JP2000215314A - Image identifying device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image identifying device, which can newly register an image, which has not been registered in the image identifying device, and can be applied even to the new design of paper money or securities by dividing an inputted image into blocks in a prescribed size, detecting a feature amount for every block unit and identifying whether a specified image is included in the inputted image or not on the basis of this feature amount. SOLUTION: A specified image recognizing part 37 detects whether the specified image exists in read image data or not. A second resolution converting part 38 converts the read image data to a prescribed resolution. On the basis of the image data converted to the prescribed resolution by the second resolution converting part 38, a recognizing part 39 recognizes the specified image. A mode selecting part 41 selects whether the input image is to be registered or to be recognized. A register part 42 registers the feature amount of the image data. When the mode selected by the mode selecting part 41 is in a recognition mode, a CPU 32 dispatches recognition mode information through a serial communication line 40 to the recognizing part 39 and the recognizing part 39 receives this information and executes recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image discriminating apparatus for bills and securities. Especially in addition to copiers, scanners, facsimile machines, etc., currency exchange machines, vending machines,
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device having a function of reading an image, such as a banknote pay-in / pay-out device, and relates to an image identification device for identifying authenticity of banknotes, securities, and the like.

[0002]

2. Description of the Related Art In recent years, in addition to copiers, scanners, facsimile machines, etc., currency exchange machines, vending machines, banknote pay-in / pay-out machines, and the like, determine whether or not an input image matches an image of a banknote, securities, or the like. There is a need for a sophisticated image identification device.

[0003] As such a conventional image identification device,
The image reading unit reads a characteristic pattern based on a specific shape or a specific color distribution of bills, securities, etc.,
2. Description of the Related Art There is known an image identification apparatus that identifies authenticity of bills, securities, and the like by comparing the characteristic pattern with a reference pattern registered in advance in a storage unit.

[0004]

However, new designs of banknotes and securities are issued one after another.
It is impossible to register all bills and securities. However, the above-described conventional image identification device has a problem that it cannot be used for a design of a new bill, securities, or the like because only registered bills, securities, and the like can be identified. .

[0005] The present invention solves the above-mentioned conventional problems, and makes it possible to newly register an image that has not been registered in the image identification device, and it is also possible to design a new bill or securities. It is an object to provide an image identification device that can be adapted.

[0006]

In order to solve the above-mentioned problems, an image discriminating apparatus according to the present invention comprises: an extracting means for dividing an input image into blocks of a predetermined size and extracting a feature amount for each block; Identification means for identifying whether or not a specific image is included in an image input based on the feature quantity extracted by the means, and registration means for registering the feature quantity of the specific image extracted for each block unit by the extraction means; , A registration data storage memory for storing the characteristic amount of the specific image registered by the registration means, and a mode selection means for selectively operating the identification means or the registration means.

[0007] With this configuration, it is possible to newly register an image that is not registered in the image identification device, and the image identification device can be applied to a design of a new bill or securities. Can be provided.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to achieve this object, an image identification apparatus according to a first aspect of the present invention divides an input image into blocks of a predetermined size and extracts a feature amount for each block unit. Means, identification means for identifying whether a specific image is included in the input image based on the feature quantity extracted by the extraction means, and feature quantity of the specific image extracted for each block by the extraction means. Registering means, a registration data storage memory for storing a feature amount of a specific image registered by the registering means, and a mode selecting means for selectively operating the identifying means or the registering means. With this configuration, the following operation is obtained.

(1) The mode selection means selects either a recognition mode for operating the identification means or a registration mode for operating the registration means.

(2) When the registration mode is selected, the extracting means extracts the characteristic amount in blocks of a predetermined size from the input specific image, and the registering means extracts the characteristic amounts of the specific image extracted in block units. Is registered in the registration data storage memory.

(3) When the recognition mode is selected, the extraction means extracts a feature amount from the input image in block units of a predetermined size, and the identification means based on the feature amount extracted by the extraction means. It is determined whether or not the specific image is included in the input image.

(4) A specific image according to the use environment can be registered at any time.

Here, as the registered data storage memory,
Examples include a rewritable nonvolatile memory such as an EEPROM and a flash memory, and a rewritable recording medium such as a magnetic disk device, a floppy disk device, and an optical disk device.

The feature quantity includes a color, a frame mask, template data,
The color range of the characteristic color, the weight coefficient in the intermediate layer of the neural network, and the like are used.

According to a second aspect of the present invention, in the image identification device according to the first aspect, the registration means stores the characteristic amount of the specific image extracted in units of blocks by the extraction means in a registration data storage memory. A new registration unit for newly registering, a specific image selection unit for selecting a specific image registered in the registration data storage memory, and a specification newly extracted by the extraction unit for a feature amount of the specific image selected by the specific image selection unit Rewriting registration means for rewriting and changing the image feature value;
And a registration data selecting means for selectively operating the new registration means or the rewriting registration means. With this configuration, the following operation can be obtained.

(1) When the registration mode is selected, the registration data selection means selects whether to perform new registration or rewrite registration.

(2) When the new registration is selected, the new registration unit newly registers the feature amount of the specific image extracted for each block by the extraction unit in the registration data storage memory.

(3) When rewriting registration is selected, the specific image selecting means selects a specific image to be rewritten from the specific images registered in the registration data storage memory, and the rewriting registration means selects the specific image to be rewritten. The characteristic amount of the specific image is rewritten and changed to the characteristic amount of the specific image newly extracted by the extracting unit.

(4) It is possible to adjust the identification accuracy.

According to a third aspect of the present invention, in the image identification apparatus according to the second aspect, when the input specific image includes a registered specific image, the rewriting registration unit is operated, When the input specific image does not include a registered specific image, the input specific image includes a registration data selection unit that performs control to operate the new registration unit, and the identification unit includes the registered specific image in the input specific image. This configuration is characterized in that it is determined whether or not it is performed. With this configuration, the following operation is obtained.

(1) When the registration mode is selected, first, the identification means identifies whether or not the input specific image includes a registered specific image, and then, the registration data selection means If the input specific image includes a registered specific image based on the identification result of the identification unit, the user selects to perform a new registration, and the input specific image includes the registered specific image. If not, select to perform rewrite registration.

(2) It is possible to prevent a plurality of specific images of the same type from being erroneously registered.

According to a fourth aspect of the present invention, there is provided the image identification apparatus according to any one of the first to third aspects, wherein:
The registering means is characterized in that a feature amount of an image other than the specific image to be identified is registered as non-identifying image feature data. With this configuration, the image data to be identified and the other image data are registered. And the effect of improving the identification accuracy is obtained.

According to a fifth aspect of the present invention, in the image identification device of the fourth aspect, the registering means selects a non-identifying image corresponding to the specific image to be identified from the registered non-identifying image feature data. This configuration is characterized in that the feature data is selected for each specific image. With this configuration, an effect that the data amount of the non-identified image can be reduced can be obtained.

According to a sixth aspect of the present invention, in the image identification apparatus of the fourth aspect, the registration means creates and registers non-identification image feature data corresponding to the specific image to be identified from the specific image to be identified. This configuration is characterized in that it is stored in a data storage memory. With this configuration, it is not necessary to register non-identified image feature data in advance, and an effect that a significant reduction in data amount can be obtained. .

An embodiment of the present invention will be described below with reference to the drawings.

(Embodiment 1) In the following embodiment,
A case where the present invention is applied to a copying apparatus will be described as an example of the image identification apparatus of the present invention.

FIG. 1 is a diagram showing an image copying system using the image identification apparatus of the present invention.

In FIG. 1, reference numeral 1 denotes an image reading device for reading a document and outputting digital color image data to an external device; 2, an image recording device for forming a color image based on image data transferred from the outside; Is a host computer that outputs a plurality of types of commands to the image reading device 1 to acquire image data or outputs image data to the image recording device 2, and 4 is an image reading device 1
And a cable for mutually connecting the image recording apparatus 2 and the host computer 3. Image data and commands are
Communication is performed bidirectionally between the devices by the cable 4.

In this embodiment, the image reading device 1, the image recording device 2, and the host computer 3
(Small Computer System In
The image recording device 2 issues a plurality of commands to the image reading device 1 without the intervention of the host computer 3 to obtain image data from the image reading device 1, and It is also possible to form an image based on this. Note that the image reading device 1 also has the image identification device of the present embodiment.

FIG. 2 is a sectional view of a main part of the image reading apparatus of the image copying system of FIG.

In FIG. 2, reference numeral 1 denotes the image reading apparatus described with reference to FIG. 1, 5 denotes a main body of the image reading apparatus, 6 denotes a document glass on which a document to be read is placed, and 7 denotes a carriage for scanning and reading the document. The carriage 7 is supported by a support member (not shown) such as a shaft and a rail, and the movement direction is restricted to one direction.

Reference numeral 8 denotes a driving source for driving the carriage, 9 denotes a driving pulley, and 10 denotes a timing belt. The drive source 8 employs a stepping motor. The power generated by the drive source 8 is transmitted to the drive pulley 9 by the timing belt 10.

