106 is performed by changing the driving conditions including the number of pulses for the row to the stepping motor driving circuit 2.
19 is realized.

For printing, the image data of the last column is transferred to the thermal head driving circuit 213 and
06 is driven and its driving circuit 213 outputs the interrupt signal I
It is completed by confirming that NT6 has been output. The form is conveyed and stored in the reject box 103 after printing is completed. By the storage, the processing for the form inserted from the insertion slot 101 is completed.

In the present embodiment, the digits (input patterns) read by the image sensor 104 are recognized as follows. This will be described in detail with reference to the explanatory diagrams shown in FIGS. Note that the recognition is performed on numbers that have been cut out from image data and subjected to preprocessing such as normalization. The numbers read by the image sensor 104 are hereinafter also referred to as scanner read characters for convenience.

FIGS. 3 and 4 are diagrams for explaining features extracted from character patterns. FIG. 3 illustrates the features extracted from the standard pattern of the character “large” as an example, and FIG. 4 illustrates the features extracted from the standard pattern of the character “small” as an example. Things. First, referring to FIGS. 3 and 4,
Features extracted from the character pattern for recognition will be described in detail.

The standard pattern stored in the ROM 203 is obtained by developing (arranging) a black dot having a dot (pixel) value of 1 (a black solid circle in the figure) in a 24 × 24 dot area. It was created. In the standard pattern of the character “large”, black dots are developed as shown in FIG. The position of the dot is based on the dot at the upper left corner, that is, the position is expressed as (0, 0). The reference dot is referred to as a reference dot.

FIG. 3B is a projection histogram in the X direction (horizontal direction in the normalized pattern) created from the standard pattern shown in FIG. 3A. The projection in the X direction represents a histogram created by counting the number of black dots existing in the X direction for each position in the Y direction (vertical direction in the normalized pattern). . FIG. 3C is a projection histogram in the Y direction created from the standard pattern shown in FIG. The projection in the Y direction indicates that it is a histogram created by counting the number of black dots existing in the Y direction for each position in the X direction. The projection in the X direction and the projection in the Y direction will be used in the same meaning hereinafter unless otherwise specified.

FIG. 3D is a projection histogram of the right side of the standard pattern shown in FIG. 3A in the X direction.
FIG. 3E is a projection histogram of the lower part of the standard pattern in the Y direction. Here, the right side portion is the rightmost partial region when the standard pattern is divided into four partial regions having the same length (width) in the X direction along the Y direction. is there. The other lower part is the length (height) of the standard pattern in the Y direction along the X direction.
Is the lowermost partial region when divided into four equal partial regions. Otherwise,
Unless otherwise specified, the left part means the leftmost partial area when the standard pattern is divided along the Y direction into four partial areas having the same length (width) in the X direction. The upper part refers to the uppermost partial area when the standard pattern is divided along the X direction into four partial areas having the same length (height) in the Y direction. I will use it in the meaning of pointing.

When the standard pattern shown in FIG. 3A is divided into four partial areas so that the lengths in the XY directions are equal to each other, the number of black dots existing in each partial area becomes as shown in FIG. ). The four partial areas thus divided are hereinafter referred to as an upper left area, an upper right area, a lower left area, and a lower right area, respectively, according to their positions. The original area divided into the four partial areas is hereinafter referred to as an entire area for convenience.

On the other hand, in the standard pattern of the character "small" shown in FIG. 4A, the projection histogram in the Y direction is shown in FIG.
The projection histogram in the X direction is shown in FIG.
The result is as shown in FIG. The projection histogram of the upper part in the Y direction is as shown in FIG. When the standard pattern shown in FIG. 4A is divided into four partial areas, the number of black dots existing in each area becomes as shown in FIG.

FIGS. 3 (b) to 3 (e) and FIGS. 4 (b) to 4 (b)
As is clear from (d), the histogram created focusing on the position where the black dot exists as described above has a unique shape. For this reason, in the present embodiment, such a histogram is used for recognition as information representing the characteristics of a pattern (character).

Further, as is apparent from FIGS. 3F and 4E, the number of black dots present in each of the four divided partial areas and the ratio between them also differ depending on the pattern. . For example, in the standard pattern of the character "Large", the difference in the number of black dots between partial areas is relatively small,
In the standard pattern of the character "small", the number of black dots existing in the lower right area is clearly smaller than the others. For this reason, in the present embodiment, the distribution state of the number of black dots in each partial area is also used for recognition as information representing the characteristics of a pattern (character). Recognition using the information is performed in order to specify a category candidate to which the input pattern belongs, because the processing is relatively light and high-precision recognition is difficult. Thereby, the recognition is performed prior to the recognition using the histogram.

In the present embodiment, the recognition is performed for numbers. FIG. 5 is a diagram for explaining features extracted from the standard pattern of each numeral. In FIG. 5, X represents projection in the X direction, and Y represents projection in the Y direction. The standard pattern is created by developing (arranging) black dots in a 24 × 24 dot area.

