JPH118763A - Image reading device - Google Patents

Abstract

(57) [Summary] When reading an image of a document, a variation in output due to shading distortion caused by a change over time of a device or a variation in light amount is corrected. SOLUTION: When read image data (1B) is input to a shading correction unit 18, the shading correction unit 18 changes with time of an image reading apparatus and variation in light amount of an optical system for reading an image. And outputs the image data as normal image data 1C. The shading correction unit 18 includes a neural network operation unit 43 and a storage unit 44 that stores shading correction data for correcting shading distortion at each position of the document table (11). When the position coordinates of the document table are input to the storage unit 44, shading correction data capable of outputting a shading correction value at that position is stored in advance, and the shading correction process is performed using the neural network operation unit 43. A stable shading correction is realized for the entire document surface.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image reading apparatus which is mounted on a digital copying machine, a facsimile, a color scanner or the like, reads an image of an original to be read by optical scanning, and outputs the image in the form of an electric signal. Related to the device.

[0002]

2. Description of the Related Art Conventionally, as an image reading apparatus, an original placed on a transparent original table surface mainly made of glass is optically scanned with light emitted from a light source, and the resulting reflection is obtained. When reading a color image with light, the R
(Red) Decomposes into three components of G (green) and B (blue) light and forms an image on a light receiving part of a photoelectric conversion element such as a CCD which is a reading element, and these are analog of each component of R, G and B. In general, an electric signal is converted into an electric signal, and the electric signal is outputted in the form of a digital electric signal through means such as an A / D converter.

Here, in the above-described image reading apparatus, the light amount unevenness of the light source, the variation of the conversion characteristic of each pixel of the photoelectric conversion element, the fluctuation of the light amount of the light source due to the voltage fluctuation inside the apparatus, and the photoelectric conversion due to aging use. Even if an image of the same density is formed on the photoelectric conversion element by optical scanning due to various factors such as a change in the conversion characteristic of the conversion element and a variation in the amount of transmitted light of the imaging lens, the output electric signal from the photoelectric conversion element Does not always have a constant value, and a phenomenon of constantly varying shading distortion occurs.

[0004] When reading an image of the same density while suppressing variations in output signals as much as possible against such a shading distortion phenomenon, a constant output electric signal is always obtained irrespective of the above factors. The correction to be performed is called shading correction, and various attempts have been made.

[0005] Above all, a commonly used correction white plate (hereinafter referred to as a reference white plate) is provided at a position other than the platen in the movement path of an optical system for optically scanning and reading an image of a document. Then, before the original is optically scanned, the reference white plate is optically scanned in advance to make the photoelectric conversion element read the image,
Based on the read image data, a correction coefficient for each of the R, G, and B component signals is calculated for each photoelectric conversion element constituting each pixel, and the correction coefficient is added to the digital electric signal of the read read image data for correction. It is a method of performing.

[0006] Japanese Patent Application Laid-Open No. 7-336535 discloses such an example.
In the publication, a bar code representing side color data is attached to a reference white plate provided for each photoelectric conversion element, and the bar code data is read and an optical scan of the reference white plate is performed before reading an original image. There is disclosed a method of performing stable shading correction while suppressing density variations at the time of manufacturing a white plate.

On the other hand, in Japanese Patent Application Laid-Open No. Hei 8-172553, the above-mentioned reference white plates are dispersedly arranged on a scanning line of a document table, and a signal obtained by reading the signals of these reference white plates has no reference white plate therebetween. A method of calculating a shading correction value of a portion by performing an estimated straight line calculation is disclosed.

[0008]

In the above-described shading correction, by performing correction on actual read image data in consideration of the output state of the white reference plate, a certain effect, that is, image Data can be output at the same density throughout.

However, in the shading described in Japanese Patent Application Laid-Open No. Hei 7-336535, the dispersion of output signals between photoelectric conversion elements can be suppressed to some extent, but the dispersion over the entire document reading surface cannot be suppressed. . That is, when an original on the original platen is optically scanned, various factors such as a change in the light amount of the light source and a change in the angle of the scanning light illuminated from the light source with respect to each scanning line cause an image of the read original to be scanned. Even if images having the same density exist, a phenomenon occurs in which the value of the output signal changes depending on the location.

In the shading correction described in JP-A-8-172531, it is very difficult to perform high-precision shading correction on a portion having no reference white plate.

Therefore, it is very difficult to always perform good shading correction on the entire document table by any of the shading correction methods and means, and some shading distortion is inevitably generated depending on the position of the document table. Would. In addition, before reading an image of a document, it is necessary to read a reference white plate or the like each time and perform a shading correction process based on the reference white plate, and the reading time tends to be long.

An object of the present invention is to provide an image reading apparatus which takes into account various factors such as a change in characteristics of a photoelectric conversion element and a light source which generate shading distortion and performs stable shading correction over the entire document table. .

In particular, an object of the present invention is to perform shading correction, and when an image of the same density is placed on a document table, a constant density is obtained regardless of a change over time, a change in light amount of a light source, and other changes. Is output.

[0014]

According to a first aspect of the present invention, there is provided an image reading apparatus for optically scanning an image of a document placed on a document table. The reflected light of the image is converted into an electric signal, and the image is formed on a photoelectric conversion unit that outputs the image as read image data. In an image reading apparatus provided with a correction means for performing shading correction processing based on a shading correction value such that an output state of the same density becomes constant, a plurality of arbitrary points on the platen are designated for shading correction. The correction means may be configured to perform the photoelectric conversion based on the same reference density at the plurality of designated points for shading correction with respect to the reference data having the same reference density in advance. The shading correction read data read by the conversion means obtains a shading correction value serving as the reference data, obtains a shading correction value between the positions of the plurality of designated points, and obtains the read image based on the shading correction value. The shading correction processing of the data is performed, and the image data is output as image data. By doing so, the entire area of the platen is complemented by complementing the shading correction values at a plurality of arbitrary points. A shading correction value that covers the entire area is obtained. Thereby, shading correction processing can be performed on the read image data of the document. That is, the shading correction processing can be performed so as to cover the entire area of the document table. In particular, at any point,
The shading correction can be stabilized by designating and inputting a position where shading correction is required, for example, a change in the scanning position of the optical scanning unit or a position where the output state changes due to aging.

Therefore, in the image reading apparatus having the above-described configuration, according to the second aspect of the present invention, the correcting means includes:
Using a neural network, position information of an arbitrarily designated point on the platen is used as an input element, and a shading correction value obtained from the reference data and the shading correction reading data according to the position information is used as a teacher signal. Learning is performed, an engagement coefficient at which the learning result finally becomes a shading correction value is obtained as shading correction data, and a storage unit for storing the shading correction data and each position information of the read image data are added to the input layer. A neuro operation unit that outputs a shading correction value corresponding to the shading correction data stored in the storage unit from an output layer, and a shading correction process of the read image data that corresponds to the shading correction value output from the neuro operation unit And a correction operation unit for performing .

