JPH08255244A - Image processor - Google Patents

Abstract

PURPOSE: To make it possible to perform the image processing of only image data to be a correction object for the image data on a screen and to obtain the image data in which an object is made attractive in appearance in an image by correcting the image data to be the correction object which is preliminarily set from the inputted image data on the screen. CONSTITUTION: This processor is provided with an image recognition means 10 and an image correction means 11. The image recognition means 10 recognizes the image data of the object which is preliminarily set as a correction object from the image data on a screen that a data input means 2 inputs from an original. The image correction means 11 corrects only the image data of the recognized object for the image data on the both surfaces by a preliminarily set image processing. Therefore, the image data of the object can be recognized from the image data by the image recognition means 10 and the image data on the screen can be corrected by the image correction means 11. In this case, because only the image data of a specified object of the image data on the screen which is copied from the original is corrected, only the image of a table can be made conspicuous when the indoor image is copied, for instance.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus for inputting image data, correcting it, and outputting it.

[0002]

2. Description of the Related Art At present, there is a digital copying machine as an image processing apparatus for inputting and outputting image data. The digital copying machine can be recognized as an image processing apparatus in which an image scanner and a laser printer are connected, and the image data input from the screen of the original by the scanning scanning of the image scanner can be output by the printing of the laser printer. .

[0003]

The image processing apparatus such as the digital copying machine described above can input and output image data.

Conventionally, a copying machine was required only to faithfully reproduce image data of an original document, but nowadays, it is desired to perform various image processing on the image data of the original document, and solarization is required. There is also a product that realizes image processing such as mosaic and mosaic. However, such image processing is uniformly executed on the entire screen of image data, and it is not possible to emphasize only a specific object located on the screen. For example, when copying indoor images,
In some cases, it is desired to make only the image on the table stand out, but it is difficult to execute such image processing by a digital copying machine.

This is because the DT in which an image scanner or a laser printer is connected to a personal computer.
It can be executed by a P (Desk Top Publishing) system or the like. However, this requires the user to operate the personal computer, and the work is extremely complicated.

[0006]

According to the first aspect of the present invention,
Data input means for inputting image data of the screen is provided, image recognition means for recognizing preset image data of the correction target from the input image data of the screen is provided, and the recognized image data of the correction target is corrected. The image correction means is provided, and the data output means for outputting the image data of the screen in which the image data to be corrected is corrected is provided.

According to a second aspect of the invention, in the first aspect of the invention, the image recognition means recognizes the image data to be corrected by using a neural network.

According to a third aspect of the present invention, in the first aspect of the invention, the image recognition means corrects the first neural network for identifying a preset shape from the image data to be corrected and the corrected shape according to the identified shape. A second neural network for discriminating target image data.

According to a fourth aspect of the invention, in the third aspect of the invention, there is provided color information detecting means for detecting the color ratio of the image data to be corrected and inputting it to the second neural network.

According to a fifth aspect of the invention, in the invention of the fourth aspect, the color information detecting means outputs the color ratio as a ratio of the color components of the three primary colors existing in the image data to be corrected.

According to a sixth aspect of the present invention, in the fourth aspect of the present invention, the color information detecting means also detects the number of colors of the image data to be corrected and inputs it to the second neural network together with the color ratio.

According to a seventh aspect of the invention, in the sixth aspect of the invention, the color information detecting means detects the number of colors as a ratio of the sum of the color components of the three primary colors existing in the image data to be corrected.

According to the invention of claim 8, in the invention of claim 3, the first neural network discriminates the shape as a ratio of the probabilities of a plurality of preset candidates.

[0014]

According to the invention described in claim 1, when the image data of the screen is inputted by the data input means, the image data of the correction target which has been preset from the image data of the screen is recognized by the image recognition means, and this recognition is performed. The corrected image data to be corrected is corrected by the image correction means. Since the image data of the screen in which the image data of the correction target is corrected is output by the data output unit, the image data of the screen in which the image data of the correction target is corrected is output from the input state.

According to the second aspect of the present invention, since the image recognition means recognizes the image data to be corrected by using the neural network, it is possible to automatically recognize the image data to be corrected.

According to the third aspect of the present invention, when the preset shape is identified from the image data to be corrected by the first neural network, the image data to be corrected is detected by the second neural network according to the identified shape. Since it is determined, the image data to be corrected is well recognized according to the shape.

