JPH05233818A - Image forming device - Google Patents

Abstract

PURPOSE:To smooth a density change by making linear relation between the number of black pixels in an area in a decided size and that density in respect to input image data in the configuration of the image forming device such as a printer. CONSTITUTION:In the image forming device provided with a means for smoothing an image by correcting the input image data corresponding to the arrangement of dot image data to be inputted, a smoothing means 1 is provided with a neural network 2 to transform one dot of the input image data into a subdot pattern composed of plural subdots and to output that pattern.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the structure of an image forming apparatus, such as a printer, and more specifically, an image in which the relationship between the number of black pixels in a certain area and the density is made linear to smooth the density change. Forming apparatus

[0002]

2. Description of the Related Art Currently, printers used as image forming apparatuses are mainly 300 dpi. Therefore, many of the signals output from electronic computers are compatible with 300 dpi. However, 300dpi printers have the drawback that images do not connect smoothly, that is, jaggies are noticeable. In order to eliminate this defect, the pixel density should be increased. However, if the pixel density is increased very simply, in addition to an increase in page buffer and an increase in printer cost due to the high precision of the engine, (1) Bitmap fonts for 300dpi, which are widely used, (2) There is a drawback that the widely used 300dpi input devices (scanner etc.) cannot be used. By the way, in the laser printer, it is difficult to increase the pixel density in the sub-scanning direction, that is, to increase the pitch of paper feed / drum feed, and even if it is possible, it will be expensive. On the other hand, in order to increase the pixel density in the main scanning direction, it suffices to increase the frequency for modulating the laser light, which can be realized relatively easily and at low cost. Therefore, a method has been proposed in which the positioning accuracy of pixels in the main scanning direction is tripled and the size of pixels is changed in 12 steps to improve the image quality related to jaggies (USP 4,847,641). In this method, the pixels of the input image are cut out with a mask of a predetermined size, compared with a pattern written in ROM in advance, and when the pattern matches, the position and size of the corresponding pixel are determined. How to fix it.

FIG. 10 is an explanatory diagram of a conventional pixel correction method using this correction method. In the figure, input data 1
Is cut out in the sample window 2 and compared with the template 3 on the right side of the figure. When the data match, the position and size of the corresponding pixel are changed.

Next, a case where the image quality improving method using the template described in FIG. 10 is applied to smoothing the density change will be described. FIG. 11 shows the relationship between the normalized density and the number of black pixels in, for example, a partial area of an image, here, a 4 × 4 dot matrix. As shown in the figure, the normalized density is not linear with respect to the number of black pixels, and in order to make the density linear, it is necessary to make the ratio of the number of black pixels smaller than the value given by the normalized density. Is.

In FIG. 12, the density is smoothed according to the number of black pixels in a 4 × 4 dot matrix, and the density pattern in the matrix is distributed from black pixels in a region to a black pixel position above and below. It is explanatory drawing of the conversion example at the time of converting to a left-right symmetrical and centralized type which concentrated on one part. In the figure, levels 0, 1, 2, ...
The number of black pixels in each dot matrix is 0, 1, 2, ...
.. represents 15 or 16 cases, and for level 1, for example, 16 types of matrices exist depending on the position of the black pixel.

In FIG. 12, the matrix image data of levels 0 to 16 shown on the left side of the arrow is converted into a pattern of the number of dots shown on the right side of the arrow to smooth the density change. .. For example, for 16 types of level 1 matrix, for example, by using the sub-dot pattern described later, each of the four central dots is partially black, and the total black area corresponds to 0.5 pixel area. Is converted to.

[0007]

However, as shown in FIG.
In general, the method described in Section 1 needs to have many template patterns, which slows down the speed for comparison and increases the capacity of the memory for storing template patterns, and the pixel arrangement that matches the limited template patterns. There was a problem that only the correction was done about.

When smoothing the density change as described with reference to FIG. 12, the level 2 shown in FIG.
It was necessary to prepare 120 types of template patterns, and it was almost impossible to prepare and modify template patterns corresponding to all levels.

According to the present invention, one dot of input image data is
As will be described later, conversion to a subdot pattern composed of a plurality of subdots realizes smoothing of density change.

[0010]

FIG. 1 is a block diagram showing the principle of the present invention. FIG. 1 is a principle block diagram of an image forming apparatus having means for smoothing an image by correcting the input dot image data according to the arrangement of the input dot image data.

In FIG. 1, the smoothing means 1 converts one dot of the input image data into a plurality of subdots, for example, 8 dots.
The neural network 2 is provided which converts a sub-dot pattern consisting of individual sub-dots and outputs the sub-dot pattern. This neural network is a hierarchical neural network in which the layers are fully connected, for example, an input layer, an intermediate layer, and an output layer.