If the shading correction process is performed using the neural network as described above, the conversion characteristics of each element of the photoelectric conversion means, the light amount unevenness of the light source for optical scanning, and various factors of shading distortion are reflected. Thus, shading correction can be performed at each position on the document table. In addition, by using a neural network for shading correction, it is possible to limit information necessary for shading correction to only equations for a weight coefficient (engagement coefficient) for neural network calculation and a response function. Further, since shading correction data is calculated at any of a plurality of points (positions) specified in advance on the document table, it is possible to output the shading correction value between the points by analogy. Further, it is not necessary to optically scan an image having the same density, for example, a reference white plate before reading an image of a document. Therefore, the reading time of the document can be reduced.

According to the third aspect of the present invention, in the image reading apparatus having the above-described configuration, the storage unit sets an arbitrary area on the document table in advance, and sets the area in accordance with each area. The shading correction data learned and calculated by the neural network is stored separately.
The correction means includes a discriminating unit for discriminating a preset area on the platen based on the read image data and position information corresponding to the data, wherein the discriminating unit determines a predetermined area on the platen according to the position information. The correction unit outputs the shading correction data from the storage unit to the neuro calculation unit according to the determination result of the determination unit, and performs the shading correction process for each of the predetermined regions. It is characterized by having done.

According to such a configuration of the correction means, the shading correction by the neural network can be performed as described above, and at the same time, the fine shading correction processing can be performed for each area on the document table. For example, a central portion on a document table and a region around the central portion are set separately, and shading correction processing can be performed for each region.
Therefore, more accurate shading correction processing can be performed. In addition, the region may be set arbitrarily, and can be set arbitrarily so as to take into consideration fluctuations in the scanning position of the optical system, and more accurate shading correction becomes possible.

Further, in the image reading apparatus having the above-described configuration, according to the fourth aspect of the present invention, in addition to the position information of a plurality of arbitrarily designated points on the document table, the time elapsed since the power of the image reading apparatus is turned on. Learning is performed based on the shading correction value at each designated point using a neural network for shading correction using time as an input element, and the shading correction data obtained as a result of the learning is stored in a storage unit. Thus, shading correction from power-on until power supply voltage is stabilized can be performed. That is, it is possible to output shading correction data reflecting various factors such as the internal voltage of the image reading apparatus in a relatively short period such as several hours after the power of the image reading apparatus is turned on. Thus, even if the light amount or the light amount distribution of the light source of the optical scanning system changes due to the voltage fluctuation inside the apparatus after the power is turned on, it is possible to always perform stable shading correction at each pixel point on the document table.

According to the fifth aspect of the present invention, in addition to the position information of a plurality of arbitrarily designated points on the platen, the elapsed time in the operation state of the image reading apparatus is provided. The learning based on the shading correction value at each designated point is performed using a neural network for shading correction using as an input element, and the shading correction data of the learning result is stored in the storage unit, so that it can cope with a temporal change. Shading correction processing can be performed. In other words, it is reflected on various factors such as the conversion characteristics of photoelectric conversion elements such as CCDs and the characteristics of light sources, whose characteristics change over time due to the reading operation of the image reading device, for example, the characteristics change over a long period of time such as several months or several years. The shading correction process can be performed. Therefore, even if a characteristic change occurs due to the continuous use of the photoelectric conversion element and the light source, it is possible to always perform stable shading correction at each pixel point on the document table.

According to the image reading apparatus having the above-described configuration, in addition to the position information of a plurality of arbitrarily designated points on the document table, the image reading apparatus according to the sixth aspect of the present invention can also detect the environmental change around the image forming apparatus. The learning based on the shading correction value at each designated point is performed using a neural network for shading correction using the detection signal as an input element, and the shading correction data of the learning result is stored in the storage unit, and the learning result is stored in the storage unit. Shading correction that has been dealt with can be made possible. That is, environmental changes around the image reading device,
For example, it is possible to output shading correction data reflecting various factors such as clouding due to water droplets on the inner surface of the document table due to a change in humidity. Therefore, stable shading correction can be always performed at each pixel point on the document table even when clouding or the like due to water droplets occurs on the inner surface of the document table due to the humidity in the image reading apparatus.

Finally, in the image reading apparatus having the above-described structure, according to the invention of claim 7, in addition to the position information of a plurality of arbitrarily designated points on the document table, the position of the image reading apparatus from when the power of the image reading apparatus is turned on. Counting means for counting the time of reading, and second counting for counting the reading operation time of the image reading apparatus.
The learning means based on various shading correction values at designated points using a neural network for shading correction, in which each signal from the detection means for detecting environmental changes around the image reading device is used as an input element. Then, each shading correction data of the learning result is stored in the storage unit, and it is arbitrarily determined whether any one of the signals from the first and second counting means and the detection sensor is added to the input element. , And any of the shading corrections can be arbitrarily performed. Therefore, if the input element of whether to add to the input layer of the neural network operation unit can be arbitrarily selected from the outside, according to the situation of the image reading apparatus when actually reading the image and the accuracy required by the input image, It is possible to select whether or not to perform the shading correction in consideration of the above input elements. Therefore, when outputting image data that does not require high-precision shading correction, the processing time can be significantly reduced by selecting a neural network for shading correction that does not consider these input elements. Become.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Various embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing details of a shading correction unit using a neural network for performing shading correction processing by the image reading apparatus of the present invention, and FIG. 2 is a plan view showing a transparent platen for placing a document. In particular, an example for arbitrarily setting and inputting a coordinate point for shading correction according to the present invention is shown. FIG. 3 is an explanatory diagram showing an example of a neural network used for shading correction in FIG. FIG. 4 is a schematic diagram of the overall structure of an image reading apparatus including the shading correction unit shown in FIG. 1 and a block diagram for performing shading correction processing on read image data.

First, referring to FIG. 4, a schematic configuration of an image reading apparatus according to the present invention will be described. The image reading apparatus according to the present invention converts a read document placed on a transparent platen 11 into a platen 1.
1 is scanned by scanning light from a light source 12 disposed on the lower surface of the light source 1, and the reflected light is reflected by reflecting mirrors 13a, 13b, 13b.
The light is guided to the imaging lens 14 by c. Imaging lens 14
, That is, the image of the reflected light from the original is formed on a CCD element 15 which is a photoelectric conversion element. CCD
A CCD driver 16 is connected to the element 15.
An analog electric signal 1A (each color signal Ra, Ga, Ba) indicating the density of each of the R, G, and B components is output via the driver. The analog signal 1A is then supplied to the A / D converter 1
7, and the conversion unit outputs each of the R, G, and B components as, for example, an 8-bit digital electric signal (read image data) 1B (Rd, Gd, Bd).