According to the fourth aspect of the invention, the color information detecting means detects the color ratio of the image data to be corrected and inputs it to the second neural network, so that the second neural network outputs the image data to be corrected. Good discrimination by shape and color ratio.

According to the invention of claim 5, the color information detecting means outputs the color ratio as the ratio of the color components of the three primary colors existing in the image data to be corrected, so that the color ratio can be easily detected.

According to the sixth aspect of the present invention, the color information detecting means inputs the number of colors and the color ratio of the image data to be corrected into the second neural network. Can be satisfactorily identified by the shape, the color ratio, and the number of colors.

In the invention according to claim 7, since the color information detecting means detects the number of colors as a ratio of the sum of the color components of the three primary colors existing in the image data to be corrected, the number of colors can be easily detected. It

In the eighth aspect of the present invention, the shape is discriminated by the first neural network as the ratio of the probabilities of a plurality of preset candidates, so that the shape of the object is easily identified.

[0022]

An embodiment of the present invention will be described below with reference to the drawings. First, the digital copying machine 1 which is the image processing apparatus of the present embodiment reads and scans a screen of an original document, inputs full-color image data, and copies the image data to a print sheet for output. When the image data includes preset image data to be corrected, it is possible to print out the image data of the screen in which only the image data to be corrected is emphasized. In this case, the image data to be corrected is a specific image requested by the user for image processing, and is generally set as a specific object.

Therefore, the digital copying machine 1 of this embodiment is
As shown in FIG. 2, the data input means 2, the arithmetic processing means 3, the data storage means 4, the program storage means 5, the controlled quantity detection means 6, the state quantity detection means 7, the data output means 8, and the operation input means. 9. Here, the functions of these various means will be sequentially described below as a configuration.

First, the data input means 2 is composed of a color-compatible image scanner or the like, and the image data of the screen of the original is RG in three primary colors by reading and scanning the original.
Input as a 256-color image by the color components of B (Red, Green, Blue). The arithmetic processing means 3 is a CPU (Centr).
al Processing Unit) and performs various data processing such as image data correction. The data storage means 4 comprises a RAM (Random Access Memory) or the like,
Various data such as the processing data of the arithmetic processing means 3 is temporarily stored in an updatable manner. The program storage means 5 is
It is composed of a ROM (Read Only Memory) or the like, and fixedly stores various data such as a control program of the arithmetic processing means 3 in advance.

The controlled amount detecting means 6 and the state amount detecting means 7 are composed of a circuit sensor, a potential sensor, a P sensor, a print density sensor, etc., and are controlled by the data output means 8 to be controlled. The quantity and the state quantity involved in the quantity are respectively detected. The data output means 8 is composed of a full color compatible laser printer or the like and outputs image data to YMC.
The color toner of K (Yellow, Magenta, Cyanide, Black) is used to print out as a color image of 256 colors on the printing paper.
The operation input means 9 is composed of a touch panel or the like, can input and operate various data such as change of operation mode and start of copy operation, and can display and output various data such as operation state and error content.

The digital copying machine 1 of this embodiment can output the image data input by the data input means 2 from the data output means 8. In this process, the program stored in the program storage means 5 is stored. Accordingly, the image processing can be performed on the image data by the arithmetic processing means 3.

Therefore, the digital copying machine 1 of the present embodiment
Has an image recognition means 10 and an image correction means 11, as shown in FIG. As will be described later in detail, the image recognizing means 10 is an object such as “can, table, chair, building” which is preset as a correction target based on the image data of the screen input from the document by the data input means 2. Recognize the image data of. The image correction means 11 corrects only the image data of the recognized object with respect to the image data of the screen by preset image processing.

The image recognition means 10 includes a character / pattern classification means, an object area extraction means, an object area division means, a first neural network 12, a candidate selection means 13, a color information detection means 14, and a second neural network 15. Have

The character / pattern classification means separates the image data input from the document by the data input means 2 into a character portion and a pattern portion, and outputs the pattern portion to the object area extraction means as image data of the screen. The object area extracting means extracts the image data of a preset object from the image data on the screen as an object area, converts the shading of the object area into a size suitable for identifying the first neural network 12, and It is output to the object area dividing means. The object region dividing means converts the image data of the object region into “4 ×
The first neural network 1 is divided into 4 "areas and the ratio of shading of these areas is used as area data.
2 and the color information detecting means 14.