As shown in FIG. 5, in this embodiment, features for recognition are extracted for each standard pattern of numbers "0" to "9". Specifically, as the histogram, besides the whole (entire area),
The projection histogram in the X direction is divided into four partial areas, that is, a left part, an upper part, and a lower part.
The projection histogram of the direction is created. As shown in FIG. 3 (f) and FIG. 4 (e), the black dot distribution includes, for each of the upper left area, lower left area, upper right area, and lower right area, a black dot existing in the partial area. Is counted.

The number "1" among the numbers "0" to "9"
“4” and “7” are significantly different from the others in the shape of the projection histogram in the X direction of the entire area. The number "2" appears to be similar to the numbers "3" and "5". But,
Unlike the numbers "3" and "5", there is no bent line that wraps around the center in the Y direction,
The number of black dots near the center is smaller than that of the numbers “3” and “5”. Therefore, the number “2” can be recognized from the projection histogram in the X direction of the entire area. Since a loop-shaped line exists in the lower part of the number “6” and in an upper part of the number “9”, in the Y direction, the portion where the loop-shaped line is formed and the portion where it is not are: There is a clear difference in the total number of black dots. From this, the numbers “6” and “9” can also be recognized from the projection histogram in the X direction of the entire area. Therefore, what is difficult to recognize from the projection histogram in the X direction of the entire area is the numbers “0”, “3”, “5”, and “8”. The number “0” is difficult to distinguish from the number “8”, and the number “3” is difficult to distinguish from the number “5”.

The number “8” is different from the number “0”,
And a pattern in which lines intersect near the center in the Y direction. Therefore, the number of black dots near the center in the X direction is
Overall, it is more than that of the number "0". In the right part and the left part, the number of black dots near the center in the Y direction is locally reduced, which is a feature that is not in the number “0”. From this, the numbers "8" and "0"
Is finally the projection histogram in the Y direction of the whole area,
X-direction projection histogram in right part or X in left part
The direction can be recognized from the projection histogram.

Regarding the numbers "3" and "5", the numbers "3" and "5" are both patterns having a horizontal line at the top, but a line continues to the right of the horizontal line in the number "3". On the other hand, in the case of the number “5”, a line continues to the left. For this reason, the shape of the upper part projection histogram in the Y direction has a clear difference due to the difference.
From this, the numbers “3” and “5” can be recognized from the projection histogram in the Y direction of the upper part.

As is apparent from the above, by comparing the histograms created from the patterns, in other words, by manipulating the patterns to create another new pattern and comparing the created patterns,
Pattern recognition can be performed. In the present embodiment, a histogram is used for pattern recognition in a manner that assumes a category to which an input pattern belongs (specifies a candidate).

As a method for creating a histogram,
It is not limited to the above. Various modifications can be made, including how to divide the entire area when creating the histogram, the partial area to be created, and the like. If the category to which the input pattern belongs is assumed, it is not necessary to create many histograms.

On the other hand, in the black dot distribution (the number of black dots in each partial area when the entire area is divided into four partial areas),
The numbers make a clear difference. For example, the numeral “1” has a feature that no black dot exists in the lower left area, which is not found in other numerals. The numeral "7" has a characteristic that no black dot exists in the lower right area, which is not found in other numerals. Therefore, those numbers “1” and “7” can be recognized from the black dot distribution.

As for other numbers, the numbers "0" and "8" are characterized in that there is no large difference in the number of black dots in each partial area. The number "2" is characterized in that the number of black dots in the upper left area and the lower right area is smaller than that in other partial areas. The number "3" is characterized in that the number of black dots in the lower left area is smaller than that in other partial areas. The number of black dots in the upper right area is smaller than that of the number “5”. The number "4" has a feature that the number of black dots in the upper right area is very small as compared with that of the other partial areas. In the number "5",
There is a feature that the number of black dots in the lower left area is smaller than that of other partial areas, and the number of black dots in the upper left area is largest. The numeral "6" has a feature that the number of black dots in the upper right area is smaller than that in the other partial areas. The numeral "9" is characterized in that the number of black dots in the upper left area and the upper right area located on the upper side is larger than that in the lower left area and the lower right area located on the lower side.

As described above, each numeral has the above-described characteristics in the black dot distribution. For this reason, it is possible to specify a candidate considered as a category to which the input pattern (number) belongs from the black dot distribution, or to specify a category to which the input pattern belongs depending on the input pattern. The identification can be performed by dividing the input pattern into partial areas, counting the number of black dots for each partial area, and comparing the number with that of the standard pattern.

The candidate specified from the black dot distribution is
If the input pattern is the number “0”, for example, the number “0”
And “8”. Hereinafter, similarly, if the input pattern is the number “1”, the number is “1”, for example, if the input pattern is the number “2”, the number is “2”, and if the input pattern is the number “3”. For example, the numbers are "3", "5" and "9". If the input pattern is the number "4", the number is "4". If the input pattern is the number "5", the numbers are "3" and "9". 5 "and" 9 ", and if the input pattern is the number" 6 ", for example, the number is" 6 ";
If the input pattern is the number "7", it is, for example, the number "7". If the input pattern is the number "8", it is, for example, the numbers "0" and "8". For example, the numbers "3", "5" and "9".