Reference numeral 11 denotes a belt, and 12 denotes a driven pulley.
The belt 11 is stretched between the driving pulley 9 and the driven pulley 12, and the carriage 7 moves in the direction d1 (hereinafter, this direction is referred to as a sub-scanning direction) and the reverse direction with the rotation of the driving pulley 9. Let it.

Reference numeral 13 denotes an original placed on the original glass 6,
Reference numeral 14 denotes a document cover for blocking external light when reading an image of the document 13, reference numeral 15 denotes a supporting portion which supports the document cover 14 so that the document cover 14 can be opened and closed, 16 denotes a reference acquisition position,
Reference numeral 16a denotes a white reference plate attached to the reference acquisition position 16 on the document glass 5, and po1 denotes a home position of the carriage 7. The carriage 7 includes the image reading device 1
Is always at the home position po1 when is waiting. The original 13 is read line by line by the movement of the carriage 7.

FIG. 3 is a sectional view of a main part showing the internal structure of the carriage of the image reading apparatus of FIG. 2, and FIG. 4 is a schematic view of the carriage of the image reading apparatus when viewed from the side.

3 and 4, reference numeral 6 denotes an original glass;
Reference numeral 7 denotes a carriage, and 13 denotes an original. These are the same as those shown in FIG.

7a is the carriage 7 on the upper surface of the carriage 7.
Opening formed in the horizontal direction perpendicular to the moving direction of the
Is a lamp provided at both front and rear of the opening 7a inside the carriage 7, 18 is an aperture provided at the middle of both the front and rear lamps 17 just below the opening 7a, 19-1, 19-
Reference numeral 2 denotes a reflection mirror that reflects light reflected from a document that has passed through the aperture 18, reference numeral 20 denotes an image sensor fixed inside the carriage 7, and reference numeral 21 denotes light reflected from the reflection mirror 19-2 on the image sensor 20. Imaging lens,
22R, 22G, and 22B are line sensor arrays vertically arranged on one side of the image sensor 20 facing the reflection mirror 19-2.

The lamp 17 irradiates the original 13 placed on the original glass 6 with light through the opening 7 a and the original glass 6. The reflected light reflected by the document 13 and incident on the opening 7a is narrowed down to only the image reading position by the aperture 18, so that the image reading position is substantially specified. The image sensor 20 is reflected by the original 13, and reflected by the reflection mirrors 19-1 and 19-2 and the imaging lens 21.
The optical information reduced and formed into an image is converted into an electric signal, and is read in a one-to-one relationship with the document surface.

In FIG. 4, the lamp 17 and the aperture 18 shown in FIG. 3 are omitted to simplify the description of the operation described later.

FIG. 5 is a perspective view showing the details of the optical system of the image reading apparatus.

In FIG. 5, reference numeral 6 denotes an original glass;
And 19-2 are reflection mirrors, 20 is an image sensor, 2
Reference numeral 1 denotes an imaging lens, which is the same as that shown in FIG. FIG.
In order to make the drawing easier to see, the reflection mirrors 19-1 and 19-1
-2 is represented by a line.

22R is a line sensor array for reading a red signal, 22G is a line sensor array for reading a green signal, 22B is a line sensor array for reading a blue signal, and 23R is a line sensor array that is read by the line sensor array 22R. A reading line indicating a position, 23G is a reading line indicating a position on the original glass 6 read by the line sensor array 22G, and 23B is an original glass 6 read by the line sensor array 22B.
This is a reading line indicating the upper position.

Each of the line sensor arrays 22R, 22G, 2
2B is composed of a plurality of pixels arranged perpendicular to the moving direction of the carriage 7 (direction d1 in FIG. 5). A color filter corresponding to the color to be read is mounted on the surface of each of the line sensor arrays 22R, 22G, and 22B. Thus, in the present embodiment, the so-called 3
Line color sensor (three line sensor arrays 22
R, 22G, and 22B).

Three line sensor arrays 22R, 22
G and 22B are located at different positions.
Two positions (lines) cannot be read simultaneously. Therefore, as described later, each line sensor array 2
A means for delaying the image data obtained by 2R, 22G, and 22B by a predetermined amount is required.

In the following, each of the line sensor arrays 22
The arrangement direction of R, 22G and 22B (d2 direction in FIG. 5)
Hereinafter, it is called a main scanning direction.

FIG. 6 is a block diagram showing the configuration of the image data processing circuit of the image reading device.

In FIG. 6, reference numeral 17 denotes the aforementioned lamp,
Is the above-mentioned image sensor, which is composed of three line sensor arrays 22R, 22G, and 22B, and outputs analog image information in units of R, G, and B lines.

An amplifier 24a amplifies the analog image data signal output from the image sensor 20 with a predetermined gain, and 24b an A / D converter for converting the analog signal amplified by the amplifier 24a into a digital signal.
Reference numeral 5 denotes a shading correction unit for normalizing an input digital image signal to a previously acquired white and black dynamic range. Reference numeral 26 denotes image data of each color input from the shading correction unit 25 for each color. A line correction unit 27 that outputs image data with a fixed timing and outputs image data in which the image input timing shift due to the difference in the position of the line sensor arrays 22R, 22G, and 22B is corrected; Converting the resolution of the image data output from the line correction unit 26 based on the parameters specified by the first
The resolution converter 28 is a line sensor array 22R, 22
A color processing unit 29 for reducing the influence of unnecessary absorption bands on the spectral spectrum existing in the color filters on G and 22B and correcting the image data of each color so that vivid color reproduction can be performed. 29 is the image data processed in the above process. , A buffer for temporarily storing an image, 30 an interface, 31 an image reading device 1
Is another device (substantially refers to the host computer 3 or the image recording device 2 in the present embodiment).

The buffer 29 is provided for absorbing a difference in communication speed with the outside and outputting image data to an external device at a higher speed. Further, the image reading device 1 and the other device 31 are connected to a SCSI (Small Computer).
(System Interface). The image reading device 1 can output image data to another device 31 via the interface 30 via SCSI, and can obtain reading parameters such as a reading range and a reading resolution from the other device 31.

Reference numeral 32 denotes a CPU for controlling the operation sequence and the like of the image reading device 1, and reference numeral 33 denotes a driving signal (more precisely, an excitation signal for a stepping motor) to a driving source 8 for moving the carriage 7 of the image reading device.
Are motor control units, 34, 35, 36 and 36a are control signals.

The CPU 32 controls the operation of the line correction section 26 by a control signal 34, controls the operation of the first resolution conversion section 27 by a control signal 35, and controls the control signal 36.
Controls the rotation speed of the drive source 8 via the motor control unit 33, and controls the lighting / extinguishing of the lamp 17 by the control signal 36a.

Reference numeral 37 denotes a specific image recognizing unit for detecting whether or not a specific image exists in the read image data. 38 denotes a predetermined image (for example, 75 dpi (do)).
2) a second resolution conversion unit 39 for converting the resolution into a specific resolution based on the image data converted to a predetermined resolution by the second resolution conversion unit 38; 40, a serial communication line; Is a mode selection unit that is a means for selecting whether to register or recognize the input image,
Reference numeral 42 denotes a registration unit for registering a feature amount of image data.

The serial communication line 40 connects between the recognition section 39 and the CPU 32, and the recognition section 39 and the CPU 3
2 can exchange information by performing bidirectional communication.

When the mode selected by the mode selection unit 41 is the recognition mode, the CPU 32 passes the recognition mode information to the recognition unit 39 via the serial communication line 40, and the recognition unit 39 recognizes that the recognition mode information has been received. Execute

When the mode selected by the mode selection section 41 is the registration mode, the CPU 32 passes the registration mode information to the recognition section 39 via the serial communication line 40, and the registration section 42 performs the second resolution conversion. Using the image data converted to a predetermined resolution by the unit 38, the feature amount of the image data is registered. At this time, the recognition unit 39
Recognition may be performed to recognize whether data registered in the registration unit 42 is data of a new specific image or data already registered in the registration unit 42.

FIG. 7 is a schematic plan view of the image sensor viewed from the line sensor array side.

In FIG. 7, the line sensor arrays 22R, 22G and 22B of each color are arranged in a line in the main scanning direction d2, and the line sensor arrays of each color are arranged at intervals of L1 and L2 in the sub-scanning direction d1. Have been.

'□' indicates line sensor arrays 22R, 22
G and 22B are shown, and for the sake of simplicity, '□' is the size of one pixel at 600 dpi of the image reading apparatus.

In a general image sensor, L1 and L2
Are equal and L1 and L2 each have a value that is an integral multiple of the read pixel size. In this embodiment, for example, L1 and L2 are 8 when converted to a line of 600 dpi.
The line sensor arrays 22R, 22G for each color
22B is arranged in parallel with the sub-scanning direction at a pitch of 600 dpi / 8 = 75 dpi. As described above, the image sensor having such a structure cannot read the same position (line) at the same time, and the line correction unit 26 (see FIG. 6) corrects this.