Next, the digital signal 1B is transmitted to the shading correction unit 18. The read image data 1B, which is a digital signal, is supplied to the shading correction unit 18
After performing the shading correction processing according to the present invention, the corrected image data 1C (Rd1, Gd1, Bd
Output to the external device as 1). For example, in the case of a digital copying machine, the data is transferred to a color image forming unit such as a laser printer. The data is transferred to an external color printer or the like.

The CCD element 15 is configured as a line sensor. That is, the photoelectric conversion elements of the CCD element 15 are arranged in the X direction with respect to the document table 11 in FIG. Further, the A / D conversion unit 17 and the shading correction unit 18 are each configured by a dedicated IC circuit.

In the image reading apparatus of the present invention, the light source 12 and the reflection mirrors 13a, 13b, 13c move in the directions indicated by arrows, thereby optically scanning the original line by line, and the reflected light by the obtained image. Is repeated by the above method, the image of the document can be read, and the electrically converted image data can be output.

Further, the shading correction unit 18 according to the present invention calculates a shading correction value, performs shading correction processing on the read image data 1B, and outputs the corrected image data 1C. I have. Therefore, at the initial stage, that is, before shipment of the product, the reference white plate 10 is placed on the document table 11 in advance, and read data serving as a reference for shading correction from the CCD device 15 by the reference white plate 10 is obtained. deep. Then, a shading correction value is determined in advance so that the data becomes the same with respect to the reference read data serving as the reference. In order to obtain this shading correction value, a detailed description will be given in the following embodiments. As a result, a result of calculation in advance by a neural network, that is, an engagement coefficient which becomes the above-mentioned shading correction value is stored as shading correction data. With the shading correction value obtained based on the shading correction data, the read data from the CCD element 15 is subjected to shading correction processing,
Output as image data.

The shading correction section 18 executes a shading correction process using a neural network in accordance with shading correction data obtained in advance.
Therefore, the shading correction read data from the CCD element 15 by the reference white plate 10 and the reference white plate 10
For example, a comparison with reference data (colorimetric data) such that all areas have the same output, for example, a difference is determined as a shading correction value, and based on the shading correction value, the shading correction data is Ask. That is, the light source 1 will be described in detail later.
The shading correction unit 18 obtains in advance the shading correction data to obtain the same output data based on the shading correction value in accordance with the shading correction value in accordance with the light amount unevenness, the voltage fluctuation, the aging change, etc. , A shading correction process based on data obtained by actually reading an image of a document.

In this manner, the shading correction data is obtained. When the document to be read is actually placed on the document table 11, the read image data 1B output from the CCD element 15 is obtained first. A shading correction process is performed using a shading correction value based on the shading correction data, and finally the digital signal 1C after the shading correction is output as image data.

By doing so, the CCD device is protected against voltage fluctuations, light amount unevenness of the light source, differences in light amount distribution, changes over time of the image reading device, and variations in the output of each pixel of the CCD device. Even when the output from the image 15 is not constant for an image of the same density, it is possible to always output a stable image data in a constant state.

(First Embodiment) Therefore, in the present invention, if shading correction is performed on the document table 11 over the entire area, it takes a very long time and a large-capacity storage unit is required. Therefore, a plurality of arbitrary points on the platen 11 are set, a shading correction value at the set point is obtained, and shading correction data is calculated in advance using a neural network based on the shading correction value. The shading correction processing is performed based on the data. As a result, the capacity of the storage unit can be reduced. Also, the processing time is not required as much. In addition, a shading correction value is output each time an actual document is scanned for reading an image, thereby enabling correction using the shading correction value. Further, before reading the document, it is not necessary to scan the reference white plate or the like each time to obtain a shading correction value or the like, so that the reading time of the document from the first sheet can be shortened.

Hereinafter, the shading correction processing of the present invention will be described step by step. First, an example for obtaining shading correction data will be described in detail. A reference white plate 10 for shading correction is installed on the entire document surface of the document table 11 shown in FIG. As the reference white plate 10, as described above, a white plate with a document cover (not shown) may be used. Then, at the initial stage, the colorimetric data (for a color image, R, R of reflected light,
The above-described reference read data obtained by converting the values obtained by reading the G and B components into digital data are obtained.

FIG. 2 is a plan view of the document table 11 shown in FIG. A plurality of arbitrary pixels on the document table 11 are designated in advance as measurement points. 2, reference numerals 21, 22, 23
Is an example of such a measurement point. The number of these measurement points is
It is optional and may be set as appropriate. Further, as the number increases, shading correction according to the present invention described below becomes more accurate. On the other hand, the calculation time by the neural network is longer, but since this is performed in advance before shipping, when performing shading correction on the read data after actually reading the image of the original, the processing time is It will not be long.

Here, if an arbitrary point on the document table 11 is designated as the origin, the measurement points 21 and
The coordinates of 22, 23... Can be uniquely determined.
In this case, the position of the origin and the coordinate system may be arbitrarily set. Therefore, to simplify the explanation, FIG.
If one of the vertices of the platen 11, that is, the point 24 is set as the origin (0, 0), the scanning line direction (CCD
X is the X axis, and the moving direction Y of the light source 12 is Y.
A coordinate system as an axis is set. Thereby, the platen 11
Each measurement point 21, 22, 23 arbitrarily specified (set) above
, The position coordinates indicated by (x1, y1)... Are uniquely determined.

After inputting such an arbitrary measurement point, the light source 12 is moved and the reference white plate 10 is optically scanned.
The digital data output from the / D conversion unit 17 is recorded corresponding to each pixel. As a result, digital signals (reading data for shading correction) of the R, G, and B components when the reference white plate 10 is read are obtained at the measurement points designated in advance as described above.

Then, by comparing the read data for shading correction at each measurement point with the colorimetric data of the reference white plate 10 obtained in advance as described above, each of the R, G, and B components is obtained. A shading correction value at the measurement point is calculated. The colorimetric data is, for example, reference data obtained as data having the same density in the scanning direction when the reference white plate 10 is actually read by the reading element.

Therefore, if the reference read data obtained by actually reading the reference white plate 10 by the CCD element 15 ideally matches the above-described colorimetric data, it is possible to obtain read image data without shading distortion. Become. However, in practice, the platen 11, the lens 14, the CCD element 15
The output is not constant due to shading distortion due to the variation of each read pixel. Therefore, the reference white plate 10
In order to obtain ideal colorimetric data when the image data is read, a difference between the colorimetric data and the reference read data obtained by actually reading the reference white plate 10 and a shading correction value based on a ratio are obtained.

The above-mentioned shading correction value is a value for ensuring that the output of each pixel from the CCD element 15 has the same density, that is, the same value when the reference white plate 10 is read. Further, the shading correction values are set at the measurement points 21, 22 and 22 set in advance as described above.
23 and so on. That is, at each measurement point,
Data of shading correction values as shown in Table 1 below is created.