[0012]

In the present invention, the smoothing means 1 of FIG. 1 is used for the purpose of smoothing the density change of the image.
Then, the smoothing means 1 makes an area composed of a plurality of dots into one block, that is, a matrix, and the output result of the sub-dot pattern for each input dot is a matrix as described as the conversion result in FIG. The input dot image data is corrected so as to form vertically and horizontally symmetrical image data (concentrated pattern).

For example, in FIG. 12, 16 indicating level 1 is shown.
When any one of the types of matrixes is input, the image data which is vertically symmetrical with respect to the matrix image data is output. This output image data is a sub-dot pattern consisting of eight sub-dots for each dot, and only one of the eight sub-dot patterns for the central four pixels. Black, total 4
By making each sub-dot black, the portion corresponding to the dot area of 0.5 becomes black. For each of the 12 surrounding pixels, all 8 sub-dots are white, so that the density change is smoothed and there is a concentrated area gradation in which black sub-dots are gathered in the center of the matrix. Will be realized.

[0014]

EXAMPLE FIG. 1 shows an example of a sub-dot pattern used in the present invention. In the figure, the input image data of one dot is divided into eight subdots. Original 1
For each black dot, all eight subdots are black.

FIG. 3 is an explanatory view of the outline of image data correction in the present invention. In the figure, if the input bitmap is composed of 8 × 8 pixels, the image data is corrected in units of a matrix of 4 × 4 pixels. That is, as shown in FIG. 12, the conversion result shown on the right side is output according to the level of the matrix composed of 4 × 4 pixels. In FIG. 3, the upper right and lower left matrices of the four matrices are for level 2, and the number of black subdots is smaller than that for level 1 of the upper left and lower right. Is doubled.

FIG. 4 shows an image forming apparatus according to the present invention.
FIG. 3 is a configuration block diagram of an embodiment of a neural network 2 that constitutes the smoothing means 1. In the figure, the input layer only distributes the output to the units in the intermediate layer, and is therefore represented by the register 10. On the other hand, the middle layer has 16 neurons 11 ₁ ...
11 ₁₆ and the output layer is composed of 8 neurons 12 _{1 to} 12 ₈ for outputting data of 8 subdots. Here, the number of neurons in the middle layer is "16"
Corresponds to the size of the matrix, that is, the number of 16 pixels in the matrix, as will be described later.

In FIG. 4, the neuron in the middle layer has, as will be described later, 1 bit of 63 bits each indicating input image data of 9 × 7 dots indicating the size of the window, as one of the inputs, as described later. 63 AND circuits 16 _{1 to} 16 ₆₃ in which the determined coefficients are given to the other input, an adder 17 for adding the outputs of these AND circuits, a register 18 for storing the addition result by the adder 17, a register It is composed of, for example, a read only memory (ROM) 19 for scaling the output of 18 and a three-state register 20 for outputting the output of the ROM 19 to the output layer when the output enable (OE) signal is input. .. In addition, each unit of the output layer adds a coefficient buffer 22 in which coefficients described later are stored, a multiplier 23 that multiplies the output of each neuron of the intermediate layer and the output of the coefficient buffer 22, and the output result of the multiplier 23. And an register 24 for holding the output of the adder 24.

The operation of the neural network of FIG. 4 will be described with reference to FIGS. FIG. 5 shows the movement of the matrix within the window, and FIG. 6 shows the coefficients given to the 63 AND circuits in the hidden layer neuron 11 ₁ shown in FIG.

In FIG. 5, a window is formed by 9 × 7 pixels, and at this time, for example, 9 × 7 lines of data are input from a bitmap memory (not shown), and the window is It is assumed that a sub-dot pattern indicating the correction result for the central dot in the above is output from each unit (neuron) of the output layer in FIG. In FIG. 5, ...
The position of the 4 × 4 matrix in the window at each successive time point is shown. That is, at a certain point of time, for example, if the matrix of 4 × 4 dots in the upper left of FIG. 3 is in the hatched portion, the pixel A in the center of the window corresponds to the upper left position in this matrix, and the data Is white. At the next time,
Since the window is laterally displaced by one dot, the data at this position, 'A' in FIG. 5, comes to the left of the center pixel of the window.

In FIG. 5, ~ indicates the position of the matrix when the window is laterally shifted by one dot, while ~ indicates the dot of the second line of the matrix composed of 4 × 4 dots. It shows the case where it comes to the center of the window. As a result, for example, FIG.
Of the four matrices, the upper left matrix is the win

[0021]

[Outer 1]

Since the hatched portion shows the data in one matrix, the number of black pixels in the hatched portion is ~.
At the time of outside 2, they are all equal.