The first neural network 12 is
The input layer 16, the intermediate layer 17, and the output layer 18 are formed by a hierarchical structure, and each of these layers 16 to 18 is composed of a plurality of neurons 20 connected by synapses 19. The input layer 16 is composed of 16 neurons 20 corresponding to area data, and the output layer 18 is a "circle," which is preset as a shape to be identified from image data of an object.
It is composed of four neurons 20 corresponding to "ellipse, triangle, square". Therefore, the first neural network 1
2 identifies the preset shape from the image data of the object by outputting the probabilities of “circle, ellipse, triangle, and quadrangle” in response to the input of the area data.

As shown in FIG. 3, the neuron 20 is a multi-input / single-output non-linear element, and the coupling coefficient ω is variably set for each of the plurality of synapses 19 on the input side. When the signal yi is input to the i-th synapse 19, since the coupling coefficient ωi is set in the synapse 19, one signal x is output to each of the plurality of synapses 19 on the output side correspondingly. To be done. In this embodiment, since the discrete time model is adopted as the neuron 20, the output value x of the neuron 20 at time (t + 1)
(t + 1) is as follows.

[0032]

[Equation 1]

F: output function θ: threshold value N: number of layers ε: non-negative parameter that determines the nonlinearity of the output function When “ε −> + 0”, the output function f becomes a unit step function and the neuron 20 The output value of each is a binary value of "1, 0" and is in an excited (firing) state and stationary (non-firing)
Express the state and.

In the digital copying machine 1 of this embodiment,
The neural networks 12 and 15 are trained by back propagation. In this case, learning and organization are realized by changing the coupling coefficient ωi of the synapse 19 and the threshold value θ of the neuron 20. The coupling coefficient ωi of the synapse 19 is as follows.

[0035]

[Equation 2]

K: parameter (0 <k <1) c: parameter (0 <c) δi: learning signal back-propagating from the output layer i di: teacher signal f ′: differentiation of output function That is, the input signal and the ideal signal Output signal of the neural networks 12 and 15 with respect to this input signal is compared with the teacher signal, and this error is used as a learning signal in the output layers 18 and 23.
To the input layers 16 and 21 through back propagation. As a result, the coupling coefficient ω of the synapse 19 is changed so that the output signal converges to the teacher signal with respect to the input signal.
The neural networks 12 and 15 are trained to generate an ideal output signal.

The color information detecting means 14 detects the color ratio and the number of colors from the area data and outputs them to the second neural network 15. The color ratio is set as the ratio of the three main color components of the image data of the object, and the number of colors is the number of color components forming the image data of the object, but here the total number of color components that can be expressed. (256
Since it is set as a ratio with the number of colors), it will be expressed as a color number ratio below.

The candidate selecting means 13 is, as described above,
The “circle, which the first neural network 12 outputs,
From the probabilities of “ellipse, triangle, and square”, the top three are selected, and the shapes identified as these from the image data of the object are input to the first to third input parts of the second neural network 15 in order of probability. Output.

This second neural network 15 also
Like the first neural network 12, the input layer 21, the intermediate layer 22, and the output layer 23 are formed in a hierarchical structure, and each of these layers 21 to 23 is a synapse 19.
It consists of a plurality of neurons 20 connected by. The input layer 21 has seven neurons 20 corresponding to the output signals of the candidate selecting means 13 and the color information detecting means 14.
The output layer 23 corresponds to "cans, tables, chairs, buildings" which are the discrimination results of the image data of the object.
It consists of 20 neurons.

Therefore, the second neural network 15 has the shape identified by the first neural network 12 and selected by the candidate selecting means 13 and the image data of the object detected by the color information detecting means 14. The image data of the object is discriminated by a neural calculation according to the color ratio and the color number ratio. Therefore, the neural networks 12, 15 and the color information detecting means 1
The image recognition means 10 composed of 4 or the like recognizes preset image data of the object from the image data of the input screen.