[0040]

[Table 1]

Based on the data of the shading correction values at the respective measurement points shown in Table 1 above, the input layer has the position coordinates (x,
y), the output layer is a shading correction value Rh, Gh, B shown in Table 1 for each of the R, G, B components of the pixel.
A neural network for shading correction (Rh, Gh, Bh) at each measurement point is learned as a teacher signal (hereinafter, referred to as a neural network).

In this case, in the neuro shown in FIG.
The coordinate positions (x1, y1), (x2, y2), (x3, y3) of the measurement points 21, 22, 23,.
Of the shading correction values (Rh, G
h, Bh). At this time, for example, at the measurement point 21, the intermediate layer (Hidden Laye
r) is set to an arbitrary value, and the value of the output layer at that time and the shading correction value (Rh1, Gh
, Bh1) are compared with each other, and the engagement coefficient is sequentially changed such that the shading correction value (Rh1, Gh1, Bh1) is finally output from the output layer (Output Layer) as the teacher data, and learning is performed. Will be The value finally obtained in the output layer is the shading correction value (Rh
1, Gh1, Bh1) is stored in a storage unit described later as shading correction data.

In this way, the measurement points 22, 23,.
, The same neural learning is performed, and the engagement coefficient of the intermediate layer when the shading correction values of the measurement points 22, 23,... Are finally output to the output layer is stored as shading correction data.

This neural network for shading correction uses a CC having a variation in conversion characteristics for each pixel.
Including the D element 15 and the light source 12 having uneven light amount,
This reflects the characteristics of each mechanism in the reading device when the reference white plate 10 is scanned.

The shading correction data obtained by the neural network shown in FIG. 3 according to the present invention is shown in FIG.
The shading correction unit 18 is used in the above. The details of the shading correction unit 18 will be described below with reference to FIG.

The shading correction section 18 shown in FIG.
A correction operation unit 41, a determination unit 42, a neuro operation unit (equivalent to the one configured by the neural network shown in FIG. 3) 43, and a shading correction data storage unit 4.
4 is provided.

In FIG. 1, reference numeral 1B denotes read image data which is a digital signal obtained by A / D conversion of analog data 1A output from the CCD element 15 via the driver 16 as described with reference to FIG. This is input to the shading correction unit 18. The correction operation unit 41 that performs shading correction is based on the shading correction data stored in the shading correction data storage unit 44, as described above, and calculates the shading correction value output by the neuro operation unit 43 by calculation. Applying a shading correction process to the read image data 1B;
Output as processed image data 1C.

The shading correction data storage section 44
In the table, shading correction values at each measurement point shown in Table 1 and an engagement coefficient of the intermediate layer obtained by learning based on the correction values are stored as shading correction data.
The stored contents are sent to the neuro operation unit 43, so that the neuro operation unit 43 outputs a shading correction value. As a result, the shading correction process is performed on the read image data 1B, and the output image data 1
Output processing is performed as C.

Further, as described above, the discrimination unit 42 stores the read image data 1 in the form of the position coordinates (x, y) of the read pixel at the arbitrary measurement points 21, 22, 23, etc. on the document table 11.
B, and a function of determining a document range or the like for performing shading correction in accordance with the area of each document based on the input position coordinates.

First, a procedure for storing the shading correction data based on the result of the neuro operation in advance in the above-described shading correction data storage unit 44 will be described. Therefore, the reference white plate 10 is set on the document table 11 at the initial stage. The image of the reference density on the reference white plate 10 is read through an apparatus shown in FIG. 4, that is, an optical system constituting a scanner including a document table 11, a mirror, a lens, and the like. At this time, the read data of the entire area of the reference white plate 10 set on the document table 11 is converted into digital signals 1B of R, G, and B components by the A / D converter 17 and the reference white plate 10 is
Is compared with the colorimetric data directly read by a pure reading means instead of a reading device as shown in FIG. 1 and the value at which the digital signal 1B becomes colorimetric data is obtained as a shading correction value.

The shading correction values are, for example, the measurement points 21, 22, 23,.
Are obtained as shown in Table 1 at each point. Measurement point 2
Based on the shading correction values of 1, 22, 23,..., The engagement coefficient of the intermediate layer is variously changed using the neuron shown in FIG. 3 with the coordinate values of the measurement points as the input layer and the shading correction values as the output layer. Then, the engagement coefficient at which the shading correction value at the measurement point is finally output from the output layer is obtained. The engagement coefficient obtained as a result of this determination is used as shading correction data as the shading correction data storage unit 44.
Is stored. In this case, a neuro operation is similarly performed at each measurement point, and the result is sequentially stored in the storage unit 44 as shading correction data.

Using the storage unit 44 storing the shading correction data obtained as described above, the read data 1B from which the image of the original to be read is actually read by the shading correction unit 18 shown in FIG. The shading correction calculation unit 41 performs shading correction processing based on the shading correction value output from the neuro calculation unit,
It is output as output data 1C.

The document to be read is actually placed on the document table 11 by the image reading apparatus shown in FIG. The read image data 1B thus read is input to the correction operation unit 41 shown in FIG. At this time, the coordinate data (x, y) is input to the determination unit 42 in accordance with each read image data.
The coordinate data is obtained by determining the coordinate positions of the measurement points 21, 22, 23,... On the document table 11, as shown in FIG. In the Y-axis direction, this is the position when the lamp 12 and the reflection mirror 13a constituting the scanner in FIG. Thus, in the read image data 1B, coordinate data determined by the positions of the CCD reading element 15, the lamp 12, and the like is input.

The coordinate data (position information) is transmitted to the discriminating unit 4
2 is applied to the input layer shown in FIG. Then, the shading correction data storage unit 44 corresponding to the position outputs the shading correction data as the engagement coefficient of the intermediate layer of the neuro operation unit 43. Accordingly, the neuro operation unit 43 outputs the shading correction value corresponding to the shading correction data to the correction operation unit 41, and the correction operation unit 41 performs the shading correction processing of the read image data at the specific coordinate data to be input. Will be After the correction processing, the image is output as image data 1C from the correction calculation unit 41. That is, the read image data 1B corresponding to the image is sequentially subjected to the shading correction based on the shading correction data obtained by the neuro operation unit 43, and is output as the image data 1C.

Here, in the present invention, as shown in FIG. 2, an arbitrary point on the document table 11 is designated, and shading correction values at the designated measurement points 21, 22, 23,. The learning operation based on the neuron is performed based on the value. For this learning, learning is performed using the shading correction values shown in Table 1 obtained at each of the arbitrary measurement points 21, 22, 23,... As teacher data, and while the engagement coefficient of the intermediate layer is updated, the learning is performed. Learning is performed until a shading correction value is obtained.