FIG. 8 is a block diagram showing the structure of the specific image recognition unit.

In FIG. 8, reference numeral 37 denotes a specific image recognition unit;
8 is a second resolution converter for converting the input image data to a predetermined resolution, 54 is a memory for temporarily storing the image data converted to the predetermined resolution by the second resolution converter 38,
Reference numeral 55 denotes a feature color counter that counts the number of image data in a predetermined range of a plurality of colors to generate feature vector data to be described later. Reference numeral 56 denotes a plurality of templates prepared in advance by using the feature vector data generated by the feature color counter 55 A template selecting unit for selecting a template having the closest Euclidean distance as compared with the template selecting unit;
6, a template storage memory for storing a plurality of templates used for comparison with the feature vector; 58, a count value of the feature color counted by the feature color counter 55, a template number selected by the template selection unit 56, and feature vector data; A buffer for storing the Euclidean distance of the template, a recognition CPU 59 for recognizing whether or not a specific image is included in the image, and a counter 60 for counting the number of pixels of the input image data in the main scanning direction and the sub scanning direction. A main / sub pixel counter for outputting an interrupt signal 63 to the recognition CPU 59 every time a predetermined count is reached, 61 is a ROM storing a recognition program and the like, 43 is template data and a characteristic color counter stored in a template storage memory 57 The value of the characteristic color used in 55 and the information of its color range are Has been registered data storage memory, 62
Is a control signal for notifying control parameters and the like from the recognition CPU 59 to the second resolution converter 38;
AM.

The registration data storage memory 43 stores, for example, E
It is a rewritable nonvolatile memory such as an EPROM or a flash memory, and can add or delete registration data relating to a specific image such as template data or characteristic color information.

FIG. 9 is a block diagram showing the configuration of the characteristic color counter 55 of FIG.

In FIG. 9, 70C1, 70C2, 70C
C3 is a characteristic color detecting unit that detects an independent characteristic color. In the present embodiment, three characteristic colors (not only red, green, and blue, but also a mixture thereof) may be detected, and therefore, three characteristic colors are detected. Each characteristic color detection unit 70C1, 70C2, 70C3
There is no difference in configuration except that each detects a different color.

Reference numerals 71a to 71r denote each characteristic color detecting section 70C.
Comparators 72a, 72b, 72c for comparing the input RGB image data with a predetermined value and detecting whether or not the image data is within a predetermined range, provided for each of the 1, 70C2, and 70C3. Represents three characteristic color detection units 70
C1, 70C2, and 70C3, one for each of the characteristic color detection units 70C1, 70C2, and 70C3.
Six comparators 71a to 71f, 71g to 71l, 71
AND gates 73a, 73b, and 73c that perform product logical operation processing on the outputs of m to 71r and output the results are AND gates 72a and 72 for the three characteristic colors, respectively.
A counter 74 counts the number of times the outputs of b and 72c become 1, and a count buffer 74 accumulates the count results of the counters 73a, 73b and 73c.

The operation of the image identification apparatus of the present embodiment configured as described above will be described below.

First, the operation when the power is turned on will be described with reference to FIGS.

First, when the power of the image copying system is turned on, the CPU 32 (see FIG. 6) sends a control signal 36 to the motor control section 33 to control the drive source 8 so that the carriage 7 of the image reading apparatus 1 Is returned to the home position po1 regardless of the initial position. afterwards,
The CPU 32 moves the carriage 7 to the reference acquisition position 16 in which the aperture 18 is directly below the reference plate 16a, and turns on the lamp 17 by the control signal 36a to turn on the reference plate 16a.
Is actually read by the image sensor 20. At this time, the gain of the amplifier 24a is determined for the analog image data signal output from the image sensor 20, and the black and white level correction (shading correction) of the shading correction unit 25 is performed. After that, CPU3
2 returns the carriage 7 to the home position po1 again, and enters a standby state.

Here, the optical system of the carriage 7 of the image reading apparatus 1 will be described in detail with reference to FIG.

The line sensor array 22R arranged on the image sensor 20 reads the red image information.
The position of the reading line on the original glass 6 is PR. The line sensor array 22G reads green image information, and the position of the reading line on the original glass 6 is PG. The line sensor array 22B reads blue image information, and the position of the reading line on the original glass 6 is PB.

During the operation of reading an image, the carriage 7 moves in the sub-scanning direction (d1). At this time, with respect to the document 13, first, the position of PB is a reading line, the position of PG is next, and finally the position of PR is a reading line. That is, based on the same position (line) of the document, blue image data is obtained first, then green, and finally red image data. Therefore,
The blue image data obtained first and the green image data obtained next are held for a predetermined number of lines, and when the red image data is obtained, the blue and green images held are held. By outputting the data, it is possible to output the data by aligning the R, G, and B line positions.

Next, the independent reading operation of the image reading apparatus 1 will be described with reference to FIGS.

First, the user sets reading resolution, reading range, and the like from an external host such as the host computer 3 shown in FIG. 1, and then issues a document reading command from the external host. When a read command is issued, C
The PU 32 turns on the lamp 17 by the control signal 36a, and sends the control signal 36 to the motor control unit 33 to control the drive source 8, thereby moving the carriage 7 in the sub-scanning direction d1. At this time, the driving force of the driving source 8 is
The signal is transmitted to the carriage 7 via the timing belt 10, the driving pulley 9, the belt 11, and the driven pulley 12 (FIG. 2).
reference).

The CPU 32 (see FIG. 6)
Changes the driving speed of the carriage 7 to a speed corresponding to a reading resolution set in advance by the host computer 3 immediately before reaching the head position of the area corresponding to the reading range set by the host computer 3. The reading of the document 13 placed on the document 6 is started.
The document 13 is illuminated by a lamp 17 through the document glass 6. The reflected light from the original 13 is reflected by a reflection mirror 19-
1, 19-2, are reduced and imaged on the image sensor 20 by the imaging lens 21. The image sensor 20 converts reflected light from the document 13 imaged on the front surface into an electric signal, and outputs analog image data signals for each of red, green, and blue. Amplifier 2
Reference numeral 4a amplifies an analog image data signal input from the image sensor 20 with a set gain. A /
The D converter 24b converts the analog signal amplified by the amplifier 24a into a digital image data signal and outputs it. The shading correction unit 25 normalizes the input digital image data signal to a previously acquired dynamic range of black and white, and outputs a normalized image data signal. The line correction unit 26 outputs the image data of each color input from the shading correction unit 25 with a delay at a fixed timing for each color, and outputs the image data based on the difference in the position of the line sensor arrays 22R, 22G, and 22B. And outputs image data in which the timing deviation of the image is corrected. The first resolution conversion unit 27 converts the resolution of the image data output from the line correction unit 26 based on preset parameters and outputs image data having a predetermined resolution. The color processing unit 28 reduces the influence of unnecessary absorption bands on the spectral spectrum of the color filters on the line sensor arrays 22R, 22G, and 22B by using the image data input from the first resolution conversion unit 27 to achieve vivid color reproduction. The image data of each color is corrected as much as possible, and the image data from which the influence of the color filter has been removed is output. The image data output from the color processing unit 28 is temporarily stored in the buffer 29 and then transmitted from the interface 30 to the host computer 3 or the image recording device 2 via the cable 4.

When the reading operation for the designated reading range is completed, the CPU 32 moves the carriage 7 in the direction opposite to the sub-scanning direction d1, and moves the carriage 7 to the home position p.
Return to o1.

Next, the operation of the line correction section 26 will be described in detail.

FIG. 10 is a diagram showing the structure of data stored inside the line correction section 26.

In FIG. 10, reference numeral 50 denotes a memory area for storing green image data in line units, and reference numeral 51 denotes a memory area for storing blue image data in line units.

Image reading device 1 according to this embodiment
Are read in the order of blue, green, and red on the same line of the document as described above. Eight lines of green image data read by the line sensor array 22G are stored in the memory area 50, and 16 lines of blue image data read by the line sensor array 22B are stored in the memory area 51. Minutes are memorized. Each line sensor array 22R, 22G,
Since the interval of 22B is equivalent to eight lines of 600 dpi (see FIG. 7), when reading an image at 600 dpi, image data of eight lines is obtained for green image data and 16 lines for blue image data. When the red image data is read, the image data eight lines before the green image data and the image data sixteen lines before the blue image data are output. This is the same as reading at the same position on the document.

As described above, once the image is read at 600 dpi in the sub-scanning direction d1 and converted to a low resolution after performing the above-described line correction, the image is read at all resolutions if the resolution is lower than 600 dpi. be able to. However, in this case, it is premised that the image is once read at 600 dpi, so that the reading speed cannot be increased. This problem can be dealt with by reading the image while moving the carriage in the sub-scanning direction at a high speed, and changing the setting of the line correction unit 26. The method will be described below.