The recognition based on the black dot distribution has such characteristics that the processing is relatively light and high-precision recognition is difficult. Therefore, in the present embodiment, this is performed in order to specify a category candidate to which the input pattern belongs. By narrowing down candidates by relatively light processing and recognizing categories by relatively heavy processing, the overall processing load is reduced. As a result, the time required for pattern recognition is reduced, that is, the speed of the recognition process is increased without reducing the recognition accuracy.

The method of dividing the whole area when counting the number of black dots and the partial area to be counted are as follows.
It is not limited to the above. It may be appropriately determined according to the shape of the input pattern to be recognized.

For example, the ROM 203 stores the histogram, data indicating an area for creating the histogram, the number of black dots in each partial area, and data indicating each partial area. Thereby, the MPU 201 refers to the data stored in the ROM 203 and performs recognition based on the histogram and the black dot distribution. The histogram and the number of black dots in each partial area may be created or counted as necessary.
It does not have to be prepared in advance, for example, by storing it in the storage device 3.

In the present embodiment, in addition to the recognition based on the histogram and the black dot distribution, recognition based on pattern matching is performed. The pattern matching is performed as follows. This will be described in detail with reference to FIGS.

FIG. 6 is a diagram showing a state where the standard pattern is superimposed on the input pattern. For example, a standard pattern of a number “0” is shown for each input pattern, which is superimposed on each of the input patterns of numbers “0” to “9”.

The image sensor 104 reads an image on a form, for example, at 200 DPI (the number of dots per cm is about 79; hereinafter, expressed as 79 DPC). The standard pattern is 24 × 24 dots. From this, the normalization is performed by the image sensor 1
The size of the input pattern (number) read at step 04 is adjusted to the size of the standard pattern. 6A to 6J are obtained by superimposing an input pattern matched to the size of the standard pattern on the basis of the normalization by the input patterns (numbers “0” to “9”).

The numerical value representing the degree of similarity between the standard pattern and the input pattern is calculated by paying attention to the ratio of coincidence of the black dots. Specifically, the number of dots that match each other in the total number of black dots of the pattern, that is, the number of dots in which both dots associated with the pattern are black dots, is counted, and the counted number of dots (the number of matching dots) Is divided by the total number of black dots. The calculated value is further multiplied by 100 to make the unit a percentage. The calculated value is referred to as a coincidence ratio, and a value expressed as a percentage is referred to as a coincidence ratio.

There are two types of patterns, a standard pattern and an input pattern. For this reason, in the present embodiment, the coincidence rate is calculated for each pattern. Further, the matching rate of a pattern obtained by superimposing the standard pattern and the input pattern (hereinafter, referred to as a combined pattern) is also calculated. Hereinafter, the matching rates are referred to as an input pattern matching rate if the total number of black dots is an input pattern, a standard pattern matching rate if the total pattern is a standard pattern, and a combined pattern matching rate if the total pattern is a synthetic pattern. I will.

FIG. 7 is a diagram for explaining the coincidence rate between the standard pattern and the input pattern. Each of the standard patterns of numbers “0” to “9” (expressed as standard characters in the figure) and the numbers “0” to
For each input pattern "9" (indicated as a scanner read character in the figure), the matching rate between them is shown.
The upper part shows the input pattern matching rate, the lower part shows the standard pattern matching rate, and the whole shows the combined pattern matching rate.

As shown in FIG. 7, the coincidence rate between the standard pattern and the input pattern has the largest value when the patterns belong to the same category. As is apparent from this, the category to which the input pattern belongs can be determined from the matching rate.

From the value calculation method, the combined pattern coincidence rate is equal to or less than the input pattern coincidence rate and the standard pattern coincidence rate. Despite the different categories to which it belongs, the input pattern matching rate,
Alternatively, even if the standard pattern coincidence rate is a large value, the composite pattern coincidence rate tends to be a relatively small value.
In other words, if the categories to which they belong match, there is a tendency that the value of each matching rate is large (the average value is large) and those values fall within a small range. For this reason,
Recognition accuracy can be further improved by calculating a plurality of matching rates having different calculation methods and referring to those values. For this reason, in the present embodiment, the input pattern is recognized by calculating a plurality of matching rates.

The matching rate is highest when the categories have the same pattern. However, for some reason, for example, how the patterns are overlapped, the matching rate of the input pattern with the standard pattern of a category different from the category to which the input pattern originally belongs is different. Can be the largest. For this reason, in the present embodiment, the following pattern matching is also performed.

Each of the patterns "0" to "9" has a unique shape. Therefore, there is an area where a black dot is unlikely to exist even in a handwritten pattern. In the present embodiment, attention is paid to this, and for each category, a pattern in which black dots are arranged (developed) in an area (blank area) where it is considered that the possibility of the presence of black dots is the lowest (hereinafter referred to as a blank pattern) ) Is prepared, and the blank pattern is overlapped with the input pattern to perform matching for obtaining a matching rate.