By performing the learning at each point in this manner, the neural characteristics are thereafter obtained. By performing the learning at each point as described above, the learned neuron is, for example, measured at the measurement point 21. And the shading correction value of each pixel point up to the measurement point 22 can be output as a continuous value. Therefore, it is possible to output a shading correction value that covers the entire area of the document table 11 other than the measurement points as shown in FIG. 2 in accordance with the input of each coordinate data.

For example, based on the coordinate positions of the measurement points 21 and 22, a shading correction value between the points is estimated. By using this, the shading correction processing at all points can be performed on the read image data 1B. Moreover, it is only necessary to specify a plurality of arbitrary points.
By learning this in advance using the above-described neuro, shading correction processing that covers the entire area of the document table 11 is enabled. In addition, before scanning the document to be read, the reference white plate 10
Therefore, it is not necessary to perform the shading correction processing to obtain the reading result of the first sheet, so that the reading time of the first sheet can be reduced.

Further, the shading correction values from the measurement point 21 to the origin 24 can be obtained in a similar manner. In addition, the measurement point 23 and the end point on the document table 11 are similarly obtained. In this case, the origin 24 and the end point 25 (x
If (n, yn) is designated as a measurement point, the shading correction value during that period can be accurately estimated. Further, if the number of measurement points is increased, a more accurate shading correction value can naturally be obtained as covering the entire area.

(Second Embodiment) In the original to be read placed on the original platen 11, the read image data is sequentially input to the correction operation section 41 by optical scanning. At this time, any point on the platen 11, that is, the measurement points 21, 22,
In the range of 23, a shading correction value based on the shading correction data learned as described above is output, and a correction process based thereon is performed.

In this case, in FIG. 2, arbitrary measurement points 21, 22, 23,... That can cover the entire area from the origin 24 to the end point 25 of the document table 11 are designated. This specified area reflects, for example, shading correction in the substantially central portion of the document table 11.

On the other hand, as shown in FIG.
An area 11a shown at a predetermined distance from the tip (upper part in the figure) of the document 1;
c, 11d, and the peripheral shading correction indicated by the area 11b from the rear end (lower part in the figure) of the document table 11 to a predetermined distance, in the areas,
A plurality of arbitrary measurement points are designated, and the shading correction values based on the comparison between the colorimetric data as described above and the reference read data obtained by actually reading the reference white plate 10 by the CCD element 15 are separately shown in Table 1. Ask.

For example, in the upper region 11a of the document table 11, the positions of the measurement points 21a, 22a, 23a,. Their coordinate data (x1a, 1a),
(X2a, y2a)... Learning is performed by neuron as described in the first embodiment, and the result is stored in the shading correction data storage unit 44 shown in FIG. 1 in the first embodiment. Is stored separately from the shading correction data obtained in. in this case,
If shading correction data is obtained for each of the four regions 11a, 11b, 11c, and 11d at the leading edge, the trailing edge, and the left and right sides of the document table 11 excluding the central portion, more accurate shading correction processing can be performed. . However, in order to avoid an increase in the stored data, the shading correction data can be obtained by using the above-described peripheral area as one area.

Therefore, the discriminating section 42 shown in FIG. 1 determines whether to perform shading correction processing corresponding to the area of the document table 11 from outside or to perform shading correction processing using one area as shown in FIG. Input settings can be made externally. Therefore, the storage area of the shading data storage unit 44 is designated so that shading correction can be performed according to the input conditions, and the shading correction data corresponding to the shading correction data is stored in the neuro operation unit 4.
3 is output.

That is, when a setting input for performing the shading correction process according to the reading area is made, the determination unit 42 uses the coordinate data (x, y) input together with the image data 1B read by the reading device. 5 and the central area other than the areas 11a, 11b, 11c, and 11d, and the storage unit 44 outputs shading correction data corresponding to the area. . Therefore, it is possible to favorably perform shading correction of the entire document table 11.

For this reason, according to this embodiment, it is possible to perform output correction favorably due to a drop in output data due to the periphery of the document table 11, a light amount in each area, a blur of the scanning optical system, and the like.

(Embodiment of the present invention for coping with aging)
Here, in the reading apparatus, even if the reference white plate 10 is read due to the change with time, initial reference read data cannot be obtained. Therefore, by using the present invention, even when the image data 1C is output, the same image data 1C as in the first reading state can be output by performing the shading correction processing on the output data from the reading apparatus even with the lapse of time. . Therefore, the shading correction data is obtained by learning using the neural network described in the first embodiment.

For this purpose, a state is created in which reading is actually performed based on the aging of the image reading apparatus. For example,
A plurality of measurement time points are set for each required time from the beginning until the life of the image reading apparatus. This is an image reading device,
In the aging process that actually operates, the state one year later,
Simulate the state two years later and the state three years later. By performing such aging processing or the like, the image reading apparatus becomes similar to a state that actually changes with time, even if it is pseudo.

Then, in each time-dependent change state, the reference white plate 10 was read, and the measurement point 2 as shown in FIG.
The shading correction value based on 1, 22, 23.
As described above in Table 1, it is determined by comparison with the colorimetric data. This shading correction value and the measurement point 2
, 22, 23,... Are trained using a neuro, and the shading correction data based on the learning result is stored in the storage unit 44 as one time-dependent change data in the shading correction data described in the first embodiment. And stored in a different area.

For example, as described above, the time (period) until the life of the image reading apparatus is divided into a plurality of times, for example, every year.
The shading correction data for each year is obtained in advance, and this is separately stored in the storage unit 44.

For this reason, as shown in FIG. 1, a timer 46 for counting the operation time of the image reading apparatus from the beginning is separately provided outside the shading correction section 18. The output from the timer 46 is, for example, 1 as described above.
It is configured to output a signal on a yearly basis.
The output from the timer 46 is input to one terminal of the neuro input layer. Thereby, every predetermined time (1
In response to the output from the timer 46 (every year has passed), the shading correction processing corresponding to the temporal change can be similarly performed.

For this purpose, a neuro operation is performed to obtain shading correction data in a state that has changed over time every year as described above. This is because, in the neuro shown in FIG. 3, an element due to aging time is added to the input layer. In FIG. 3, the operation time of the device is the same. Then, as described above, the aging process of the image reading apparatus is used to create a state that has actually changed over time, and the reference white plate 10 in that state is created.
The shading correction value based on the comparison between the read data and the colorimetric data is obtained as shown in Table 1, and the relationship between each measurement point is learned by neuro, and the result is stored in the storage unit 44 as shading correction data. The data is stored separately for each divided period, for example, for each year.

Based on the shading correction data thus obtained, the following actual reading process of the original image is performed. Therefore, in FIG. 1, the image of the original is actually read by the image reading apparatus,
The same shading correction as in the embodiment is performed.