The image correction means 11 performs image processing on the image data of the object recognized as one of "can, table, chair, building" according to the characteristic data preset for each of these objects. The data output means 8 prints out the image data of the screen in which the image data of the object is corrected. For example, if the image data of the object is a table,
The image correction means 11 emphasizes the sharpness of the straight line portion,
The data output means 8 adjusts the exposure amount of the edge of the straight line portion. Further, the image correction means 11 adds a shadow to the root portion of the leg of the table, and the data output means 8
Increases the density of the black toner by emphasizing the intensity of the exposure amount in the shaded area.

The image processing of the image data of the object by the image correction means 11 can be set in various forms as desired, but here, as a general theory, it is set in consideration so that the object looks good. In addition, the digital copying machine 1 of the present embodiment includes a stage designating means, a correction adjusting means, as various means related to the image processing of the image data of the object.
It has an object designation means, a correction control means, and the like.

The stage designating means can designate the degree of image processing of the image data of the object stepwise such as "strong, medium, weak" by the input operation of the operation inputting means 9. The correction adjusting unit adjusts the degree of image processing of the image data of the object at a designated stage. The object designating unit can designate that the object of the image data to be image-processed is set as “table only” or “other than chair” by the input operation of the operation inputting unit 9, and the correction control unit designates The image data of the object to be image-processed is limited to the content.

With such a configuration, the digital copying machine 1 of the present embodiment, as shown in FIG. 6, operates in the normal copying mode as the operation mode by the data input means 2 as in the conventional case. The document is read and scanned to input image data, and the image data is printed out by a data output means 8 on a printing paper.

Then, in the state where the object correction mode is set as the operation mode, in the process of copying the image data of the document on the printing paper as described above, as shown in FIGS. The image data of the object can be recognized from the image data of the screen, and the image data of this object can be corrected by the image correction means 11 with respect to the image data of the screen. In this case, of the image data of the screen copied from the original, only the image data of a specific object is corrected, so, for example, when copying an image in a room,
It is possible to make only the image of the table stand out.

Therefore, the processing operation of the digital copying machine 1 in such an object correction mode will be sequentially described in detail below with reference to FIGS. 6 to 8. First, in the digital copying machine 1 of the present embodiment, "cans, tables, chairs, buildings" are preset as objects to be recognized and corrected.

Then, as shown in FIG. 6, in the initial setting, the operation input means 9 is input and operated, the object correction mode is set as the operation mode, the degree of the image data of the object is designated if desired, and the object to be image-processed. Is specified. And
When the data input means 2 scans and scans a document in which the image data of the object is written on the screen, the image data of the screen including the image data of the object is input from the document by this scanning. At this time, it is determined whether or not the object correction mode is set, and when the mode setting is determined, the object correction is executed as a processing operation.

Then, as shown in FIG. 7, the character / pattern classification means separates the image data on the screen into a character portion and a pattern portion, and outputs this pattern portion to the object area extraction means as image data on the screen. At this time, the character / pattern classification means examines the grayscale difference between adjacent pixels of the image data, and if the high grayscale difference is continuous, the character part is obtained. Sort into. Since the image data of the character portion does not exist as the image data of the object, the description is omitted because it is irrelevant to the present invention.

Next, as shown in FIG. 8, the object area extracting means detects a blank or a background from the image data of the screen consisting of the picture part, and as shown in FIG. It is sequentially excluded from the outer peripheral portion of the data. Then, since the minimum area in which the image data of the object exists is extracted as the object area, the shading of this object area is converted into a preset size. This size is a size used by the first neural network 12 for learning to identify an object. For example, the total length of an arbitrary line in the object region is 3 cm, and the density (assumed to be image data of the object) is 1 cm. If the line of the time area is 6 cm, then the 1 cm shading of the object area is converted to 2 cm.

Next, the object area dividing means is shown in FIG.
As shown in FIG. 3, the image data of the object area is divided into “4 × 4” areas, and the ratio of shading of these areas is output to the first neural network 12 and the color information detecting means 14 as area data. At this time, if the shade ratio of a certain area is 30 (%), this is output as 0.3.

Therefore, as shown in FIG. 7, the first neural network 12 corresponds to the image data of the object 16
When individual area data are input, the probability of "circle, ellipse, triangle, square" is calculated by neural calculation. The probability is 70% for a circle, 20% for an ellipse, 8% for a triangle, 2% for a square, and 100% in total. Is output.