FIG. 11 is a diagram showing the operation of the line correction unit when reading an image at a resolution of 300 dpi in the sub-scanning direction.

In FIG. 11, reference numeral 50 denotes a memory area for storing green image data in line units, and reference numeral 51 denotes a memory area for storing blue image data in line units, which are the same as those in FIG.

Assuming that the moving speed of the carriage 7 when reading the original at 600 dpi, that is, the moving speed in the sub-scanning direction d1, is V, the moving speed of the carriage 7 when reading the original at 300 dpi is set to 2V. In other words, the moving speed of the carriage 7 is set to twice that at the time of reading at 600 dpi.

Here, the moving speed Vx of the carriage 7 at an arbitrary reading resolution is, for example, a reference reading resolution of 600 dpi, a moving speed of the carriage 7 at a reading of 600 dpi of V, and an actual reading resolution of X [dpi]. Then, it can be expressed by (Equation 1).

[0086]

(Equation 1)

Therefore, when reading an image at 300 dpi, the moving speed of the carriage 7 is twice as high as 600 dpi, so that the moving distance of the carriage 7 per unit time is also 2 times.
Double. Line sensor arrays 22R, 22G for each color,
Since the distance between the carriages 22B is always constant, if the moving speed of the carriage 7 doubles, the distance that the image reading device 1 moves when reading one line of image data also doubles, and the line correction unit 26 stores the moving distance. The number of lines of the image data to be stored may be １ ／. That is, as shown in FIG. 11, the interval between the line sensor arrays 22R, 22G, and 22B is 6
Eight lines of 00 dpi, that is, four lines of 300 dpi. Therefore, the line correction unit 26
When reading an image at dpi, image data for 4 lines is stored for green image data, and image data for 8 lines is stored for blue image data, and when the red image data is read, By outputting the image data four lines before the green image data and the image data eight lines before the blue image data simultaneously with the red image data, the same position on the document is read. Is the same as doing

Table 1 shows a generalization of the above.

[0089]

[Table 1]

That is, in this embodiment, as shown in (Table 1), the reading resolution is set to N times (N is an integer) based on 75 dpi. At this time, generally, the number of delay lines of the green image data stored in the memory area 50 for storing the green image data in line units is N, and the number of blue lines stored in the memory area 51 for storing the blue image data in line units is N. The number of delay lines of the image data is 2N.

These settings are made to the line correction unit 26 by the control signal 34 from the CPU 32 (see FIG. 6). Also, the carriage moving speed V at each resolution
x is given by (Equation 1). This setting is performed on the motor control unit 33 by the control signal 36 from the CPU 32 in FIG.

As described above, the line correction unit 26
Line sensor arrays 22R, 22 of the image sensor 20
The difference in the reading position caused by the difference in the positions of G and 22B is corrected, and the image data output from the line correction unit 26 is in a state equivalent to reading the same line of the document. At this time, the designation of the reading resolution is a discrete value, but the actual image reading device 1 accepts the designation of the reading resolution in units of 1 dpi from the host computer 3 or the image recording device 2 and corrects the image data. Output.

The processing performed by the line correction unit 26 is alignment in the sub-scanning direction d1, and no conversion is performed on image data in the main scanning direction d2. The first resolution converter 27 performs these processes. Hereinafter, processing in the first resolution conversion unit 27 will be described.

Here, for the sake of simplicity, it is assumed that a reading of 200 dpi is externally designated for the image reading apparatus.

When reading at 200 dpi is designated, the CPU 32 (see FIG. 6) sets the carriage moving speed for the reading resolution of 225 dpi for the motor control unit 33. According to (Table 1), this is 60
The speed is 2.7 times the carriage moving speed V at 0 dpi.

Next, the CPU 32 similarly sets the reading resolution of 225 dpi for the line correction unit 26. That is, the delay amount of the memory area 50 for storing the green image data in line units is reduced to three lines,
The delay amount of the memory area 51 for storing the blue image data in line units is set to six lines (see FIG. 10 or 11).

When the image is read with these settings,
From the line correction unit 26, 22 in the sub-scanning direction d1
The image data converted to 5 dpi is output.

Here, as an example, a case where an image is read at a resolution of 225 dpi when a resolution of 200 dpi is specified has been described. Motor control unit 33 and line correction unit 26
(Table 2) shows a general relationship of the contents set in the table, that is, a relationship between the set reading resolution and the actual reading resolution.

[0099]

[Table 2]

Next, the resolution conversion algorithm in the main scanning direction d2 will be described.

FIG. 12 is a diagram showing an algorithm for resolution conversion.

In FIG. 12, 53 is 1 at 600 dpi.
One pixel is shown. However, for ease of explanation, the actual pixel size is ignored, and only the center position of one pixel of 600 dpi is shown.

In the pixel 53 of 600 dpi, P600 — 0, P600 — 1, P60
0_2... P600_6 are numbered. Hereinafter, for the sake of convenience, the pixel values for these positions will be referred to as P6, for example.
The pixel value corresponding to the position of 00_0 is represented as * P600_0 (the concept of the pointer in the C language is loosely used).

First, 600 dpi image information is stored in 200
The case of converting to dpi will be described.

Here, the position of the first pixel after conversion to 200 dpi is always the first pixel of 600 dpi, ie, P
It should be aligned to the position of 600_0. Therefore, 200
The head pixel position of dpi is P200 which is the same as P600_0.
_0. Since the location is the same, the pixel value is also P600
The same value as _0, that is, * P600_0 is adopted. The next pixel position is P200_1, but in order to obtain this pixel value, consider that the location of P200_1 is represented by a pixel position of 600 dpi. Using a simple proportional equation (600
/ 200) × 1 = 3, so P200_1 = P60
0_3. Therefore, the pixel value at the position of P200_1 is *
P200_1 = * P600_3. Similarly, *
P200_2 = * P600_6 can also be obtained.

Next, the image information of 600 dpi is converted to 300d.
The case of conversion to pi will be described.

The position of the first pixel after conversion is always 600d
It is assumed that the pixel is aligned with the first pixel of pi, that is, P600_0. The head pixel position of 300 dpi is P600_
0, the pixel value is also the same value as P600_0,
That is, * P600_0 is adopted. The next pixel position is
P300_1, but in order to obtain this pixel value, P3
Consider that the location of 00_1 is represented by a pixel position of 600 dpi. Using a simple proportional equation (600/300) ×
Since 1 = 2, P300_1 = P600_2. Therefore, the pixel value at the position of P300_1 is * P300
_1 = * P600_2. Similarly, * P300
_2 = * P600_4, further * P300_3 = * P60
0_6 can be obtained.

Next, the image information of 600 dpi is converted to 400 d.
The case of conversion to pi will be described.

The position of the first pixel after conversion is always 600 dp.
It is assumed that the pixel is aligned with the head pixel of i, that is, the position of P600_0. Since the head pixel position of 400 dpi is the same as P600_0, the pixel value also adopts the same value as P600_0, that is, * P600_0. The next pixel position is P4
00_1, but in order to obtain this pixel value, P400
Consider that the location of _1 is represented by a pixel position of 600 dpi. When calculated using a simple proportional equation, (600/4
00) × 1 = 1.5, and P400_1 becomes P600_
It can be seen that it exists between 1 and P600_2. Therefore, the pixel value of P400_1 is calculated as (Equation 2) using the position information of 1.5.

[0110]

(Equation 2)

In this method, the position where the pixel after the resolution conversion exists is determined based on the pixel position of 600 dpi, and a weighting operation is performed based on the distance from the adjacent pixel of 600 dpi to obtain the pixel value after the resolution conversion. It is nothing but being.

When the above concept is similarly applied to P400_2, (600/400) × 2 = 3, and
It can be seen that P400_2 exists at the position of P600_3. Therefore, * P400_2 = * P600_3. Further, when the above concept is applied to P400_3, (600/400) × 3 = 4.5, and P4
It can be seen that 00_3 exists between P600_4 and P600_5. Therefore, the pixel value of P400_3 is calculated as (Equation 3) using the position information of 4.5.

[0113]

(Equation 3)

For the subsequent pixels, the pixel values can be obtained in the same manner.

The resolution conversion to 500 dpi can be processed in exactly the same way.

As described above, the conversion from 600 dpi to another resolution has been described as the resolution conversion processing in the main scanning direction. However, this is not an arithmetic method applied only to a reading resolution of 600 dpi, but is different from the original resolution. If the resolution after conversion is known, it is a method applicable in every case.

For example, image data in the sub-scanning direction output at a resolution of 225 dpi by line correction can be converted to, for example, 200 dpi in the same manner. In the present embodiment, the method described above is used for resolution conversion in the sub-scanning direction d1.

Next, the operation of the specific image recognition section 37 will be described with reference to FIGS.

The input to the second resolution converter 38 is performed from a stage preceding the first resolution converter 27 (see FIG. 6).
First, the reason will be described.