One year after the image reading apparatus, the data is added as an input element of the neuro-operation unit 43, and the data is added one year after the apparatus in order to perform the shading correction according to the above-mentioned aging. The storage unit 44 outputs the shading correction data. Therefore, after one year has elapsed, the read image data 1B can be subjected to shading correction processing by the correction calculation unit 41 using shading correction data based on one year later, and the processed data 1C is output. As a result, a fixed output state can be corrected and output irrespective of a temporal change.

(Embodiment of the present invention for coping with voltage fluctuation)
On the other hand, the output state of the image reading apparatus greatly changes according to the voltage fluctuation. For this reason, a plurality of measurement time points are set during the required time from when the power is turned on to when the voltage inside the device is stabilized, in the same manner as the above-described temporal change. The shading correction value of each of the R, G, and B components at each measurement point shown in FIG. 2 when the aging process is performed up to each measurement time point is obtained. As a result, a similar shading correction value as shown in Table 1 is obtained at each measurement time point. This means that the reference white plate 10 is actually read at each measurement time point,
The relationship between the read data and the colorimetric data is as described above.

Therefore, the neural network in which the time since power-on is added to the input layer shown in FIG. 3 is made to learn the shading correction values of the measurement points 21, 22, 23,... Keep it. This is subjected to learning calculation as a shading correction neuro that takes into account the time since power-on, and the result is stored in the above-described shading correction data storage unit 44 as shading correction data.

Therefore, when the image of the original to be read is actually read, the shading correction unit 18
A timer (first counting means) 45 for counting the time from power-on is prepared. The output from the timer 45 is added as one of the input elements to the input layer in the neural network as shown in FIG.

Therefore, even when an image of a document is actually read in a state where the power supply voltage is unstable when the power is turned on, shading correction based on an appropriate shading correction value in the unstable state at that time can be performed. That is, if the image reading operation of the document is started immediately after the power is turned on, the shading correction data corresponding to the time since the power is turned on is output from the storage unit 44 to the neuro calculation unit 43. Therefore, the neuro calculation unit 43 outputs to the correction calculation unit 41 a shading correction value corresponding to the time at which the reading operation starts from the time when the power is turned on. As a result, the data 1C output from the correction operation unit 41 is subjected to shading correction in consideration of an unstable state, and becomes accurate read image data 1C.

That is, during the image reading operation after the power is turned on,
Even if the light amount changes or the light amount distribution of the light source 12 fluctuates due to the voltage fluctuation, stable shading correction processing can always be performed at each pixel point on the document table 11.

(Embodiment of the present invention for coping with environmental changes)
In addition to the above-described voltage fluctuations immediately after the power is turned on, it is conceivable that a change in the amount of light of the light source 12 and a change in output characteristics from the CCD element may also occur in an environmental change or the like that occurs in a place where the image reading apparatus is installed. Can be

For this purpose, for example, a humidity sensor (hygrometer) 47 for detecting an environmental change is provided inside the image reading apparatus to detect a change in humidity, and based on shading correction data obtained as a result of performing a neuro operation in advance in accordance with the humidity sensor. Thus, shading correction processing can be performed.

Therefore, the reference white plate 10 is placed on the document table 11 with the image reading apparatus actually in each humidity state, for example, in the state of 60%, 70%, and 80% humidity, and the optical system shown in FIG. CCD device 1 in the same manner as in each of the above-described embodiments.
5, and based on the read reference data and the colorimetric data, a shading correction value is obtained in a humidity change state as shown in Table 1. This is referred to as each measurement point 21 shown in FIG.
, And the relationship between the shading correction value and
As shown in FIG. 3, a learning operation is performed using a neural network in which a humidity signal is previously added to one of the input layers. The shading correction data (the engagement coefficient of the intermediate layer) based on the result of this calculation is stored in the shading correction data storage unit 44 in advance.

As described above, the shading correction processing by the shading correction unit 18 shown in FIG. 1 is performed by using the shading correction data storage unit 44 storing the calculation results by various neural networks according to the humidity change. Performed on the manuscript. This will be described with reference to FIG. 1. A humidity sensor 47 for measuring the humidity inside the image reading apparatus is prepared outside the shading correction unit 18. The output from the humidity sensor 47 is input to one input terminal of the input layer shown in FIG. Thus, when the humidity changes, the storage area of the shading correction data storage unit 44 is read from the hygrometer 47 so that the shading correction data corresponding to the humidity is output from the storage unit 44 to the neuro calculation unit 43. Select according to the signal of

Thus, even when dew condensation or the like occurs on the document table 11 due to a change in the environment inside the reading apparatus, particularly a change in humidity, stable shading correction can be always performed at each reading point of the document image on the document table 11. It becomes possible. That is, individual shading correction according to the humidity state can be performed. Therefore, the data 1C subjected to appropriate shading correction can be output without being affected by environmental changes.

(Other Embodiments of the Present Invention) As described above,
In the image reading device, shading correction data is obtained by performing learning using a neuro based on an arbitrary designated point on the document table 11 before actually reading an image. Then, based on the shading correction data that is the learning result, shading correction processing is performed on the data obtained by actually reading the image, and the image data can be output as fixed image data.

Therefore, in the present invention, not only the single shading correction data but also the shading correction data due to the aging of the image reading apparatus, the humidity change, and the voltage fluctuation until the power supply is stabilized until the power is turned on. , Respectively, and these can be used as needed.

An example will be described below. It is possible to arbitrarily input the setting of the shading correction processing based on them. That is, it is not necessary to perform shading correction according to the humidity change in a place where the humidity change is hardly considered. Therefore, such processing can be omitted.

Therefore, in order to select and use them, the discriminating unit 42 in FIG. 1 determines whether or not to perform shading correction processing in response to a change in the power supply voltage of a light source or the like, a change over time in a reading device, or the like. To be able to input the condition setting signal and the like. On the basis of this input condition setting signal, the neuro-calculating section 43 performs x, as shown in FIG.
Signals from timers 45 and 46 as shown in FIG. 1 corresponding to respective condition setting signals and a humidity sensor 47 for detecting an environmental change are selectively inputted to input terminals different from the position information of y. Keep it.

As a result, the neuro operation unit 43 performs the neuro operation for shading correction on the condition that the input from each of the timers 45 and 46 or the humidity sensor 47 is a condition.

For this reason, the user can arbitrarily set the respective condition settings, and when unnecessary, do not set the condition. The setting input state is discriminated by the discriminating section 42 and transferred to the shading correction data storage section 44 via the discriminating section 43 as a condition setting signal. In response to this, the storage unit 44 outputs the shading correction data previously obtained and stored according to the condition to the neuro calculation unit 43. And the neuro operation unit 43
Calculates a shading correction value corresponding to the shading correction value
Output to 1. Further, an input request signal is output to the timer 45, the timer 46, and the humidity sensor 47 in accordance with the above-described condition setting. In response, the timer 45, the timer 46, the hygrometer 47
Outputs an elapsed time or humidity signal to the neuro operation unit 43. From these signals and the operation data, the neuro operation unit 43 calculates the shading correction values of the R, G, and B components and outputs the result to the correction operation unit 41.