Then, the candidate selecting means 13 selects the upper three from the four probabilities output from the first neural network 12, and inputs them to the second neural network 15 as a shape identified from the image data of the object. Layer 2
The first to third neurons 20 of 1 are output in the order of probability. That is, the four shapes identified by the first neural network 12 are circle → 0.1, ellipse → 0.3, triangle →
Since the identifier is set in advance such as 0.7, square → 1.0, if the probability of the circle is maximum, the “main shape 1” is assigned to the first neuron 20 of the input layer 21 of the second neural network 15. Is output as 0.1.

On the other hand, the color information detecting means 14 detects the color number ratio and the color ratio from the 16 area data, and outputs these to the fourth to seventh neurons 20 of the input layer 21 of the second neural network 15. Output. The color number ratio is the number of color components forming the image data of the object and the number of all color components.
Since it is detected as a ratio with (256), the number of color components is added up for each pixel forming the area data, and this number is divided by the total number of 256 and output. For example, if the image data of the object area is represented by 128 colors, the number of color components forming the area data is 128, and the color number ratio is "128/256 = 0.5".

Since the color ratio is detected as a ratio of three representative colors of the image data of the object, the number of each of the 256 color components is integrated in the pixel unit forming the area data, and the higher order It is output as an identifier of the three color components.
Therefore, 256 color components have (red: 0.1 purple: 0.2 yellow:
The identifiers are individually set like 0.3 green: 0.4 ...). For example, if the most color component of area data is red,
As one of the color ratios, 0.1 is output as "main color 1" to the fifth neuron 20 of the input layer 21 of the second neural network 15.

As described above, when the identification result of the image data of the object, the color number ratio and the color ratio are input to the input layer 21 of the second neural network 15, the object is output to the output layer 23 by the neural calculation. The determination result of the image data is output. Since "can, table, chair, building" which is the discrimination result of the image data of the object is preset in each of the four neurons 20 of the output layer 23, each of the four neurons 20 is set in advance. The image data of the object is determined to have the highest output probability. If the maximum probability output by the second neural network 15 does not reach 60%, it is determined that the image data of the object is neither "can, table, chair, building", and the following processing is stopped. .

Then, as described above, the image recognition means 10 including the neural networks 12 and 15 and the color information detection means 14 converts the image data of the input image data into "can, table, chair, building". When the image data of the object is recognized, the image correction unit 11 performs image processing on the image data of the recognized object according to the characteristic data set in advance, so that the image data of the screen printed out by the data output unit 8 is displayed. The image data of the object is corrected.

For example, when the image data of the object is a table, the image correction means 11 emphasizes the sharpness of the straight line portion and adds a shadow to the root of the table or the like. Then, the data output unit 8 adjusts the exposure amount of the edge of the straight line portion and emphasizes the intensity of the exposure amount of the shaded portion to increase the density of the black toner. For this reason,
In the image data of the screen that is printed out, the image data of the table is corrected to a shape that looks good, and the stereoscopic effect and presence are emphasized. Similarly, when the image data of the object is a can, since the arc portion is smoothed and the shadow is added, the shape is also corrected to a good-looking shape and the stereoscopic effect and presence are emphasized.

That is, the digital copying machine 1 of this embodiment is
If the image data of the screen of the document to be copied includes the image data of the object "can, table, chair, building", this is recognized by the image recognition means 10 and the image correction means 11 is detected.
The image data of the object is corrected to a good-looking shape among the image data of the screen to be copied.

At this time, like a conventional digital copying machine,
Image processing is not uniformly performed on the entire image data of the screen, and only the image data of the object is image-processed, so that the reproducibility of the entire image does not deteriorate. Further, the image processing of the image data of the object is automatically executed only by setting the operation mode, and does not require a complicated operation unlike the conventional DTP system, so that it can be easily executed by a general user.

In particular, since the image recognizing means 10 has the neural networks 12 and 15, automatic recognition of the image data of the object is satisfactorily executed, and the accuracy of recognizing the object can be improved by repeating the learning. Moreover, since the image recognition means 10 has the first neural network 12 for identifying the shape of the image data of the object and the second neural network 15 for determining the image data of the object, the image data of the object follows the shape. Well recognized.

