JPH10294875A - Color printer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To vary color conversion characteristic automatically with high accuracy by relieving the load of deciding a conventional color conversion characteristic required by the color printer having a nonlinear input/output characteristic. SOLUTION: Only input color information in the learning mode is used and a 1st neural network 11 applies color conversion to the input color information to provide the output of a learning signal, and the input output characteristic decision section 13 uses a 2nd neural network 14 to produce a teacher signal. Both the signals in pairs are given to the 1st neural network 11, where an error is calculated and conducts learning till the error is sufficiently small by the error inverse propagation method to decide a color conversion characteristic automatically. Or the 1st neural network is eliminated and a print section 12 is connected directly in series with the 2nd neural network so as to change the characteristic of the entire printer.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a color printing apparatus which receives color image data as input color information and prints it on paper or the like, corrects the color data, and easily changes the color conversion characteristics of the entire apparatus. For example, a color printer or a color facsimile
It can be applied to such as.

[0002]

2. Description of the Related Art As shown in FIG. 7, a color printing apparatus mainly comprises a color conversion section 71 and a printing section 12. Conventionally, color conversion in this type of apparatus is performed by a method using a correction function or an LUT (LUT). A look-up table). In the method using the correction function, a simple linear conversion or a function obtained by adding a high-order term to the non-linear characteristic of the printing unit is used as the correction function. However, determination of the correction function must be manually performed, and it is difficult to find an optimum correction function. On the other hand, the LUT method is a method in which output color information corresponding to input color information is paired into a table, but a huge memory is required to make all combinations into a table. For example, an input value is calculated using a certain interpolation method with a value stored in a table for input color information not included in the table. Generally, interpolation methods include eight-point interpolation, six-point interpolation, five-point interpolation, and four-point interpolation. However, since the calculation formula is a simple formula such as linear interpolation, the conversion accuracy may not be sufficient. . In any case, these methods do not provide optimal color conversion, because they cannot cope with the nonlinear input / output characteristics of the printing unit.

Therefore, in recent years, it has been considered to use a neural network for the color conversion unit. As the neural network, a hierarchical neural network having an input layer, a plurality of intermediate layers (hidden layers), and an output layer is used. ing. Generally, a color can be represented in a three-dimensional space, so that the neural network used is often a three-input three-output hierarchical neural network. When the input of the printing unit is CMYK (cyan, magenta, yellow, black), a three-input four-output hierarchical neural network, a three-input three-output hierarchical neural network, and a three-input one-output hierarchical neural network are used. Are combined. Neural networks have the ability to reproduce non-linear inputs and outputs, so it is thought that good color conversion can be performed depending on the learning method.

[0004]

However, at present,
For example, as in the color data correction processing method described in Japanese Patent Application Laid-Open No. 5-292332, the neural network merely learns and uses the inverse characteristic of the printing unit. The reason is realized by a simple concept, such as inputting color information to the printing unit as a teacher signal, and pairing output color information of the printing unit as an input signal to the neural network as learning data. Because you can. If a neural network based on such a concept is used as a color conversion unit, it is true that the input color information to the color printing apparatus and the output color information coincide with each other in theory, so that it is chromatically correct. However, actually, for example, when a digital still image displayed on a CRT is printed, the displayed color and the printed color are different. This is because the color reproduction range of the printing unit does not reach the image reproduction range of the CRT. Therefore, even though the above method is theoretically correctly reproduced in chromaticity, it is difficult to say that appropriate color conversion is performed at present. On the other hand, a method and a system configuration for providing an arbitrary input / output to the color printing apparatus have not been established, and it is difficult to realize the design even if the input / output characteristics are freely designed.

The present invention has been made in view of such problems in the prior art, and enables a color conversion unit to use a neural network to freely and accurately set the color information input / output characteristics of the entire apparatus. Usually, a color printing apparatus has only one color conversion characteristic,
It is an object of the present invention to provide a variable color printing apparatus.