The flow of control by shading correction according to the present invention and the flow of processing by shading correction when the above-mentioned input conditions are arbitrarily selected and set will be described below with reference to flowcharts shown in FIGS. 6 and 7 indicate processing in each circuit of the block diagram of FIG. 1, respectively.
For the sake of simplicity, the same reference numerals are used.

First, when the image reading of the document placed on the document table 11 starts, the read image signal (data) 1
B (Rd, Gd, Bd in the case of color) and (x, y) of the coordinate signal (position information data) at that time are input to the respective circuits shown in the block diagram shown in FIG. 1 (step 6A). Is done. Therefore, as for each input signal, the read image data 1B is converted into the shading correction operation unit 41,
The position coordinate signal (x, y) at that time is sent to the determination unit 42 (step 6B). Therefore, the read image data 1B is sent to the shading correction calculation unit 41 shown in FIG. 1 as it is, and the shading correction processing in step 6M is performed. This process is performed based on a shading correction value output from a neuro operation unit 43 described later.

Then, the position coordinate signal (x, y) sent to the discriminating section 42 is first sent to the discriminating section 42 in step 6C.
It is determined whether the pixel is within a coordinate range that requires shading correction by a neuro set in advance from outside. Here, for the pixels excluded from the target, a predetermined value for shading correction is output to the correction calculation unit 41 (Step 6M). This is a case where the shading correction processing by the neuro according to the present invention is not required. In other words, conventionally well-known shading correction is executed without targeting the shading correction according to the present invention. For this reason, the user sets in advance such that the shading correction processing according to the present invention is performed for each of the areas, for example, only the peripheral areas 11a, 11b, 11c, and 11d on the document table 11 as described with reference to FIG. As a result, if the coordinate position is at the center, it is determined that the shading correction processing according to the present invention is not to be performed, and the processing in step 6M is executed.

In step 6M, for example, a shading correction value for shading correction that has been conventionally performed, instead of a shading correction using a neuron, is output to the correction calculation unit 41 as it is. . Therefore, the shading correction value is stored in advance in the storage unit 44a or the like, and the shading correction value corresponding to the position coordinates is output. For example, the shading correction value is stored in association with each pixel in one line, and the shading correction value is output based on the value of x regardless of the y value of the coordinate position (x, y). In this case, the storage capacity is only required for one line. This is a simple method.
Although it is possible to correspond to the entire area, the storage capacity of the shading correction value becomes enormous.

On the other hand, when the position coordinate data (x, y) is
If it is determined that the image is within the shading target due to neuro, in steps 6D to 6F, shading correction using neuro is performed in consideration of factors such as the time from power-on, the elapsed time of the reading operation of the image reading device, and the humidity inside the device. Is determined. That is, this is the object of the shading correction processing by the neuro according to the present invention.

Therefore, after the processing in the discrimination section 42 is completed, the subsequent processing is sent to the neuro operation section 43. Therefore, in accordance with the results determined in Steps 6D to 6F, in Steps 6G to 6J, the calculation data necessary for the neuro calculation and the input values are loaded. The operation data is
The various types of data stored in the shading correction data storage unit 44 as described above, and the shading correction data is loaded into the neuro operation unit 43 from this. If the condition has not been set from the outside, the calculation process of the neuro calculation unit 43 based on the position coordinate data is executed as it is (step 6J). This is based only on the shading correction data obtained based on the measurement points 21, 22, 23,... Shown in FIG.

In step 6J, as described in the second embodiment, the document table 11 can be set in an arbitrary area in advance, and the shading correction processing can be performed for each area. That is, it is also possible to determine whether the coordinate position data is the central part or the peripheral part, and to load (output) the shading correction data obtained in advance for each area to the neuro operation part 43. This may be set in advance to perform shading correction processing for each area.

In consideration of shading correction processing related to voltage fluctuations after power-on, the shading correction data described in the above-mentioned (embodiment for coping with voltage fluctuations) is stored in the storage unit 44 by the neuro calculation unit. 43 (step 6D → 6G). Also, when dealing with the aging of the image reading device,
The corresponding shading correction data is loaded into the neuro operation unit 43 (steps 6E → 6H). Then, when coping with an environmental change, the shading correction data is loaded into the neuro operation unit 43 (Step 6F) because the conditions have been input to the determination unit 42 in advance.
→ 6I). Otherwise, it is as described above.

Next, the neuro operation unit 4 loaded with the shading correction data according to any of the above conditions.
In 3, when the shading correction data is input, the shading correction value corresponding to the shading correction data is
Output to 1. That is, in step 6L, the neuro-operation unit 43 uses the shading correction data storage unit 4
The shading correction data (Rh, Gh, Bh) is output from the output layer to the correction calculation unit 41 by performing calculation based on the shading correction data that is the engagement coefficient sent from the control unit 4.

The shading correction value is calculated by the correction operation unit 4
1, the shading correction processing (step 6M) is performed on the read image data 1B already input to the shading correction calculation unit 41 from step 6B described above, and the processed image data 1C is It is output to the outside (step 6N). In this case, if the shading correction data obtained by the neuro calculation unit 43 is not obtained, the shading correction value predetermined in step 6K is sent to the shading correction calculation unit 41 as described above. As described above, the shading correction operation unit 41 performs the shading correction process, and the read image data is subjected to the image data 1C after the correction process.
Is output as

Accordingly, shading correction processing is performed on the read image data in the entire area of the document table 11 using the shading correction value output from the neuro operation unit 43. In this case, shading correction data at positions at measurement points 21, 22, 23 and the like arbitrarily specified as shown in FIG. 2 specified in advance is obtained by performing a neuro operation, and an output process is performed based on this. For,
The output processing of the image data 1C after the actual image reading is not delayed.

Further, as described in the second embodiment, a plurality of areas can be set on the document table 11 as shown in FIG. 5, and shading correction processing can be performed for each area. That is, when the temporal change in the peripheral area of the document table 11 is large, the shading correction in that area is performed according to the temporal change, and the first portion is provided in the central portion.
The shading correction according to the position information according to the embodiment can be performed only. Therefore, since the shading correction data can be partially obtained by the neuro operation unit 43, the shading correction processing at a point where the change with time is large can be effectively performed, and the stable shading correction processing can be performed.

[0102]

According to the image reading apparatus of the present invention, when the position coordinates of a certain read pixel on the document table are input, a shading correction value for the read image data based on the position is obtained, and the shading correction processing based on the shading correction value is performed. Thus, stable shading correction processing covering the entire area of the document table can be performed.