FIGS. 8 to 17 are diagrams showing a state in which a blank pattern is superimposed on an input pattern. FIG. 8 shows the blank pattern of the number “0” as the number “0” “2” “6”.
The state of superimposition with each of the input patterns “8” and “9” is shown for each input pattern. Hereinafter, similarly,
FIG. 9 shows the blank pattern of the number “1” as each input pattern of the numbers “1” and “7”, and FIG. 10 shows the blank pattern of the number “2” as the numbers “0”, “2”, “3”, “7” and “7”. 9 is the input pattern of FIG. 11, the blank pattern of the number "3" is the input pattern of the numbers "2", "3", "5", "6" and "8", and FIG. 12 is the blank pattern of the number "4". 13 is an input pattern of the number "4", FIG. 13 is a blank pattern of the number "5", each input pattern of the numbers "3" and "5", and FIG. 14 is a blank pattern of the number "6". 15 is a blank pattern of the number "7", FIG. 15 is a blank pattern of the numbers "1", "2" and "7", and FIG. 16 is a blank pattern of the number "8". To the number "0"
Each input pattern of “2”, “3”, “6”, “8” and “9”, FIG. 17 shows the blank pattern of the number “9” as the number “0”
The respective input patterns “8” and “9” are shown for each input pattern. The input patterns selected for each blank pattern as described above are input patterns of a category in which the possibility of erroneous recognition by matching with the standard pattern is considered.

As shown in FIGS. 8 to 17, the blank pattern has no portion that matches the input pattern of the same category, that is, the dots associated between the patterns do not become black dots. ing. However, since the blank pattern is created by arranging black dots in the blank area of the pattern of the category to which the blank pattern belongs, a portion corresponding to most of the input patterns having different categories is generated. Even if there is an input pattern that does not produce a matching part, the pattern (standard pattern or input pattern) of the category to which the blank pattern belongs at that time is
A portion that matches the blank pattern of the category to which the input pattern belongs is generated. Specifically, as shown in FIG. 16 (b), the input pattern of the numeral “2” does not have a portion corresponding to the blank pattern of the numeral “8”. The input pattern of “8” has a portion that matches the blank pattern of the number “2”. This means that an input pattern can be recognized only by matching using a blank pattern.

As numerical values indicating the similarity between the blank pattern and the input pattern, three types of coincidence rates are calculated in the same manner as in the case of the matching using the standard pattern. The coincidence rate is hereinafter referred to as a prohibited coincidence rate in order to distinguish it from a coincidence rate calculated when performing matching using a standard pattern. For the three types of prohibited match rates, the input pattern prohibited match rate if the total number of black dots divided by the total number of matching black dots is that of the input pattern, and the blank pattern prohibited match if the total black dot number is that of the blank pattern. If the ratio and the total number of black dots are those of a composite pattern obtained by superimposing a blank pattern and an input pattern, it will be referred to as a composite pattern inhibition matching rate.

FIG. 18 is a diagram for explaining the forbidden coincidence rate between the blank pattern and the input pattern. Number "0" to "9"
For each blank pattern (indicated as a prohibited area of standard characters in the figure) and each input pattern of numbers "0" to "9" (indicated as a scanner read character in the figure), the prohibition matching rate between them is shown. It was done. The upper part shows the input pattern prohibition coincidence rate, the lower part shows the blank pattern prohibition coincidence rate, and the whole shows the combined pattern prohibition coincidence rate. Note that the column in which the prohibition matching rate is not described indicates a combination of categories that does not show a state in which the blank pattern and the input pattern are superimposed in the explanatory diagrams of FIGS.

Since the blank pattern is created as described above, as shown in FIG. 18, the forbidden coincidence rate between the blank pattern and the input pattern has the smallest value when they belong to the same category. In the blank pattern of the numeral "8", the prohibition coincidence rate is all 0 in each of the input patterns of the numerals "2" and "8" (see FIG. 16).
However, in the blank pattern of the numeral “2”, the prohibition coincidence rate is not all 0 when the input pattern is the numeral “8” (see FIG. 10). For this reason, the category to which the input pattern belongs can be determined from the prohibited match rate.

The blank pattern is stored in the ROM 203, for example, like the standard pattern. Thereby, the MPU 201 recognizes the input pattern using the blank pattern stored in the ROM 203.

In this embodiment, recognition of an input pattern, that is, recognition of a category to which the input pattern belongs, is performed by using the above-described plurality of methods as necessary. Thereby, the recognition accuracy is improved and the time required for the recognition is reduced. Hereinafter, the operation of the MPU 201 will be described in detail with reference to various flowcharts shown in FIGS.

FIG. 19 is a flowchart showing the flow of the entire processing. The flow is shown focusing on processing executed by the MPU 201 in order to correspond to a form (medium) inserted from the insertion slot 101. First, the overall processing will be described in detail with reference to FIG. It should be noted that the flowchart shown in FIG.
This is realized by reading and executing the program stored in the OM 202 or 203.

First, in step S 1, the document (medium) inserted from the insertion slot 101 is waited for to be detected by the sensor 110, and is conveyed to a position where it is read by the image sensor 104. Checking whether or not it has been transported to that position,
This is performed based on the detection result of the sensor 111.

When the sensor 111 detects the form, the process proceeds to step S2, and the image is read by the image sensor 104. Reading that image
As described above, the process is terminated after the sensor 112 stops detecting the form.