As described above, the image data output from the line correction section 26 is stored in the line sensor array 22 for each color.
The distance between the RGB lines in the sub-scanning direction d1 due to the different positions of R, 22G, and 22B is corrected. At this point, the image sensor 20 outputs the resolution in the main scanning direction d2, and no processing is performed. That is, in the above-described configuration, image data having a resolution of 600 dpi is output from the line correction unit 26 in the main scanning direction d2.

As described above, at the time of output from the line correction unit 26, the resolution in the main scanning direction d2 is different from that of the other devices 3.
Regardless of the reading resolution specified by 1, always 60
Since it is 0 dpi, this is set to a predetermined resolution, for example, 75
The conversion to dpi can be performed by a single, also parameter-invariant processing system. If the output of the first resolution conversion unit 27 is used, a predetermined resolution, for example, 75d
When converting to pi, hardware becomes complicated because image data of various resolutions must be handled.

In the sub-scanning direction d1, the line data output from the line correction section 26 is, as shown in Table 2, 75 dpi, 150 dpi, and 225 dpi.
i, 300 dpi, 375 dpi, 450 dpi, 52
It is either 5 dpi or 600 dpi. Most importantly, they are all multiples of 75 dpi. Therefore, these data are converted into the predetermined resolution,
Conversion to 75 dpi can be done very easily.

Recognition CPU 59 of specific image recognition section 37
Is the image reading device 1 via the serial communication line 40.
Of the CPU 32. The CPU 32 obtains the image reading conditions transferred from the other device 31 via the interface 30, and controls the line correction unit 26, the first resolution conversion unit 27, and the motor control unit 33 of the image reading device 1 based on the conditions. As described above, the CPU 32 sends the reading conditions regarding these resolutions to the recognition CPU 5 via the serial communication line 40.
9 is also notified. This allows the recognition CPU 59 to know the resolution of the image to be read.
Based on this information, the recognition CPU 59 sends the control signal 62
To the second resolution converter 38 in the sub-scanning direction d.
The process 1 is specified, more specifically, a thinning rate for all lines. Of course, the main scanning direction d2 is constant irrespective of the reading resolution, so that the pixel thinning rate in the line is fixed. (Table 3) shows the setting contents of the thinning rate for the second resolution converter 38.

[0124]

[Table 3]

As described above, by setting the fixed pixel thinning rate to 2 in the main scanning direction d2, image data of 600 dpi is always converted to 300 dpi. By performing such thinning processing, the amount of image data to be processed thereafter can be significantly reduced.

In the sub-scanning direction d1, the line thinning rate is changed according to the reading resolution. As a result, after the thinning, as shown in the resolution column of (Table 3), the resolution of main scanning × sub-scanning is 300 dpi × 75 dpi or 300 dpi.
It is converted to i × 150 dpi.

Next, the second resolution conversion section 38 converts the image data obtained by the thinning processing to 75 dpi in both the main scanning direction d2 and the sub-scanning direction d1 by averaging processing. Hereinafter, this resolution of 75 dpi is referred to as a predetermined resolution. When the image is converted into 300 dpi in the main scanning direction and 75 dpi in the sub-scanning direction by the thinning-out process, 4 pixels in the main scanning direction and 4 pixels in one line in the sub-scanning direction are used.
The value of one pixel is averaged. When the image is converted into 300 dpi in the main scanning direction and 150 dpi in the sub-scanning direction by the thinning-out processing, the values of 4 × 2 pixels are averaged using four pixels in the main scanning direction and pixels for two lines in the sub-scanning direction. To process.

With the above processing, the second resolution converter 38
Acquires image data of a predetermined resolution of 75 dpi in both the main scanning direction d2 and the sub-scanning direction d1. However, the predetermined resolution is not limited to 75 dpi.

Next, an outline of a specific image recognition algorithm in the image identification device of the present embodiment will be described first.

In FIG. 8, the RGB image signal converted to a predetermined resolution by the second resolution converter 38 is temporarily stored in the memory 54.

The RGB image signals stored in the memory 54 are cut out in blocks of a predetermined size and sent to the characteristic color counter 55 as RGB point sequential signals. The size of the block is set to, for example, 50 × 50 pixels (2500 pixels).

The characteristic color counter 55 stores the inputted RGB
The number of pixels in the range of RGB values determined in advance as the characteristic color is counted for the dot sequential signal. This count range is stored in the registration data storage memory 43 in advance, and is set in the characteristic color counter 55 during operation.
In the present embodiment, three different colors included in a specific image are defined as characteristic colors, and the number of pixels determined to be characteristic colors is counted for each block. Feature color counter 5
5 finishes counting the characteristic color of one block, transfers the result to the template selecting unit 56, and the template selecting unit 56 writes the result of the characteristic color counting transferred from the characteristic color counter 55 to the buffer 58.

The result of the characteristic color count output from the characteristic color counter 55 is obtained by counting the number of characteristic colors existing in a 50 × 50 pixel block. In the present embodiment, since the number of characteristic colors is 3, this can be regarded as outputting a three-dimensional vector (hereinafter, this is referred to as a characteristic vector). That is,
This means that the characteristic color counter 55 generates a characteristic vector.

The template selection unit 56 calculates a three-dimensional Euclidean distance between the feature vector generated by the feature color counter 55 and a plurality of templates stored in the template storage memory 57 in advance. The template selecting unit 56 compares the feature vector with each template based on the three-dimensional Euclidean distance, selects the closest template, and selects the selected template number (hereinafter, referred to as the nearest template number) and the three-dimensional Euclid. The distance is stored in the buffer 58. The nearest template number and the three-dimensional Euclidean distance serve as indices indicating the similarity between the input image data and the specific image.

The total number of pixels processed by the characteristic color counter 55 is counted and managed by the main / sub-pixel counter 60.
When the counted number of pixels reaches a predetermined amount, the main / sub-pixel counter 60 issues an interrupt signal 63 to the recognition CPU 59.
Occurs. Upon receiving the interrupt signal 63, the recognition CPU 59 reads the contents of the buffer 58 and obtains the characteristic color count result (feature vector), the nearest template number, and the three-dimensional Euclidean distance stored therein. The recognition CPU 59 selects the similarity based on the nearest neighbor template number and the three-dimensional Euclidean distance obtained from the buffer 58 for a plurality of blocks according to a certain rule, calculates the sum thereof, and determines the determination result according to the sum. Output. The recognition CPU 59 transmits the determination result to the CPU 32 via the serial communication line 40. The CPU 32 outputs this determination result to the interface 30, and the interface 30 outputs this determination result to another device 31 such as a host computer or an image recording device by SCSI.

Next, the operation of the characteristic color counter 55 will be described in detail with reference to FIG.

The characteristic color counter 55 detects three characteristic colors C1, C2, and C3 included in the specific image, and detects the respective numbers.

Here, the operation of the characteristic color counter 55 for one characteristic color C1 will be described. In addition, other characteristic colors,
The same applies to C2 and C3.

First, the characteristic color detector 70C1 compares the input red signal with the designated color signals r_ref1 and r_ref2 by the comparators 71a and 71b, and compares the input green signal with the designated colors by the comparators 71c and 71d. The signals are compared with the signals g_ref1 and g_ref2, and the input blue signal is compared with the designated color signals b_ref1 and b_ref2 by the comparators 71e and 71f. The designated color signals r_ref1, r_ref2, g_ref1, g_
ref2, b_ref1, b_ref2 are recognition CPUs
59 is set in the registers of the comparators 71a to 71f. Designated color signals r_ref1, r_ref2, g_r
ef1, g_ref2, b_ref1, b_ref2
Designates a color included in the specific image, is obtained in advance by statistically processing the color included in the target specific image, and is stored in the registration data storage memory 43. These designated color signals are generally obtained using a color used for a background color or a pattern of a specific image and distributed over a wide range, or a vermilion of an imprint. When specifying a color, each upper limit of RGB is set so that the specified color has a certain width.
The lower limit value is specified, for example, as r_ref1 (lower limit value for the R signal) and r_ref2 (upper limit value for the R signal), and pixels falling within these ranges are treated as specific color pixels. AND gate 72a outputs the product logic of the outputs of comparators 71a to 71f. Therefore, when the input image signal is within the range of the specific color C1, the outputs from the comparators 72a are all "1", and the output of the AND gate 72a is "1".

The counter 73a counts the number of pixels corresponding to the specific color C1 detected in this way. This counting is performed for each block. Here, a block is obtained by dividing a read image in units of a plurality of pixels in the main scanning direction and the sub-scanning direction. Here, 50 pixels are assigned to pixels of a predetermined resolution converted by the second resolution conversion unit 38.
A rectangular area of 50 × 50 pixels is defined as one block in units of pixels. Therefore, the counter 73a stores the count result in the count buffer 74 every time 50 pixels are input,
Reset the count value. The count result storage area of the count buffer 74 exists for the number of blocks in the main scanning direction for each of the specific colors C1, C2, and C3, and the count buffer 74 stores data for one block in the sub-scanning direction. When the count result input from the counter 73a is stored, the count buffer 74 always obtains an added value of the count value data already stored in the count buffer 74 and the newly input count result.
Update the count value to that value. That is, the count buffer 74 accumulates the number of characteristic color pixels of one block in the sub-scanning direction by performing a read-modify-write operation.