[0006]

First, the principle of the present invention will be described. In the configuration of FIG. 8, X1 is an input signal to the neural network 11, X2 is an output of the neural network 11 and an input of the printing unit 12, and X3 is an output of the printing unit 12. These X1, X2, and X3 are three-dimensional to represent color information, but are collectively represented as vectors for simplicity. When the input / output characteristics of the color printing device 81 are represented by F (), the input / output characteristics of the neural network 11 are represented by H (), and the input / output characteristics of the printing unit 12 are represented by G (), these relationships are expressed as follows. You.

X3 = F (X1) (1) X3 = G (X2) (2) X2 = H (X1) (3) The inverse function of equation (2) is represented as G ′ (). X2 = G '(X3) (4) The equation (1) is substituted into the equation (4). X2 = G '(F (X1)) (5) X2 is eliminated from the equation (5). H (X1) = G '(F (X1)) (6) From equation (6), it is understood that H (), that is, what input / output characteristics the neural network 11 should have.

Although the above is theoretically as described above, in this case, there remains a problem of how to obtain the inverse function G '() of the printing unit. This is because G ′ () is a characteristic of the printing unit 12 and cannot be generally expressed by a mathematical expression, and therefore it is difficult to obtain the inverse function. Therefore, the present invention is intended to solve this problem by preparing a second neural network and causing the second neural network to learn the inverse characteristic of the printing unit 12.

Next, a neural network learning method will be described with reference to FIG. In this configuration, the first neural network 11 for correcting the input color information or converting the color information to another color space is to be learned. Here, the input / output characteristic determination unit 13 has F (), which is the characteristic of the entire apparatus, and the second neural network 14 has learned in advance the inverse characteristic of the printing unit, which is represented by G ′. It is equivalent to ().

As the learning data for learning the first neural network, only an appropriate number of input signals may be prepared. An input signal X1 for learning is input to the first neural network to obtain an output signal H (X1). The input signal X1 is also input to the input / output characteristic determination unit at the same time, and its output is input to the second neural network, and finally obtains the output G '(F (X1)). And
This output G '(F (X1)) is given as a teacher signal of the first neural network, and H (X1) and G' (F
The error of (X1) may be calculated, and the first neural network may be trained by the error backpropagation method until the error becomes sufficiently small.

According to a first aspect of the present invention, there is provided a color printing apparatus for performing printing based on an input signal carrying color information, a first neural network for correcting the input color information or converting the color information to another color space, A printing unit for receiving and printing an output signal output from the first neural network, and determining input / output characteristics which have input / output characteristic information of the entire apparatus, convert input color information according to the input / output characteristics, and determine an output; And a second neural network that learns the inverse characteristics of the input / output characteristics of the printing unit and converts the output signal of the input / output characteristics determination unit in accordance with the inverse characteristics. When the characteristic information is changed, the output signal of the second neural network which receives and converts the output signal of the input / output characteristic determining unit is converted to a first neural network. By re-learning is given to the network, it is intended to vary the input-output characteristics of the entire device.

According to a second aspect of the present invention, there is provided a color printing apparatus for performing printing based on an input signal carrying color information, a receiving unit for receiving data from the outside, and correcting the input color information or other color space. A first neural network which converts the input color information into a color image, a printing unit which receives an output signal output by the first neural network and performs printing, and has input / output characteristic information of the entire apparatus. And a second neural network that learns the inverse characteristics of the input / output characteristics of the printing unit and converts the output signal of the input / output characteristics determination unit according to the inverse characteristics. Replacing the input characteristic determination unit with new characteristic information based on the color information from the communication unit and the input / output characteristic information of the entire device. , By re-learning gives an output signal of said second neural network to the first neural network, is intended to replace the input-output characteristics of the entire device.

According to a third aspect of the present invention, in the second aspect of the present invention, a storage unit for storing input / output characteristic information of the entire apparatus is additionally provided, and based on designated data from the communication unit, The characteristic information of the input / output characteristic determining unit is to be replaced with the input / output characteristic information already stored in the storage unit.