Therefore, a neural network is used to obtain a shading correction value, and it is possible to perform stable shading correction on the entire read document in a shorter read time than in the related art. In particular, in order to output a shading correction value using a neural network, a plurality of arbitrary points are specified in advance, and the shading correction value at the specified point is learned and calculated, so that the distance between these points is determined. Since the shading correction value can be output so as to substantially estimate the shading correction value, and the shading correction process can be performed using the shading correction value, the reading time can be reduced as compared with the case where the reference white board or the like is read in advance each time and the shading correction is performed. .

Further, by using a neural network learned by adding the elapsed time from the power-on to the input for shading correction, the light amount or light amount distribution of the light source changes due to voltage fluctuation inside the device after the power-on. However, stable shading correction can always be performed at each pixel point on the document table.

Further, even in the case of characteristic changes due to aging in the image reading apparatus, shading correction processing can be performed using a neural network, and is not affected by characteristic changes due to photoelectric conversion elements (reading elements) or optical scanning. And stable shading correction can be performed. This also enables shading correction processing to be performed in response to environmental changes.

If it is possible to arbitrarily select whether or not to use each of the above-described neural networks to perform shading correction, these shading corrections can be performed particularly when high-precision shading correction is not required. Since the shading correction processing can be performed without considering the condition setting, the processing time can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating details of an example of a shading correction unit included in an image reading apparatus according to the present invention.

FIG. 2 is a plan view showing an example of measurement points at which arbitrary position coordinates can be set on a document table of the image reading apparatus according to the present invention.

FIG. 3 is an explanatory diagram illustrating an example of a neural network used in the shading correction unit according to FIG. 1;

FIG. 4 is a schematic block diagram showing a configuration of a shading correction unit of the image reading apparatus according to the present invention, including a circuit unit;

FIG. 5 is a plan view showing an example in which an arbitrary measurement point for shading correction according to an area on a document table in FIG. 2 is designated;

FIG. 6 is a control flowchart illustrating an example of a processing procedure by shading correction in FIG. 1;

FIG. 7 is a control flowchart showing a continuation of the processing procedure of FIG. 6;

[Explanation of symbols]

REFERENCE SIGNS LIST 10 shading correction reference white plate 11 original table 12 light source for optical scanning 13 a to 13 c reflection mirror 14 imaging lens 15 line CCD (photoelectric conversion element) 16 CCD driver 17 A / D converter 18 shading correction unit 1 A photoelectrically converted Analog image signal 1B Digital image signal (read image data) converted from A / D converter 1C Digital image signal (image data) subjected to shading correction processing 21-23 Measurement points 41 for collecting data for shading correction 41 Correction calculation unit 42 Discrimination unit 43 Neural network calculation unit 44 Shading correction data storage unit 45 Timer (first counting means) for elapsed time from power-on 46 Timer for reading operation time of image reading apparatus (second timer) Counting means) 47 Hygrometer Detection sensor of environmental changes)

Claims (7)

[Claims] An image of a document placed on a document table is optically scanned, and reflected light of the scanned image is converted into an electric signal and formed on photoelectric conversion means for outputting as read image data. Then, correction is performed on the read image data output from the photoelectric conversion means corresponding to the image based on the reflected light based on a shading correction value such that the output state of the same density of the image becomes constant in advance. In the image reading apparatus provided with means, a plurality of arbitrary points on the platen are designated for shading correction, and the correction means preliminarily performs the shading correction for the reference data having the same reference density. The shading correction read data read by the photoelectric conversion means at the specified reference densities at the plurality of specified points is the shade data serving as the reference data. And calculating a shading correction value between the positions of the plurality of designated points, performing shading correction processing of the read image data with the shading correction value, and outputting the image data as image data. An image reading apparatus comprising: 2. The method according to claim 1, wherein the correction means uses a neural network to input position information of an arbitrarily designated point on the platen as an input element, and according to the position information, the reference data and the read data for shading correction. Learning is performed using the shading correction value obtained by the above as a teacher signal, an engagement coefficient whose learning result is finally a shading correction value is obtained as shading correction data, and a storage unit for storing the shading correction data; A neuro-operation unit that adds each position information of the data to the input layer, outputs a shading correction value corresponding to the shading correction data stored in the storage unit from the output layer, and a shading correction value output from the neuro operation unit. A correction calculation unit that performs shading correction processing of the read image data according to The image reading apparatus according to claim 1, characterized in that it is configured. 3. The storage unit sets a document table in an arbitrary area in advance, and stores shading correction data learned and calculated by the neural network in accordance with each area, and stores the data separately. Comprises a discriminating section for discriminating a preset area on the platen based on read image data and position information corresponding to the data, wherein the discriminating section includes a preset area on the platen corresponding to the position information. The correction unit outputs shading correction data from the storage unit to the neuro calculation unit in accordance with the determination result of the determination unit, so that the shading correction process corresponding to each of the predetermined regions can be performed. 3. The image reading device according to claim 2, wherein: 4. In addition to the positional information of a plurality of arbitrarily designated points on a platen, each designated point is input using a neural network for shading correction using an elapsed time from power-on of the image reading apparatus as an input element. The learning based on the shading correction value is performed, and the shading correction data as a result of the learning is stored in the storage unit, so that the shading correction from power-on to stabilization of the power supply voltage is enabled. 2. The image reading device according to 2. 5. In addition to the positional information of a plurality of arbitrarily designated points on the platen, each designated point is determined using a neural network for shading correction using an elapsed time in the operating state of the image reading device as an input element. 3. The image reading apparatus according to claim 2, wherein learning based on the shading correction value is performed, and shading correction data obtained as a result of the learning is stored in a storage unit, thereby enabling shading correction processing capable of coping with aging. apparatus. 6. In addition to position information of a plurality of points arbitrarily designated on a platen, each designation is made using a shading correction neural network using a detection signal of a change in environment around the image forming apparatus as an input element. The image according to claim 2, wherein learning based on a shading correction value at a point is performed, and shading correction data as a result of the learning is stored in a storage unit, thereby enabling shading correction in response to environmental changes. Reader. 7. A first counting means for counting the time from when the power of the image reading device is turned on, in addition to the position information of a plurality of arbitrarily designated points on the document table, and counting the reading operation time of the image reading device. A second counting means for converting each signal from a detection sensor for detecting a change in environment around the image reading device into various shading correction values at designated points using a neural network for shading correction as an input element. Learning based on the learning result is stored in the storage unit, and whether any one of the signals from the first and second counting means and the detection sensor is added to the input element is determined. 3. An image reading apparatus according to claim 2, wherein said image reading apparatus is arbitrarily selectable from outside so that any one of the shading corrections can be arbitrarily performed. .