When data input for one block is completed in the sub-scanning direction, the count buffer 74 stores the stored contents, that is, the characteristic color C1 (and C1) obtained for each block.
The count result of (2, C3) is stored in the buffer 58 and passed to the template selecting unit 56.

The above operation is performed by the characteristic color detectors 70C2 and 70C7.
This is also performed in parallel at 0C3, and three characteristic colors C1, C2, and C3 are counted with respect to a predetermined designated color signal. Each count result is a three-dimensional vector, that is,
After being stored in the buffer 58 as a feature vector, it is passed to the template selection unit 56.

FIG. 13 is a diagram showing a data structure of data stored in the buffer 58. As shown in FIG.

In FIG. 13, thick solid lines indicate the boundaries of each block, and C ₁ (n), C ₂ (n), and C ₃ (n)
Are the characteristic colors C1 counted in the n-th block, respectively.
The count results of the pixels C2 and C3 are shown. That is, the data stored in the buffer 58 indicates that the feature data of one block is constituted by the number of pixels of three specific colors C1, C2, and C3.

Next, the operation of the template selecting section will be described.

FIG. 14 is a flowchart showing the operation of the template selection unit.

When the feature vector composed of the count values of the three feature colors C1, C2, and C3 from the feature color counter 55 is passed for each block, the template selection unit 56 starts comparing the feature vector with the template. .

First, the template selecting section 56 acquires a feature vector for each block from the counter buffer 74 (step S1). The acquired data is represented as C _n = (C ₁ (n), C ₂ (n), C
₃ (n)) (where n is the block number).

[0149] Next, the template selecting unit 56, the size of C _n | C _n | determines whether more than a predetermined value (step S2).

| C_nIf | is greater than a certain
The plate selection unit 56 stores the
From the stored template, C_nClosest to
Search for a thing (step S3). Template storage
The template for Mori 57 is T _m= (T_C1(M), T
_C2(M), T_C3(M)) (where m is the reference data number)
m = 1 to M), and the distance D_nm= |
C_n-T_m| (Euclidean distance of 3D vector)
D when also becomes smaller_nmIs detected, and the template number m and
Distance data D_minIs stored in the buffer 58 (step
S4).

In step S2, if | C _n | does not exceed the fixed value, the buffer 58 stores the template number (for example, M + 1) for which no template is defined and the value D _max that is greater than the maximum value that D _nm can take. (Step S5).

Here, the template stored in the template storage memory 57 will be described in detail.

The template storage memory 57 is a RAM, and the recognition CPU 59 copies the template data stored in advance in the registration data storage memory 43 to the template storage memory 57 when the power is turned on. The user obtains templates in advance from the target specific image and stores them in the registration data storage memory 43.

FIG. 15 is a diagram showing the relationship between a template and a specific image.

As shown in FIG. 15, the template is based on the specific image to be positioned at the horizontal position as a reference (see FIG. 15A). When rotated (FIG. 15B)
(See FIG. 15C), and when the positional relationship between the block and the specific image is shifted in the horizontal and vertical directions by several pixels (see FIG. 15D), each characteristic color C1,
The number of pixels corresponding to C2 and C3 is obtained. However,
Since the number of templates obtained as described above is enormous, a representative one is extracted by using a clustering technique such as vector quantization and stored in the registration data storage memory 43.

Next, the process of recognizing a specific image will be described.

In FIG. 8, the output of the template selecting section 56 is temporarily stored in a buffer 58. When these processes have been completed for a predetermined number of blocks, the main / sub-pixel counter 60 generates an interrupt signal 63. C for recognition
When the interrupt signal 63 is input, the PU 59 requests the buffer 58 for its contents. The recognition CPU 59 reads the contents of the buffer 58 and stores the contents in the work RAM 64.

FIG. 16 is a diagram showing a data structure in the working RAM 64. As shown in FIG.

In FIG. 16, TN (n) is a template number closest to the feature vector obtained for each block. D (n) is the distance D _min or D _max between each block and the template closest to the feature vector.

FIG. 17A is a diagram showing an image of an image including a specific image, and FIG. 17B is a diagram showing the value of TN (n) for each block of the actual specific image. FIG. 17C shows the value of D (n) for each block.

Here, the maximum template number is 254
The number not defined as a template is 254 + 1 = 255. In FIG. 17C, 0
A portion of 0 indicates _Dmax or a value close to _Dmax, and means that the image has no color similar to the specific image. In FIG. 17C, 02 is D _min ,
The value of the distance between the feature vector and the template is 0 or a value close thereto, that is, an image whose color is very similar to the specific image. Also, in FIG. 17C, the 01 portion means an intermediate value, that is, an ambiguous image.

Next, the frame mask will be described.

FIG. 18 is a diagram showing the structure of a frame mask.

The frame mask is a mask applied to blocks constituting a frame, and is shown in FIGS.
As shown in FIG. 7, a plurality of masks having different mask angles are prepared.

In FIG. 18, a hatched square indicates a mask block, and a white square indicates a non-mask block.
The mask data of the binary code with the latter being 1 is stored in the registration data storage memory 43 of FIG.

The recognition CPU 59 determines the presence or absence of a specific image based on the distribution state of TN (n) and D (n) developed on the work RAM 64 and the frame mask.

Next, a description will be given of a frame determination process performed first using D (n).

In the frame determination processing, a group of a plurality of adjacent blocks is regarded as one frame, and the processing is performed while the center position of the frame is shifted from the upper left of the input image horizontally and vertically in units of one block.

FIG. 19 is a flow chart showing the processing contents for one frame in the frame processing.

First, when the recognition CPU 59 starts frame processing for a certain frame, the work RAM 64
Is stored, the variable D _sum and B _{n um} be used in subsequent processing is initialized to 0 (step S10).

Next, the Euclidean distance D (n) between the feature vector and the selected template is read for the center block of the frame, and D (n) and the threshold Th1 are read.
Are compared (step S11).

If D (n) is larger than the threshold Th1 (the distance is long), the recognition CPU 59 determines that there is no specific image for this frame (step S2).
1), the process proceeds to the next frame.

D (n) is equal to or less than threshold value Th1 (the distance is short)
In the case of, the recognition CPU 59 acquires one of the frame masks from the registration data storage memory 43 (step S
12).

Next, the recognition CPU 59 sequentially checks, for each block, whether or not each block is a non-mask block in the frame mask obtained from the registered data storage memory 43 (step S13).

If the block inspected in step S13 is a non-mask block, the recognition CPU 59 obtains D (n) of the corresponding block from the work RAM 64 (step S14). Next, the recognition CPU 59
The obtained D (n) is stored in the D
_{The sum} is added to _sum , and the value is updated (step S15).

The counter value B _num for counting the number of blocks subjected to the processing is incremented (step S16).

If the block inspected in step S13 is a mask block, the recognizing CPU 59 repeats step S13 for the next block without doing anything.

The above steps S13 to S16
The above processing is performed for all the blocks constituting the frame (step S17).

Next, the recognition CPU 59 calculates the average distance D _mean from the block number counter value B _num and the sum D _{sum of} distances.
= D _sum / B _num is obtained (step S18), and is compared with the threshold value Th2 (step S19).

If D _mean ≦ Th2, it is determined that there is a high possibility that the specific image is included in the image (step S2).
2).

When D _mean > Th2, it is determined that there is no specific image, and the recognition CPU 59 repeats steps S12 to S19 for the next frame mask (step S20).

In the above frame determination processing, when it is determined that the specific image is likely to be included in the image being processed, a final determination is made.

Hereinafter, the final judgment processing will be described with reference to FIGS.
1 will be described.

FIG. 20 is a diagram showing the rotation angle correction in the final determination, and FIG. 21 is a diagram showing the relationship between the frame, the block, and the recognition processing in the final determination by the recognition CPU.

For the final decision, TN (n) in which a template number whose three-dimensional Euclidean distance is closest to the feature vector of each image block is described. However, the image cut out by the frame determination processing may include a specific image that is not in the normal arrangement.

Step S1 of the above-described frame determination process
In step 9, the angle at which the specific image is arranged can be estimated at the frame determination stage depending on the type of the frame mask used when it is determined that the specific image is likely to be included in the image. T based on this information
N (n) itself is relocated to the normal position.

FIG. 20 shows the situation, and FIG.
0 (a) shows the state before rotation correction, and FIG. 20 (b) shows the state after rotation correction. In each of the figures, the positions of ○, □, and Δ correspond to each other.

After the rotation correction is performed, a final determination process is performed.

In FIG. 21, the symbols ○, □, and Δ correspond to those in FIG. 20, respectively. In FIG. 20B, the block after the normal placement is partly in a stepped shape, but in the final determination, as shown in FIG. 21, the processing is performed assuming that the specific document has been placed correctly.