[0023]

[Outside 2]

In FIG. 5, ˜out3 is thus the matrix win at each time point.

[0025]

[Outside 3]

The positions on the dough are shown, but at the same time, each of them is in each of 16 hidden layer neurons 11 _{1 to} 11 ₁₆ .
The coefficients for 63 AND circuits are also shown. In FIG. 4, 63 bits of input data for 63 dots in the window are given in parallel to all neurons 11 _{1 to} 11 _{16 in} the intermediate layer via the register 10. These 63 bits are applied to one input of each of the AND circuits 16 _{1 to} 16 ₆₃ , one bit at a time, as described above, and the logical product with the coefficient input to the other becomes the output of each AND circuit. Here, the neurons 11 _{1 to} 11 ₁₆
The coefficient for the AND circuit to which 1-bit data for each of the 63 dots in the window is input is
The data in the hatched area in Fig. 5 to 4 is input.

[0027]

[Outside 4]

It is set to "1" for the de-circuit, and set to "0" for the AND circuit to which the data of the other part is input. FIG. 6 shows the coefficients given to the 63 AND circuits in the hidden layer neuron 11 ₁ . These coefficients are all "1" in the AND circuit for the hatched part in the window and all "0" in the AND circuit for the other parts. Therefore, the output of the adder 17 eventually represents the number of black pixels in the matrix, and becomes "1" for the upper left matrix in FIG.

Similarly, in FIG.
It represents the coefficient for 63 AND circuits in 1 _2. The coefficient is "1" for the AND circuit to which the dot data corresponding to the hatched position is input, and for the other AND circuits. Will be '0'. The following is the same, and the outer 5 in neurons 11 ₁₆ of the intermediate layer

[0030]

[Outside 5]

A circuit in which "1" is input as a coefficient is represented by a hatched portion. As mentioned above, the hatched area in the outer 6 is the same mat

[0032]

[Outside 6]

Since it tracks the lix, each neuron 1 in the intermediate layer corresponds to the number of black pixels in the matrix.
1 _{1 to} 11 ₁₆ output the same value with a time lag. Next, the operation of the output layer unit will be described with reference to FIGS. In FIG. 4, each of the neurons 11 ₁ to ₁ 1 in the intermediate layer
The output of 1 ₁₆ is shown in FIG.

[0034]

[Outside 7]

In the above, when the output enable signal is input to the corresponding intermediate layer neurons, the output signals are sequentially output to each unit in the output layer. FIG. 7 shows an output process of the image data correction result for the upper half of the upper left matrix of FIG. Further, FIG. 7 also corresponds to the time points from to in FIG.

In FIG. 7, the upper left diagram corresponds to the position of FIG. 5, and at this position, each neuron 12 in the output layer is
The outputs of _{1 to} 12 ₈ need to be all white, that is, '0', corresponding to the dot at the upper left of the conversion result on the right side of level 1 in FIG. Coefficients for outputting such data are stored in the coefficient buffer 22 in advance by learning. At the time of FIG. 5, the X address indicating the horizontal movement of the matrix in FIG. 7 is 0, and the Y address indicating the vertical movement of the matrix is also 0. Correspondingly, each of the output layers in FIG. The coefficient 1 for the output of 11 ₁ of the intermediate layer neuron is stored in the coefficient buffer in the neuron, and by using this coefficient, the output of the output layer, that is, the subdot pattern is all white.

Similarly, the second diagram from the left in FIG. 7 corresponds to the position in FIG. 5, and the conversion result for level 1 in FIG. The pattern must be printed. At this position, the X address is 1 and the Y address is 0. As shown in FIG. 8, for this address, the coefficient buffer 22 is subjected to the point of FIG. 5, that is, the second neuron 11 of the hidden layer neurons. _The coefficient to be multiplied by the output of ₂ , coefficient 2, is stored. By using this coefficient, the output of each neuron in the output layer, that is, the sub-dot pattern is all white.