Further, the digital copying machine 1 of the present embodiment is formed so as to copy the image data of the original document in full color, and the color ratio and the color number ratio of the image data of the object are also input to the second neural network 15. Therefore, the second neural network 15 can better identify the image data of the object. It should be noted that the present invention is not limited to the above-described embodiment, and for example, in an image processing device dedicated to black and white, it is possible to identify the image data of the object only from the shape.

Further, in the digital copying machine 1 of the present embodiment, when detecting the color ratio of the image data of the object, the number of each of the 256 color components is integrated in pixel units of the image data of the object, By selecting the color components of in order,
Although the output of the color ratio is output as an example, the present invention is not limited to the above embodiment. For example, if the color ratio is detected as the ratio of the color components of the three primary colors existing in the image data of the object, the color ratio can be directly calculated from the image data read and scanned by the data input unit 2, so that the color component is determined pixel by pixel. Therefore, it is not necessary to integrate the values, and the color ratio can be easily output.

Similarly, in the digital copying machine 1 of this embodiment, when the color number ratio of the image data of the object is detected, the total number of existing color components is integrated, and this number and the total number of color components are calculated. Although the color number ratio is detected as the ratio of the above, the present invention is not limited to the above embodiment. For example, if the color number ratio is detected as the ratio of the sum of the color components of the three primary colors existing in the image data of the object, the color number ratio can be directly calculated from the image data read and scanned by the data input unit 2, and therefore, in pixel units. It is not necessary to determine and integrate the color components, and the color number ratio can be easily output.

When the color number ratio is thus detected by the color components of the three primary colors, the ratio of the three primary colors in each pixel is rounded off,
If they are the same, they are calculated as one. For example, if the ratio of the three primary colors of a certain three pixels is (42, 8, 50) (33,
In the case of 55, 12) (44, 11, 45), these are rounded to (4, 1, 5) (3, 5, 1) (4, 1, 5). Then, since the ratios of the three primary colors are the same in the first and third pixels, this is calculated as one, and the color number ratio detected from the three pixels is "2".

The color components of the three primary colors are generally R of the image data read from the original by the data input means 2.
In addition to the GB component, YMC (Yellow, Magent) of image data generated from this image data for printing by the data output means 8
a, Cyanide) component is also available.

Further, the first neural network 12
Determines the shape of the image data of the object as a ratio of the probabilities of a plurality of preset candidates, so that the shape of the object can be easily and accurately identified. Further, since the feature data is set in advance for each of the image data of the object to be recognized, and the image processing is executed in accordance with this feature data, it is possible to execute the optimum image processing for the object.

In the digital copying machine 1 of this embodiment,
The image data of the object to be corrected is set as "can, table, chair, building", but the present invention is not limited to the above-described embodiment, and the image data of the correction object is "apple, person, It is also possible to set such as "tulip, building". Since "cans, tables, chairs, buildings" are all artifacts, they are characterized by their shapes, but their colors are arbitrary. On the other hand, since "apples, people, tulips" are living things, they are also characterized by shape and color. Therefore, as shown in FIG. 5, when the image data of such an object is recognized and image processing is performed, the shape is corrected as in the present embodiment, and the apple is augmented with red and It is preferable to correct the color such as enhancing the skin color.

That is, when setting the image data of the object to be corrected, it is preferable to set the image processing so that the appearance of the object becomes good, and the characteristic data is set so that such image processing is executed. Will be done.

Further, in the digital copying machine 1 of this embodiment,
If the degree of image processing of the image data of the object is designated by the operation input means 9, the degree of image processing of the image data of the object is adjusted at this designated stage. The degree can be adjusted. The designation of the degree of such image processing is performed by the preset "strong,
In addition to selecting "medium and weak", you can also enter numerical values.
Furthermore, when an object to be image-processed is designated by the operation input means 9, only the image data of the designated object is image-processed, so that the image processing can be executed only for the object desired by the user.

[0070]

According to the first aspect of the present invention, there is provided image input means for inputting image data of a screen, and image recognition means for recognizing preset image data to be corrected from the input image data of the screen. By providing the image correction means for correcting the recognized image data of the correction target and the data output means for outputting the image data of the screen in which the image data of the correction target is corrected, the image data of the screen is corrected. Since only the image data to be corrected can be image-processed, it is possible to obtain image data in which an object looks good on the screen, and image processing is not performed uniformly on the entire screen image data. It is possible to prevent deterioration of reproducibility and it is not necessary to manually perform image processing on only the image data to be corrected, so that a complicated operation is not required.