In FIG. 21, reference numeral 75 designates a value of TN (n) other than 255 (= template undefined) among the blocks included in the frame determined to have a high possibility that the specific image exists in the frame determination. It is a set of blocks. Reference numeral 76 denotes blocks included in the block set, and a template is described as a histogram for each block for ease of explanation. Reference numeral 77 denotes a final decision unit constituted by a neural network, 78 denotes an input layer of the neural network of the final decision unit 77, 79 denotes an intermediate layer of the neural network of the final decision unit 77, and 80
Is the output layer of the neural network of the final determination unit 77,
Reference numeral 81 denotes a comparison unit that compares two outputs output from the output layer 80 and selects a larger one.

Each block converted to the normal arrangement position by the above rotation correction has a template number TN.
(N). The template number is actually
Since it is the feature vector itself included in the specific image,
These can be represented by histograms, as in block 76 of FIG. The frequency of this histogram is input to the input layer 78 of the final determination unit 77. The input layer 78 has three nodes for one input unit in order to correspond to the three-dimensional information included in the histogram, and the frequency of the histogram is input to each node.

The frequency is input to the nodes corresponding to all the blocks. In the neural network, the arithmetic operation is performed in the intermediate layer 79 by weighting operation learned in advance, and the output layer 80 outputs a degree that looks like a specific image and a degree that does not look like a specific image. Finally, in the comparing means 81,
Select the one that output a greater degree and output it. Therefore, the output of the comparing means 81 is a binary output indicating whether or not the input image is a specific image. With the above operation, the specific image can be detected.

As described above, it is determined whether or not a specific image exists in an image read by an image reading apparatus. However, this type of recognition always involves an erroneous determination. In particular, in the case of a device that detects a specific image for which copying is prohibited, if a general image is recognized as a specific image, there is a problem that an image for which copying is not normally prohibited cannot be copied.

Hereinafter, this solution will be described with reference to FIG.

In FIG. 8, the specific image recognizing section 37 has a memory 54, and the RGB image signal converted to a predetermined resolution by the second resolution converting section 38 is temporarily stored in the memory 54.

In step S19 of the frame determination processing, the angle at which the specific image is arranged in the frame determination stage depends on the type of the frame mask used when it is determined that the specific image is likely to be included in the image. You can get an idea. In addition, at the stage of applying a frame mask by performing a shift or the like on an image, it is possible to give an estimate of the coordinate information of the specific image.

The recognition CPU 59 is configured with hardware so as to be able to access the memory 54. When the final image is recognized as a specific image in the final determination, the recognition CPU 59 performs the processing based on the position and rotation angle information. By reading the corresponding address in the memory 54, it is possible to obtain, for a specific portion of the specific image, structural information on a printing net or the like or more detailed color information.

By performing the re-determination based on these pieces of information after the detailed determination, the possibility that an erroneous determination will occur can be greatly reduced.

By the above operation, it is determined whether or not the read image includes the specific image.

Next, the operation after the judgment result is output will be described.

The recognition CPU 59 transfers the determination result to the CPU 32 using the serial communication line 40. CPU
32 controls the interface 30 and notifies other devices 31 of the determination result. In the case of a general-purpose interface such as SCSI, an image reading device sends an Abor to another device.
t may be output to forcibly receive the result, or a command for another device to receive the result may be output to the image reading apparatus.

The determination operation is started when the reading operation of the image reading device 1 is completed. At the time when the determination is started, the image data read by the image reading device 1 is transferred to the host computer 3 or the image recording device. 2 has already been transferred to another device 31. Other devices 31
, The image processing and the recording operation are stopped until the judgment result is received. As described above, the determination result of the specific image is notified from the image reading device 1, and the other devices 31
Accordingly, for example, it is determined whether or not to shift to the image recording operation.

Next, an example of a specific image registration operation will be described.

FIG. 22 is a flowchart showing the means for registering a specific image.

The registration data storage memory 43 stores the feature amount of a specific image to be registered at the time of registration (see FIG. 8). The feature amount stored in the registration data storage memory 43 is read by the recognition CPU 59 at the time of recognition, is stored in the work RAM 64, and is referred to at the time of recognition.

The user inputs recognition mode information for selecting a registration mode to the image reading device 1 from a mode selection button (not shown) on the input panel provided on the host computer 3 or the image reading device 1 or the like. Then, the registration mode is selected by the mode selection means 41 of the image reading apparatus 1.

In this state, the user first reads the specific image to be registered by the reading device 1 (step S3).
0).

The second resolution converter 38 converts the read image into a predetermined resolution (step S31), and stores the converted image data in the memory 54 (step S3).
2). At this time, the position of the specific image placed on the platen may be at any position, but the angle is 0 degree, that is, the main scanning direction of the reading device and the long side of the specific image are placed in parallel. Read. This is the image reading device 1
Is not necessary when the position of the input image is invariable as in a money changer, but when the image reading device 1 is a copying device or the like,
Since the location of the input image is not specified, the image is read at an angle of 0 °, a white background is detected, and only meaningful data other than the background is stored in the memory 54.

Next, the recognition CPU 59 selects whether or not to identify a specific image based on the data stored in the memory 54 (step S33).

If the specific image is not identified in step S33, the feature quantity to be registered of the image stored in the memory 54 is extracted (step S34).

Next, the user operates registration data selection means (a selection screen displayed on the monitor screen of the host computer 3 or a selection button (not shown) on an input panel provided in the image reading apparatus 1). New registration means for newly registering a feature amount (a new registration screen displayed on a monitor screen of the host computer 3 or an image reading apparatus 1)
A new registration button (not shown) on the input panel provided in the printer, and a rewriting registration unit (monitor screen of the host computer 3) for additionally changing the feature amount of the specific image already stored in the registration data storage memory 43. (Step S35), and an additional registration screen displayed on the screen or a rewrite registration button (not shown) on an input panel provided in the image reading apparatus 1.

When the new registration means is selected by the registration data selection means, the recognition CPU 59 proceeds to step S3.
The feature amount extracted in step 4 is additionally registered in the registration data storage memory 43.

(Table 4) shows the structure of data stored in the registered data storage memory 43.

[0214]

[Table 4]

If the new registration means is selected, the recognition C
The PU 59 stores the registration data storage memory 4 shown in (Table 4).
The extracted characteristic amount is newly stored in the characteristic amount X1 of the new data X (step S36).

When the rewriting registration unit is selected by the registration data selection unit, the user can select a specific image selection unit (a selection screen for specific image information displayed on the monitor screen of the host computer 3 or an image reading device). 1) on the input panel provided in 1)
A specific image to be registered is selected from the existing specific images in the registration data storage memory 43 shown in (Table 4) (step S3).
7), the recognition CPU 59 rewrites and registers the feature amount extracted in step S34 for the selected specific image (step S38).

For example, when the specific image selected by the specific image selecting means is the registered data B, the recognition CPU
Reference numeral 59 changes the feature amount B1 of the registered data storage memory shown in (Table 4).

In step S33, when identifying a specific image, a specific image to be registered is selected based on the identification result.

If the identification result does not match any of the existing registered specific images (step S40), the new registration means uses the registration data storage memory 43 shown in (Table 4).
In, the feature value obtained in step is newly stored in the feature value X1 of the new data X (step 31).

In step S40, if the identification result matches the already registered specific image, for example, if it matches the specific image of the registered data B, the rewriting registration unit registers the registration data shown in (Table 4). The feature amount B1 of the data storage memory 43 is additionally changed (step S42).

Finally, a method of extracting a characteristic amount and a characteristic amount will be described.

If the input image data is directly registered as the data to be registered, the capacity of the memory becomes extremely large. Therefore, a feature amount other than the image data is registered.

First, as a feature amount for performing recognition, the first
First, a plurality of colors that clearly represent the features of the specific image are extracted. The characteristic color extraction method is described in the detailed description of the characteristic color counter 55 described above.

As a second feature, a frame mask as shown in FIG. 18 is created. Angle 0 in registration mode
Read by degree. Based on the image data at the angle of 0 degree, the image data is rotated at an arbitrary angle, for example, every 5 degrees to create frame mask data.

As the third feature, template data is created. The creation method is the same as the template selection unit 5 described above.
6 is described in the detailed description.

The color range of the characteristic color is extracted as the fourth characteristic amount. The color range is, for example, 20% of the value of the characteristic color. For example, when the value of the characteristic color green is 180, the color range at this time is ± 36.

As the fifth feature value, a weight coefficient in the intermediate layer 79 of the neural network shown in FIG. 21 is extracted. The weighting factor is obtained by neurally distinguishing the feature amount of the specific image from the non-identification image feature data that is a feature amount other than the specific image in order to clarify whether the template data as the third feature amount is a specific image or not. Learn by network.