According to a fourth aspect of the present invention, there is provided a color printing apparatus which performs printing based on an input signal carrying color information, has input / output characteristic information of the entire apparatus, and converts input color information according to the input / output characteristic. An input / output determination unit that determines an output, a neural network that learns characteristics and converts an output signal of the input / output determination unit according to the characteristics, and a printing unit that receives an output signal of the neural network and performs printing. Further, the input / output characteristics of the entire apparatus are made independent of the input / output characteristics of the printing unit by making the input / output characteristics of the entire printing apparatus independent of the input / output characteristics of the printing unit.

According to a fifth aspect of the present invention, in a color printing apparatus for performing printing based on an input signal carrying color information, a receiving unit for receiving external data, and input / output characteristic information of the entire apparatus are input. An input / output determination unit that converts color information according to the input / output characteristics and determines an output, a neural network that learns characteristics and converts an output signal of the input / output determination unit according to the characteristics, and receives an output signal of the neural network. A printing unit that performs printing by using the neural network as a reverse characteristic of the input / output characteristics of the printing unit, and the color information from the communication unit and the input / output characteristics information of the entire apparatus. It is intended to replace the input / output characteristics of the entire device by replacing the input / output characteristics of the input / output characteristic determination unit based on this. That.

According to a sixth aspect of the present invention, in the fifth aspect of the present invention, a storage unit for storing input / output characteristic information of the entire apparatus is additionally provided, and based on designated data from the communication unit, It is intended to replace the characteristic information of the input / output characteristic determination unit with the input / output characteristic already stored in the storage unit.

[0017]

Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram illustrating an example of an embodiment of a color printing apparatus according to claim 1. The input / output characteristic determining unit 13 and the second neural network 14 are directly connected, while the first neural network 11 is directly connected to the printing unit 12. The second neural network 14 is connected to the first neural network 11 so that its output can be input to the first neural network 11. The first neural network 11 must have a configuration that can be learned by the back propagation method or the like, but the second neural network 14 does not need to be capable of learning. It is sufficient that the input / output characteristics (printing unit characteristics) after learning with a plurality of learning data pairs of output values of the unit 12 as input signals. Input / output characteristic determination unit 13
May be represented by a function or a table.

This color printing apparatus has two modes, a printing mode and a learning mode. To switch the mode, the device may be provided with a mode switch, or data input to the device may be provided with a header by attaching a header, and the mode may be internally switched according to the value of the flag. . In the print mode, the input signal is converted by the first neural network and then input to the printing unit 12 to perform printing. In the learning mode, the input signal (learning data) is input to the first neural network 11 and also to the input / output characteristic determining unit 13. The output of the input / output characteristic determination unit 13 is directly input to the second neural network 14, and the output data is treated as a teacher signal of learning data. The teacher signal and the first neural network 11
Of the first neural network 1 until the error is sufficiently reduced by the error back propagation method.
Train 1 In such a configuration, when the input / output characteristic determination unit 13 is replaced because the characteristic needs to be changed, the characteristic of the entire color printing apparatus is automatically requested by learning the first neural network 11. Can be replaced.

FIG. 2 is a block diagram showing an embodiment of the color printing apparatus according to the second aspect. In this configuration, a receiving unit 21 is further added to the configuration shown in FIG. 1, and the receiving unit 21 is connected to the input / output characteristic determining unit 13 and the input side of the first neural network 11.
The receiving unit 21 has a role of receiving various data from outside such as a telephone circuit and a network. In addition, the input / output characteristic determination unit 13 must be formed of a rewritable storage element. This color printing apparatus has three modes: a print mode, a learning mode, and a download mode. For print mode and learning mode,
1 is the same as the color printing apparatus shown in FIG. 1 except that the receiving unit 21 receives data by communication. In the download mode, the content of the input / output determination unit 13 is rewritten with the data received by the receiving unit 21. ing.

FIG. 3 is a block diagram showing an embodiment of the color printing apparatus according to the third aspect. In this configuration, a storage unit 31 is further added to the input / output characteristic determination unit 13 in the configuration of FIG. The storage unit 31 holds data for rewriting the contents of the input / output characteristics determination unit 13, thereby storing the contents of the input / output characteristics determination unit 13 without downloading the input / output characteristic data at each printing. The content in the section 31 can be replaced.