Similarly, FIG. 7 shows how the sub-dot pattern for the pixel at the center of the window is output. As described above, FIG. 7 shows the process of outputting the upper half of the conversion result for level 1 in FIG.
When the address and Y address are '1', only the rightmost subdot is black, and when the X address is 2 and the Y address is 1, the leftmost subdot is black and all other subdots are white. Becomes

FIG. 9 shows a time chart of the correction process.
Both the X and Y addresses shown here represent only the lower 2 bits, and "0" is the initial value of each. When 63 dots of window data 1 from the input layer is given to the intermediate layer, each intermediate layer neuron calculates the sum of products.
The result of scaling is three-state register 2
Hold at 0 once. After that, the intermediate layer neurons 11 ₁ to
The intermediate layer output enable is sequentially set to "1" in order to serially output the output of 11 ₁₆ to the output layer. In the output layer, since the X and Y addresses are both "0", the coefficient group shown on the leftmost side is selected from the coefficient buffers shown in FIG. From then on, the coefficients are sequentially applied to the multiplier 23 to take the product with the output of each intermediate layer neuron. In other words, the multiplier 23 first outputs to the neuron 11 ₁ and the leftmost uppermost "coefficient 1" in FIG. 8, and then outputs the neuron 11 _{2 to} the _second uppermost "0", ... ..Neuron 1
The output of 1 ₁₆ and "0" in the lowermost row on the left side are sequentially given. When the sum of products calculation up to the neuron 11 ₁₆ is completed, the corrected output (sub-dot pattern) corresponding to the window data 1 is determined.

Next, the window data 2 shifted by one dot to the right which is the scanning direction of the laser beam is given, and at the same time, the X address is updated and the same operation is performed. Output layer is X
Since the address is "1", the selected coefficient becomes the second coefficient group from the left in FIG. 8, which is sequentially given to the multiplier 23 by the output layer coefficient enable, and the output corresponding to the window data 2 is determined. When these operations are sequentially performed and the processing for one dot line is completed, the Y address is updated and the processing for the next dot line is performed. This operation is performed over the entire page, and the conversion process for one page is completed.

After all, the X and Y addresses indicate the dot position to be corrected in the 4 × 4 matrix. In other words, when X = 0 and Y = 0, the uppermost left corner of the 4 × 4 matrix, and when X = 1 and Y = 0, 2 from the uppermost left.
The dots are the same as the bottom right corner when X = 3 and Y = 3. Therefore, the processing results of the X addresses 0 to 3 are for the same matrix.

[0042]

As described in detail above, according to the present invention, the density change with respect to the number of black pixels in the matrix can be made to be close to linear, and the density gradation is improved when a halftone image is recorded. Greatly improved. It is also possible to convert, for example, a dispersed density pattern for a thermal transfer printer into a concentrated density pattern for an inkjet printer or an electrophotographic printer. Furthermore, for example, although there are 120 types of level 2 patterns, they are all corrected to the same pattern, and the feature of the neural network of detecting similar patterns is fully exerted.

[Brief description of drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 is a diagram showing an example of a sub-dot pattern in the present invention.

FIG. 3 is a diagram illustrating an outline of image data correction according to the present invention.

FIG. 4 is a block diagram showing a configuration of an embodiment of a neural network.

FIG. 5 is a diagram showing movement of a matrix within a window.

FIG. 6 is a diagram showing an example of coefficients given to an AND circuit in an intermediate layer neuron.

FIG. 7 is a diagram showing an example of an output result of an output layer neuron.

FIG. 8 is a diagram showing stored contents of a coefficient buffer in each neuron of an output layer.

FIG. 9 is a diagram showing a time chart of an embodiment of matrix data correction processing.

FIG. 10 is a diagram illustrating a conventional example of an image data correction method.

FIG. 11 is a diagram showing the relationship between the number of black pixels in a matrix and the normalized density.

FIG. 12 is a diagram for explaining smoothing of density change and conversion of a dispersed density pattern into a concentrated density pattern.

[Explanation of symbols]

1 smoothing means 2 coefficients of the neural network 10 input layer register 11 ₁ to 11 ₁₆ hidden neurons 12 ₁ to 12 ₈ output layer neurons 16 ₁ to 16 ₆₃ and the circuit in the intermediate layer neuron 22 output layer neuron buffer

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kazuhiko Sato 1015 Kamiodanaka, Nakahara-ku, Kawasaki, Kanagawa Prefecture, Fujitsu Limited (72) Inventor Jun Shio, 1015, Kamedotachu, Nakahara-ku, Kawasaki, Kanagawa Prefecture, Fujitsu Limited

Claims (2)

[Claims] 1. By correcting the input dot image data according to the array of the input dot image data,
In an image forming apparatus having means for smoothing an image, the smoothing means (1) converts one dot of the input dot image data into a sub-dot pattern composed of a plurality of sub-dots, and thereby the density of the image. An image forming apparatus comprising a neural network (2) for smoothing and outputting changes. 2. The smoothing means (1) defines an area consisting of a plurality of dots as one block, and outputs the output result of the sub-dot pattern for each input dot to the upper and lower and left and right target images in the block. The image forming apparatus according to claim 1, wherein the input dot image data is corrected so as to be formed.