After the reading of the form image is completed, the process proceeds to step S3. In step S3, an OCR character recognition process for recognizing a character (input pattern) read by the image sensor 104 is executed. In a succeeding step S4, it is determined whether or not the character has been recognized. If all of the characters to be recognized in the image read by the image sensor 104 can be recognized, it is determined to that effect, and the form is conveyed to the reject box 103 while being printed and stored therein. After that, a series of processing ends. Otherwise, it is determined so and the process proceeds to step S5.

In step S5, the image sensor 104
Executes an image sensing density adjusting process for adjusting the density of the image read by. Thereafter, the process returns to step S1, and
The form is transported (reverse transport) to a position where the form is read by the image sensor 104. Thereby, the image of the form is read again to recognize the character (input pattern).

FIG. 20 is a flowchart of the OCR character recognition process executed as step S3. Next, the OCR character recognition processing will be described in detail with reference to FIG. In FIG. 20, for convenience,
Among the processing executed as the OCR character recognition processing, 1
The main processing executed to recognize a character (input pattern) is extracted and the flow is shown.

First, in step S11, a dot distribution recognizing process for performing recognition by paying attention to the distribution of black dots forming a character (input pattern) is executed. Subsequent step S12
Then, a projection histogram recognition process for performing recognition focusing on the histogram is executed. A series of processing ends after the execution.

In the dot distribution recognition processing and the projection histogram recognition processing, recognition is performed for one character (input pattern). Thus, in the OCR character recognition processing, the characters to be recognized (input patterns) are recognized by executing the processing for each character while changing the characters to be recognized.

FIG. 21 is a flowchart of the dot distribution recognizing process executed as step S11. Next, the recognition processing will be described in detail with reference to FIG.

First, in step S21, the RAM 204
, One input pattern (character) to be recognized is cut out from the image data read by the image sensor 104, and preprocessing including normalization is performed. In the following step S22, the width and height of the cut-out and normalized pattern are determined by, for example, the number of dots.
After that, it moves to step S23.

In step S23, the input pattern is divided into four partial areas based on the obtained width and height. In the next step S24, the number of black dots existing there is counted for each partial area obtained by dividing into four. When the counting is completed, the process proceeds to step S25.

In step S25, the category to which the input pattern belongs is assumed, that is, the candidate is specified from the distribution state of the counted number of black dots in each partial area. In a succeeding step S26, a matching process for performing recognition by comparing the blank pattern or the standard pattern of the candidate with the input pattern is executed. After the execution of the matching process, the process proceeds to step S27.

In the step S27, the recognition result by the matching process is determined. When the category to which the input pattern belongs can be recognized by executing the matching processing, it is determined that the category is included, and a series of processing ends here. If not, it is determined so and the process returns to step S25, where the category to which the input pattern belongs is assumed again. In step S27, although not shown, it is also determined whether another category has been assumed in step S25, and if it cannot be assumed,
After that, a series of processing is terminated.

The processing loop formed in steps S25 to S27 is repeatedly executed until another category cannot be assumed in step S25 or the category to which the input pattern belongs is recognized by the matching processing. Thereby, the category to which the input pattern belongs is recognized from among the candidates specified from the black dot distribution.

The assumption of the category from the black dot distribution can be made in a shorter time as compared with recognition using a blank pattern or a standard pattern prepared in advance. Assuming categories,
Since matching is performed using blank patterns and standard patterns of the assumed category, the number of times of matching can be greatly reduced as compared with the related art in which matching with all patterns is performed. This also reduces the time required for recognition. In addition, since the assumption is a form verified by matching using a blank pattern or a standard pattern, high recognition accuracy is maintained.

Next, the matching processing executed as step S26 will be described in detail with reference to the flowchart shown in FIG. First, step S3
In step 1, the blank pattern of the category assumed in step S25 executed immediately before (indicated as a prohibited area in the figure) is represented by R.
It reads out from the OM 203, performs matching with the input pattern (comparison by superimposition), and calculates various prohibited coincidence rates. In a succeeding step S32, it is determined whether or not the average value of the forbidden coincidence rates is less than 5%. The average value is 5%
If it is less than the predetermined value, the determination becomes YES and step S3
Move to 4. Otherwise, the determination is NO, and the input pattern does not belong to the assumed category, and a series of processing ends here.

The numerical value of less than 5% as a criterion for judging that the input pattern does not belong to the assumed category based on the matching with the blank pattern is a value other than when the category of the blank pattern and the category of the input pattern are the same as shown in FIG. Then
There is a possibility that the blank pattern of the number “8” is filled when matched with the input pattern of the number “2”. For this reason, in step S34, matching with the standard pattern of the assumed category (indicated as a standard character in the figure) is performed, and various matching rates are calculated. Next step S3
In 5, it is determined whether the average value of the coincidence rates or two or more coincidence rates among them exceeds 70%. If the average value or two or more of the calculated coincidence ratios are greater than 70%, the determination is YES and step S
The flow shifts to 36, where the category of the input pattern is determined to be the category of the standard pattern, that is, the assumed category is correct, and a series of processes is ended. If neither of them is satisfied, the determination is NO, and a series of processing ends here.