FIG. 4 is a block diagram showing one embodiment of the color printing apparatus according to the fourth aspect. This configuration is the same as the configuration of FIG. 1 except that the first neural network 11 is removed and the printing unit 12 is directly connected to the second neural network 14. As in this configuration, the second neural network 14 and the printing unit 1
2 are connected in series, the characteristics of the neural network 14 and the printing unit 12, which have learned in advance the inverse characteristics of the characteristics of the printing unit 12, cancel each other out, and the characteristics of the entire apparatus match those of the input / output characteristics determination unit 13. Will be done. According to this configuration, the characteristics of the entire device can be changed only by replacing the input / output characteristics determination unit 13.

FIG. 5 is a block diagram showing an embodiment of the color printing apparatus according to the fifth aspect. In this configuration, a receiving unit 21 is added to the configuration of FIG. Receiver 21
Is connected to the input side of the input / output characteristic determination unit 13, and the input / output characteristic determination unit 13 must be formed of a rewritable storage element as in the configuration of FIG. As a result, the input / output characteristic determination unit 13 can receive the data whose contents are rewritten by communication.

FIG. 6 is a block diagram showing an embodiment of the printing apparatus according to claim 6. In this configuration, a storage unit 31 is further added to the input / output characteristic determination unit 13 in FIG. The storage unit 31 holds data for rewriting the content of the input / output characteristic determination unit 13, so that the input / output characteristic data is not downloaded each time printing is performed.
The content of the input / output characteristic determination unit 13 can be replaced with the content in the storage unit 31.

[0024]

As described above, according to the present invention, the combination of the input / output characteristic determining unit having the characteristics of the entire apparatus and the neural network in which the inverse characteristic of the printing unit is learned in advance is used. Unlike the conventional color printing device that determines the input / output data of the color conversion unit while considering the characteristics of the printing unit so that the expected characteristics can be obtained, just determine the characteristics of the entire device. This has the effect that correct color conversion can be performed automatically.

According to the first aspect of the present invention, it is possible to make the output characteristic data of the color information of the entire apparatus variable by learning the first neural network when changing the input characteristic determining unit.

According to the second aspect of the present invention, since data representing input / output characteristics of the entire apparatus other than the color image data can be received by communication, the data is replaced with the received characteristic data, and , The input / output characteristic data of the entire apparatus can be easily replaced.

According to a third aspect of the present invention, in addition to the effect of the second aspect of the present invention, a storage unit for storing data representing input / output characteristics is added, and a plurality of characteristic data are stored in the storage unit. By doing so, it is not necessary to download the data every time the characteristic data of the entire apparatus is replaced as in the invention according to claim 2, and the data can be replaced more easily.

According to a fourth aspect of the present invention, the first neural network is eliminated, and the second neural network and the printing unit are connected in series to cancel each other's characteristics. The characteristics of the entire device can be changed without training the first neural network by simply replacing the input / output characteristics determining unit, since the characteristics of the entire device can be changed.
The time and effort required to learn the neural network can be saved.

According to a fifth aspect of the present invention, in addition to the fourth aspect of the present invention, there is further provided a receiving unit for receiving data representing the input / output characteristics of the entire apparatus from the network, and transmitting the characteristic data of the input / output characteristics determining unit by communication. Since the information can be received, the replacement of the characteristic data can be easily performed as compared with the color printing apparatus according to the fourth aspect.

According to the invention of claim 6, in addition to the effect of the invention of claim 5, a storage unit for storing data representing input / output characteristics is added. By storing the data, the necessity of downloading the data every time the characteristic data of the entire apparatus is replaced as in the invention according to claim 5 can be eliminated, and the data can be replaced more easily.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an embodiment of a color printing apparatus according to claim 1.

FIG. 2 is a block diagram showing an embodiment of a color printing apparatus according to claim 2;

FIG. 3 is a block diagram showing an embodiment of a color printing apparatus according to claim 3;

FIG. 4 is a block diagram showing an embodiment of a color printing apparatus according to claim 4;

FIG. 5 is a block diagram showing an embodiment of a color printing apparatus according to claim 5;

FIG. 6 is a block diagram showing an embodiment of a color printing apparatus according to claim 6;

FIG. 7 is a block diagram illustrating a main part of a conventional color printing apparatus.