According to a second aspect of the invention, the image recognition means is
By recognizing the image data to be corrected using the neural network, automatic recognition of the image data to be corrected can be favorably performed, and the recognition accuracy of the object can be improved by repeating the learning. .

According to a third aspect of the invention, the image recognition means is
By having the first neural network that identifies the shape of the image data to be corrected and the second neural network that determines the image data to be corrected according to the identified shape, the image data to be corrected is well recognized according to the shape. can do.

According to the fourth aspect of the present invention, the color information detecting means for detecting the color ratio of the image data to be corrected and inputting it to the second neural network is provided. Since not only the shape but also the color ratio can be referred to, the image data to be corrected can be recognized better.

According to a fifth aspect of the present invention, the color information detecting means outputs the color ratio as a ratio of the color components of the three primary colors existing in the image data to be corrected, so that the image data inputted by the data input means can be detected. Since the color ratio can be calculated directly, the detection of the color ratio can be simplified.

According to the sixth aspect of the invention, the color information detecting means recognizes the image data to be corrected by detecting the number of colors of the image data to be corrected and inputting the color number together with the color ratio into the second neural network. In this case, not only the shape but also the number of colors can be referred to, so that the image data to be corrected can be better recognized.

According to the seventh aspect of the invention, the color information detecting means detects the number of colors as the ratio of the sum of the color components of the three primary colors existing in the image data to be corrected, and the image data input by the data input means is detected. Since it is possible to directly calculate the number of colors from the above, it is possible to simplify the detection of the number of colors.

According to the eighth aspect of the present invention, the first neural network determines the shape of the image data to be corrected by determining the shape of the image data to be corrected as the ratio of the probabilities of a plurality of preset candidates. It can be easily and accurately identified.

[Brief description of drawings]

FIG. 1 is a schematic diagram showing a processing operation in which a digital copying machine that is an image processing apparatus according to an embodiment of the present invention recognizes image data to be corrected.

FIG. 2 is a block diagram showing a digital copying machine.

FIG. 3 is a schematic diagram showing neurons of a neural network.

FIG. 4 is a schematic diagram showing image data including a table as an object.

FIG. 5 is a schematic diagram showing image data including tulips as an object.

FIG. 6 is a flowchart showing a main routine of a processing operation of the digital copying machine.

FIG. 7 is a flowchart showing a correction subroutine.

FIG. 8 is a flowchart showing a subroutine of object recognition.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 image processing device 2 data input means 8 data output means 10 image recognition means 11 image correction means 12, 15 neural network 12 first neural network 15 second neural network 14 color information detection means

Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI Technical display area H04N 1/48 H04N 1/46 A

Claims (8)

[Claims] 1. Data input means for inputting screen image data, image recognition means for recognizing preset correction target image data from the input screen image data, and recognized correction target image data. An image processing device comprising: an image correction unit that corrects the image data; and a data output unit that outputs the image data of the screen in which the image data to be corrected is corrected. 2. The image processing apparatus according to claim 1, wherein the image recognition means recognizes the image data to be corrected using a neural network. 3. A first neural network in which the image recognition means identifies a preset shape from the image data to be corrected, and a second neural network that determines the image data to be corrected according to the identified shape. The image processing apparatus according to claim 1, further comprising: 4. The image processing apparatus according to claim 3, further comprising color information detection means for detecting a color ratio of the image data to be corrected and inputting the color ratio to the second neural network. 5. The image processing apparatus according to claim 4, wherein the color information detecting means outputs the color ratio as a ratio of color components of the three primary colors existing in the image data to be corrected. 6. The image processing apparatus according to claim 4, wherein the color information detecting means also detects the number of colors of the image data to be corrected and inputs the number together with the color ratio to the second neural network. 7. The image processing apparatus according to claim 6, wherein the color information detecting means detects the number of colors as a ratio of the sum of the color components of the three primary colors existing in the image data to be corrected. 8. The image processing apparatus according to claim 3, wherein the first neural network determines the shape as a ratio of the probabilities of a plurality of preset candidates.