By executing the processing of step S36,
In step S27 shown in FIG. 21, it is determined that the input pattern has been recognized (OK). Thereby, the execution of the dot distribution recognition processing is also ended. If the determination in step S35 is NO, it is determined in step S27 that the input pattern could not be recognized (NG), so the process returns to step S25 and the category is assumed again.

As described above, in this embodiment, matching with a blank pattern is performed before matching with a standard pattern. This is because matching with a blank pattern can be performed in a shorter time than matching with a standard pattern. In other words, by performing matching with a standard pattern requiring a relatively long time as needed, the processing time can be shortened as a result.

FIG. 23 is a flowchart of the projection histogram recognition process executed as step S12 in the OCR character recognition process shown in FIG. Next, the projection histogram recognition processing will be described in detail with reference to FIG. Note that the flowchart shown in FIG. 23 illustrates a process extracted when the input pattern cannot be recognized in the immediately preceding dot distribution recognition process. In the projection histogram recognition process, if the input pattern has been recognized in the dot distribution recognition process, FIG.
A series of processes is terminated without executing the processes related to recognition shown in FIG. This prevents the processing time from becoming unnecessarily long.

First, in step S41, as shown in FIG. 5, projection histograms in the X direction and the Y direction of the entire area of the input pattern (character) or its partial area are created. In the following step S42, a category to which the input pattern belongs is assumed from the shape of the created projection histogram.

The categories assumed to belong to the input pattern are as follows: if the projection histograms in the X direction and the Y direction of the whole area are respectively created, for example, if the input pattern is the number “0”, the numbers “0” and “ 8 ". Hereinafter, similarly, when the input pattern is the number “1”, the number is “1”, and when the input pattern is the number “2”, the number is “2”.
And "3", if the input pattern is the number "3", the numbers "2", "3" and "5", if the input pattern is the number "4", the number "4" and the input pattern is the number "5" If there are, the numbers "2", "3" and "5", and if the input pattern is the number "6", the number "6" and the input pattern is the number "7"
If the input pattern is the numeral "7", if the input pattern is the numeral "8", the numeral is "0" and "8". If the input pattern is the numeral "9", the numeral is "9".

In step S43 following step S42, the matching process shown in FIG. 22 is executed using the blank patterns and standard patterns of the candidates assumed in step S42. Thereafter, in step S44, a recognition result by the matching process is determined. When the category to which the input pattern belongs can be recognized by executing the matching processing, it is determined that the category is included, and a series of processing ends here. If not, it is determined so and the process returns to step S42 to again assume the category to which the input pattern belongs. In step S44, although not particularly shown, it is also determined whether another category has been assumed in step S42, and if it cannot be assumed, the fact is determined and the series of processing is terminated. It has become.

The processing loop formed in steps S42 to S44 is repeatedly executed until another category cannot be assumed in step S42 or until the category to which the input pattern belongs is recognized by the matching processing. Thereby, the category to which the input pattern belongs is recognized from among the candidates specified from the created histogram.

FIG. 24 is a flowchart of the imesene density adjustment process executed as step S5 in the process shown in FIG. Next, FIG.
This will be described in detail with reference to FIG.

As described above, in the OCR character recognition processing shown in FIG. 20, the projection histogram recognition processing in step S12 is executed for each input pattern. As the projection histogram recognition process is executed for each input pattern, FIG.
If it is determined in step S44 that another category cannot be assumed in step S42 (meaning that the input pattern cannot be recognized) in step S42 shown in FIG. 3, the recognition result is determined in step S4 in FIG. It is determined as NG. The imesene density adjustment processing is executed thereby.

First, in step S51, the line of the input pattern is compared with that of the standard pattern which was most similar to the OCR character recognition processing (the average value of the coincidence rate was high) in correspondence with the position of the line. Then, it is determined whether or not the line of the input pattern is thicker than that of the standard pattern. If the line of the input pattern is thicker than the line of the standard pattern, the determination is YES and the process moves to step S52. Otherwise, the determination is no and the process moves to step S53.

In step S52, the image data is
The image sensor 1 issues a command to set a threshold value for conversion into a value higher than the current value by a predetermined value.
04, the density of the image read by the image sensor 104 is reduced, and then a series of processing ends. In the other step S53, a command for instructing the threshold to be set lower than the current value by a predetermined value is sent to the image sensor 104,
After the density of the image read by the image sensor 104 is increased, a series of processing ends.

In the present embodiment, the form printed on the form
Standard patterns and blank patterns are prepared according to the font of the numbers to be recognized. Even if such a pattern is prepared, if the density at the time of reading the input pattern is too high, the line forming the input pattern becomes thicker, causing a decrease in the matching rate and an increase in the forbidden matching rate. Even if the density is low, the line of the input pattern becomes thin or the line disappears partially, and as a result, the matching rate decreases and the prohibited matching rate increases. However, if the density (here, the threshold value) is adjusted as described above, an image can be read at an appropriate density (density appropriate for the density of the number to be printed and recognized). Become. For this reason, it can be avoided that the input pattern (printed number) cannot be recognized depending on the density at the time of reading, and the number of forms that cannot recognize the input pattern can be minimized.