In learning a neural network,
It is not possible to recognize whether or not the image is a specific image only with the feature amount of the specific image. By using data other than the specific image and comparing it with the data, it is possible to determine the likeness of the specific image. At this time, if the non-identified image feature data is completely different from the feature amount of the specific image, there is no learning effect by the neural network, and it becomes difficult to identify the vague image. Therefore, based on the pattern of the template data that can be taken by the specific image using the mask block, the pattern of the template data that cannot be taken by the specific image is used as the feature of the non-identified image feature data by the frame mask that is the second feature. An image corresponding to each specific image to be stored or selected from the non-identified image feature data is selected.

For example, the pattern of template data that can be taken by a specific image is randomly replaced in a mask block. In addition, a template data pattern that can be taken by a specific image is deleted from an arbitrary combination of template data, and a template data pattern whose total count is closer than a certain value is selected as non-identification image feature data.

From the specific image thus input,
Unidentified image feature data corresponding to the specific image is created, and when the specific image is registered data B in (Table 4), the created data is registered in the non-identified image feature data B2. As a result of learning the feature amount of the specific image and the non-identified image feature data which is a feature amount other than the specific image by the neural network, a weight coefficient in the intermediate layer 79 of the neural network in FIG. 21 is obtained. A recognition parameter is also obtained as the feature amount of No. 6.

As described above, a plurality of characteristic colors, frame masks, template data, color ranges,
A plurality of feature amounts such as weight coefficients and recognition parameters are stored in the registration data storage memory 43 of FIG. 8 as registration data.

[0232]

As described above, according to the image recognition apparatus of the present invention, the following advantageous effects can be obtained.

According to the first aspect of the present invention, it is possible to provide an image identifying apparatus capable of registering a specific image according to a use environment at any time.

According to the second aspect of the present invention, it is possible to provide an image identification device capable of adjusting the identification accuracy.

According to the third aspect of the present invention, it is possible to provide an image identification device capable of preventing a plurality of same type specific images from being registered by mistake.

According to the fourth aspect of the present invention, it is possible to clarify the difference between the image data to be identified and the other image data, and to provide an image identifying apparatus with high identification accuracy.

According to the fifth aspect of the present invention, it is possible to provide an image identification device capable of reducing the data amount of a non-identification image and reducing the capacity of a registration data storage memory.

According to the sixth aspect of the present invention, there is no need to register non-identifying image feature data in advance, so that the amount of data can be greatly reduced and the capacity of the registered data storage memory can be reduced. An identification device can be provided.

[Brief description of the drawings]

FIG. 1 is a diagram showing an image copying system using an image identification device of the present invention.

FIG. 2 is a sectional view of a main part of the image reading apparatus of the image copying system of FIG. 1;

FIG. 3 is a sectional view of a main part showing an internal structure of a carriage of the image reading apparatus of FIG. 2;

FIG. 4 is a schematic view of the carriage of the image reading apparatus when viewed from the side.

FIG. 5 is a perspective view showing details of an optical system of the image reading apparatus.

FIG. 6 is a block diagram illustrating a configuration of an image data processing circuit of the image reading device.

FIG. 7 is a schematic plan view of the image sensor viewed from the line sensor array side.

FIG. 8 is a block diagram showing the structure of an image identification device.

FIG. 9 is a block diagram showing a configuration of a characteristic color counter of FIG. 8;

FIG. 10 is a diagram showing the structure of data stored inside a line correction unit.

FIG. 11 is a diagram illustrating an operation of a line correction unit when an image is read at a resolution of 300 dpi in the sub-scanning direction.

FIG. 12 is a diagram showing a resolution conversion algorithm.

FIG. 13 is a diagram showing a data structure of data stored in a buffer.

FIG. 14 is a flowchart showing the operation of a template selection unit.

FIG. 15 is a diagram showing a relationship between a template and a specific image.

FIG. 16 is a diagram showing a data configuration in a work RAM;

17A is a diagram showing an image of an image including a specific image. FIG. 17B is a TN (n) for each block of an actual specific image.
(C) Diagram showing the value of D (n) for each block

FIG. 18 is a diagram showing a structure of a frame mask.

FIG. 19 is a flowchart showing processing contents for one frame in frame processing;

FIG. 20 is a diagram showing rotation angle correction in a final determination.

FIG. 21 is a diagram illustrating a relationship between a frame, a block, and a recognition process in a final determination of the recognition CPU;

FIG. 22 is a flowchart showing a specific image registration unit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Image reading device 2 Image recording device 3 Host computer 4 Cable 5 Main body 6 Original glass 7 Carriage 7a Opening 8 Drive source 9 Drive pulley 10 Timing belt 11 Belt 12 Follower pulley 13 Document 14 Document cover 15 Support 16 Reference acquisition position 16a Reference plate 17 Lamp 18 Aperture 19-1 and 19-2 Reflecting mirror 20 Image sensor 21 Imaging lens 22R, 22G, 22B Line sensor array 23R, 23G, 23B Reading line 24a Amplifier 24b A / D converter 25 Shading correction unit 26 Line correction unit 27 First resolution conversion unit 28 Color processing unit 29 Buffer 30 Interface 31 Other device 32 CPU 33 Motor control unit 34, 35, 36, 36a Control signal 37 Specific image recognition unit 38 Second Image conversion unit 39 Recognition unit 40 Serial communication line 41 Mode selection unit 42 Registration unit 43 Registration data storage memory 50, 51 Memory area 53 Pixel 54 Memory 55 Characteristic color counter 56 Template selection unit 57 Template storage memory 58 Buffer 59 Recognition CPU Reference Signs List 60 main / sub-pixel counter 61 ROM 62 control signal 63 interrupt signal 64 work RAM 70C1, 70C2, 70C3 characteristic color detectors 71a, 71b, 71c, 71d, 71e, 71f, 7
1g, 71h, 71i, 71j, 71k, 71l, 71
m, 71n, 71o, 71p, 71q, 71r Comparator 72a, 72b, 72c AND gate 73a, 73b, 73c Counter 74 Count buffer 75 Set of blocks 76 Block 77 Final decision unit 78 Input layer 79 Middle layer 80 Output layer 81 Comparison Means d1 Sub-scanning direction d2 Main scanning direction P200_0, P200_1, P200_2 200d
pi pixel P300_0, P300_1, P300_2, P300
_3 300 dpi pixel P400_0, P400_1, P400_2, P400
_3, P400_4 400 dpi pixels P500_0, P500_1, P500_2, P500
_3, P500_4, P500_5 500 dpi pixels P600_0, P600_1, P600_2, P600
_3, P600_4, P600_5, P600_66
00 dpi pixel po1 home position

Continued on front page F-term (reference) 5B057 BA26 CC02 DA12 DB02 DB06 DC01 DC25 DC33 DC36 DC39 DC40 5C072 AA01 BA20 KA01 LA00 QA00 XA01 5L096 AA02 AA06 BA07 BA18 FA15 GA19 GA40 HA08 JA11 KA04 KA15 9A001 BB02 BB03 BB02 BB03 BB02 BB03 DD03 HH24 HH29 HH31 JJ64 KK16 KK37 KK42 KK58 LL02 LL03

Claims (6)

[Claims] An extracting means for dividing an input image into blocks of a predetermined size and extracting a feature amount for each block, and a specific image for the input image based on the feature amount extracted by the extracting means. Identification means for identifying whether or not a specific image is included; registration means for registering a feature amount of a specific image extracted for each block unit by the extraction means; and a feature amount of the specific image registered by the registration means. An image identification apparatus, comprising: a registration data storage memory for storing; and a mode selection unit for selectively operating the identification unit or the registration unit. 2. The registration unit according to claim 1, wherein said registration unit newly registers a feature amount of said specific image extracted in block units by said extraction unit into said registration data storage memory, and said registration unit stores said feature amount in said registration data storage memory. A specific image selecting means for selecting the specific image; and a rewriting registration for rewriting and changing a characteristic amount of the specific image selected by the specific image selecting means to the characteristic amount of the specific image newly extracted by the extracting means. 2. The image identification apparatus according to claim 1, further comprising: a registration unit that selectively operates the new registration unit or the rewrite registration unit. 3. When the input specific image includes the registered specific image, the rewriting registration unit is operated, and the input specific image does not include the registered specific image. In the case, the apparatus further comprises the registration data selection means for controlling the operation of the new registration means, wherein the identification means identifies whether or not the input specific image includes the registered specific image. Claim 2
An image identification device according to claim 1. 4. The image identification apparatus according to claim 1, wherein the registration unit registers a feature amount of an image other than the specific image to be identified as non-identification image feature data. apparatus. 5. The non-identifying image characteristic data corresponding to the specific image to be identified is selected from the registered non-identifying image characteristic data for each of the specific images. The image identification device according to claim 4, wherein: 6. The registration means according to claim 4, wherein said registration means creates said non-identification image characteristic data corresponding to said specific image to be identified from said specific image to be identified and stores it in said registration data storage memory. An image identification device according to claim 1.