FIG. 8 is a block diagram for explaining the principle of the present invention.

FIG. 9 is a block diagram showing a configuration used for learning a neural network in the present invention.

[Explanation of symbols]

11: first neural network, 12: printing unit,
13: input / output characteristic determination unit, 14: second neural network, 21: reception unit, 31: storage unit, 71: color conversion unit, 81: color printing device.

Claims (6)

[Claims] 1. A color printing apparatus for performing printing based on an input signal carrying color information, comprising: a first neural network for correcting input color information or converting the color information to another color space; and a first neural network. A printing unit that receives and outputs an output signal to perform printing, an input / output characteristic determination unit that has input / output characteristic information of the entire apparatus, converts input color information according to the input / output characteristics, and determines output, and the printing unit A second neural network that learns the inverse characteristics of the input / output characteristics of the input / output characteristics and converts the output signal of the input / output characteristics determination unit in accordance with the inverse characteristics. The output signal of the second neural network, which receives and converts the output signal of the input / output characteristic determination unit, is given to the first neural network. A color printing apparatus wherein the input / output characteristics of the entire apparatus are made variable by re-learning. 2. A color printing apparatus for performing printing based on an input signal carrying color information, a receiving unit for receiving data from the outside, and a first unit for correcting the input color information or converting the input color information into another color space. A neural network, a printing unit for receiving and outputting an output signal output from the first neural network, and having input / output characteristic information of the entire apparatus; converting input color information according to the input / output characteristic to determine an output; And a second neural network that learns the inverse characteristics of the input / output characteristics of the printing unit and converts the output signal of the input / output characteristics determination unit according to the inverse characteristics. The input characteristic determining unit is replaced with new characteristic information based on the color information and the input / output characteristic information of the entire apparatus, and the second neural network is replaced by the second neural network. A color printing apparatus characterized in that input and output characteristics of the entire apparatus are replaced by giving an output signal of the network to a first neural network and re-learning the same. 3. A storage unit for storing input / output characteristic information of the entire apparatus, wherein characteristic information of the input / output characteristic determination unit is stored in the storage unit based on designated data from the communication unit. 3. The color printing apparatus according to claim 2, wherein the input / output characteristic information is stored in the printer. 4. A color printing apparatus which performs printing based on an input signal carrying color information, has input / output characteristic information of the entire apparatus, and converts input color information according to the input / output characteristics to determine an output. An output determining unit, a neural network that learns characteristics and converts an output signal of the input / output determining unit according to the characteristics, and a printing unit that receives and prints an output signal of the neural network, further comprising the neural network. A color printing apparatus characterized in that the input / output characteristics of the entire apparatus do not depend on the input / output characteristics of the printing unit by making the input / output characteristics of the entire printing apparatus independent of the input / output characteristics of the printing unit as the characteristics learned by the printing unit. 5. A color printing apparatus for performing printing based on an input signal carrying color information, a receiving section for receiving data from the outside, and input / output characteristic information having input / output characteristic information of the entire apparatus. An input / output determining unit that converts and determines an output according to a characteristic, a neural network that learns a characteristic and converts an output signal of the input / output determining unit according to the characteristic, and a printing unit that receives and prints an output signal of the neural network Further, the characteristic to be learned by the neural network is an inverse characteristic of the input / output characteristics of the printing unit, and the input / output characteristics based on the color information from the communication unit and the input / output characteristic information of the entire apparatus. Color printing characterized by replacing the input / output characteristics of the entire device by replacing the input / output characteristics of the determination unit. apparatus. 6. A storage unit for storing input / output characteristic information of the entire apparatus, wherein characteristic information of said input / output characteristic determination unit is stored in said storage unit based on designated data from said communication unit. 6. The color printing apparatus according to claim 5, wherein the input / output characteristics are replaced with the input / output characteristics already stored in the printer.