In the present embodiment, since the image sensor 104 binarizes and outputs image data, when the density is adjusted, the image of the form is read again. If the image data is output as multi-valued data, the image data may be stored until the recognition is completed, so that it is not necessary to read the image data again. For this reason, the function of adjusting the density may be installed in the pattern recognition device.

The procedure for recognizing an input pattern and the method adopted for that purpose are not limited to the present embodiment, but can be variously changed. For example, the respective methods of recognition based on black dot distribution, recognition based on a histogram, and recognition based on a blank pattern are as described above.
Any of them can be used alone. Thus, the techniques can be combined in various ways, including the order in which they are used. As a specific procedure, for example, in the dot distribution recognition processing shown in FIG. Is set as a set, and it is performed each time a category is assumed. However, all categories considered to belong to the input pattern are assumed at once, and the category to which the input pattern belongs is recognized from among them. May be. Even in such a case, since the number of patterns to be matched with the input pattern can be suppressed, the processing time required for recognition can be reduced. In addition, since the pattern that is most similar to the input pattern can be specified from the assumed category, improvement in recognition accuracy can be expected. This is the same in the projection histogram recognition processing shown in FIG.

The input pattern to be recognized is a numeral in the present embodiment, but is not limited to this. Regardless of whether it is handwritten or not, other characters or symbols may be the target of recognition. The device on which the image reading device and the pattern recognition device to which the present invention is applied is not limited to the form processing device.

A program for realizing the operation of the above-described image reading apparatus or pattern recognition apparatus is as follows.
The program may be distributed on a recording medium such as a ROM, a CD-ROM, a floppy (registered trademark) disk, or a magneto-optical disk. Alternatively, part or all of the program may be distributed via a transmission medium used in a public network or the like. In such a case, the worker (user) obtains the program and loads it onto an existing device (including the exchange of a recording medium (ROM, etc.), etc.) to apply the present invention to that device. Can be. For this reason, the recording medium may be one that can be accessed by an apparatus that distributes the program.

[0115]

As described above, according to the present invention, the input pattern is divided into a plurality of regions, and the number of pixels having a pixel value within a predetermined range is counted for each region, and for each region, The category candidate to which the input pattern belongs is specified from the distribution state of the counted number of pixels, and the category to which the input pattern belongs is recognized from the specified candidates.

The shape of the input pattern differs depending on the category. As a result, the number of pixels whose values within a region obtained by dividing the input pattern are within a predetermined range, the difference or ratio of the number of pixels between regions, and the like also change depending on the category. For this reason, the category candidate to which the input pattern belongs can be specified (recognized) from the distribution state of the number of pixels counted for each region.

Since there is no need to compare patterns, the candidate can be specified relatively easily, that is, in a short time. When the category to which the input pattern belongs is identified from the identified candidates, the number of patterns to be compared with the input pattern for the recognition is reduced as a whole. As a result, the input pattern can be recognized in a shorter time.

In the present invention, a blank pattern is created for each category based on an area where an input pattern belonging to the category does not exist, and the blank pattern created for each category is compared with the input pattern to determine the category to which the input pattern belongs. recognize.

The shape of the input pattern differs depending on the category. As a result, an area (part) in which a line or the like constituting the input pattern does not exist or is considered to be unlikely to exist is unique for each category. Therefore, by comparing the blank pattern with the input pattern, the category to which the input pattern belongs can be recognized.

In the present invention, one or more histograms are created by paying attention to the positions of the pixels constituting the input pattern, and the category to which the input pattern belongs is recognized based on the created histogram.

The shape of the input pattern differs depending on the category. As a result, when a histogram is created by focusing on the pixel position, the shape becomes a shape in which the shape of the input pattern is reflected on the shape, and the shape becomes unique for each category. Therefore, based on the histogram thus created,
The category to which the input pattern belongs can be recognized.

[Brief description of the drawings]

FIG. 1 is a schematic cross-sectional view of a form processing device equipped with an image reading device and a pattern recognition device according to an embodiment.

FIG. 2 is a circuit block diagram of a controller.

FIG. 3 is a diagram for explaining features extracted from a character pattern (part 1).

FIG. 4 is a diagram for explaining features extracted from a character pattern (part 2).

FIG. 5 is a diagram illustrating features extracted from a standard pattern of each numeral.

FIG. 6 is a diagram showing a state in which a standard pattern is superimposed on an input pattern.

FIG. 7 is a diagram illustrating a matching rate between a standard pattern and an input pattern.

FIG. 8 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (when a blank pattern is 0).

FIG. 9 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 1).

FIG. 10 is a diagram illustrating a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 2).

FIG. 11 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (when the blank pattern is 3).

FIG. 12 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 4).

FIG. 13 is a diagram illustrating a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 5).

FIG. 14 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 6).

FIG. 15 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (when the blank pattern is 7).

FIG. 16 is a diagram illustrating a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 8).

FIG. 17 is a diagram showing a state in which a blank pattern is superimposed on an input pattern (in the case of a blank pattern of 9).

FIG. 18 is a diagram illustrating a prohibited match rate between a blank pattern and an input pattern.

FIG. 19 is a flowchart showing the flow of the entire processing.

FIG. 20 is a flowchart of an OCR character recognition process.

FIG. 21 is a flowchart of a dot distribution recognition process.

FIG. 22 is a flowchart of a matching process.

FIG. 23 is a flowchart of a projection histogram recognition process.

FIG. 24 is a flowchart of an imesene density adjustment process.

[Explanation of symbols]

 102 Conveyance path 104 Image sensor 107 Stepping motor 109 Controller 110-114 Sensor 201 MPU 202, 203 ROM 204 RAM 206 Interrupt control register 209 CIS data transfer circuit 210 CIS I / F 212 Built-in I / O 214-217 Input / output register 218 Built-in ADC 219 Stepping motor drive circuit

Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (reference) G06T 7/00 300 G06T 7/00 300G F term (reference) 5B064 AA01 AB13 BA01 CA11 DA05 DB06 DB12 DC27 DC37 DC39 DC42 EA38 5L096 BA17 CA02 EA13 FA36 GA19 GA28 JA03 JA06

Claims (13)

[Claims] 1. An apparatus for recognizing a category to which an input pattern read by being decomposed into pixels, wherein the input pattern is divided into a plurality of regions, and for each of the regions,
A pixel number counting unit that counts the number of pixels whose pixel values are within a predetermined range; and a category candidate to which the input pattern belongs from a distribution state of the pixel number for each area counted by the pixel number counting unit. A pattern recognition apparatus comprising: a candidate specifying unit for specifying; and a category recognizing unit for recognizing a category to which the input pattern belongs for the candidates specified by the candidate specifying unit. 2. An apparatus for recognizing a category to which an input pattern read by being decomposed into pixels belongs, wherein a blank pattern created based on an area in which the input pattern belonging to the category does not exist does not exist. Pattern recognition means comprising: a stored pattern storage means; and a category recognition means for recognizing a category to which the input pattern belongs by comparing a blank pattern stored in the pattern storage means with the input pattern. apparatus. 3. An apparatus for recognizing a category to which an input pattern read after being decomposed into pixels, wherein a histogram creating unit which creates a histogram by focusing on a position of a pixel constituting the input pattern; A category recognition unit for recognizing a category to which the input pattern belongs based on the histogram created by the pattern recognition apparatus. 4. The histogram creating means creates the histogram by counting the number of pixels in which the value of the pixel is within a predetermined range for each position in a predetermined direction. The pattern recognition device according to claim 3, wherein 5. The pattern recognition device according to claim 4, wherein the predetermined direction includes at least one of a vertical direction and a horizontal direction of the input pattern. 6. The apparatus according to claim 1, further comprising a density adjusting unit for adjusting the density of the input pattern when the category recognizing unit cannot recognize the category to which the input pattern belongs. The pattern recognition device according to any one of claims 1 to 5. 7. An apparatus for reading an image by decomposing the image into pixels and for causing a pattern recognition device to recognize a category to which a pattern present in the read image belongs, an image reading means for reading the image, A recognition requesting unit for requesting the pattern recognition device to recognize a category to which a pattern present in an image read by the image reading unit belongs; and if the pattern recognition device cannot recognize the category to which the pattern belongs, the image An image reading apparatus comprising: a density adjusting unit that adjusts the density when the reading unit reads the image. 8. A method for recognizing a category to which an input pattern read after being decomposed into pixels, wherein the input pattern is divided into a plurality of regions, and for each of the regions, the value of the pixel is Counting the number of pixels falling within a predetermined range, identifying a candidate for a category to which the input pattern belongs from the distribution state of the number of pixels counted for each area, and targeting the identified candidate, among the candidates Recognizing a category to which the input pattern belongs from the following. 9. A method for recognizing a category to which an input pattern read by being decomposed into pixels belongs, wherein a blank pattern is classified for each category based on an area where an input pattern belonging to the category does not exist. And a blank pattern created for each category is compared with the input pattern to recognize a category to which the input pattern belongs. 10. A method for recognizing a category to which an input pattern read after being decomposed into pixels, wherein a histogram is created by focusing on positions of pixels constituting the input pattern, and the created histogram is created. And recognizing a category to which the input pattern belongs based on the following. 11. A recording medium readable by an apparatus that recognizes a category to which an input pattern read by being decomposed into pixels belongs, wherein the input pattern is divided into a plurality of areas, and for each of the areas,
A function of counting the number of pixels whose pixel values are within a predetermined range; and a function of specifying a category candidate to which the input pattern belongs from a distribution state of the number of pixels for each area counted by the counting function. And a function of recognizing a category to which the input pattern belongs with respect to a candidate specified by the specifying function. 12. A recording medium readable by an apparatus that recognizes a category to which an input pattern read by being decomposed into pixels belongs, wherein the recording medium is created based on an area where an input pattern belonging to the category does not exist. And a function of comparing the blank pattern read by the reading function with the input pattern and recognizing a category to which the input pattern belongs. Recording medium on which a program for recording is recorded. 13. A recording medium readable by an apparatus for recognizing a category to which an input pattern read after being decomposed into pixels, the function of creating a histogram focusing on the positions of pixels constituting the input pattern And, based on the histogram created by the creating function,
And a function of recognizing a category to which the input pattern belongs.