JP2002052758A - Method of processing image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method of processing images which enables to obtain appropriate, good and desired output images even when a frequency response becomes different by a direction of an X-Y space or when a frequency response of an output device becomes unstable in a high frequency region. SOLUTION: According to this method of processing images, output images expressing a halftone are obtained by executing dithering with the use of a stochastic screen to each pixel of input images. The stochastic screen is formed with two frequency characteristics, i.e., frequency characteristics of the output device and blue noise characteristics taken into account.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method, and more particularly to a dithering process using a stochastic screen, which is capable of obtaining a desired image even when the frequency response of an output device in a high frequency region is unstable. And an image processing method capable of obtaining an output image of

[0002]

2. Description of the Related Art A stochastic screen used in an image processing method is also called an FM screen.
Includes pattern dither, white noise masks, more recently blue noise masks, green noise masks, and the like. Generally, the error diffusion method is also included in the stochastic screen, but in the present invention, the stochastic screen is limited to a method using the dither method. At present, FM screens used for shading such as printing are close to pulse frequency modulation but have a large degree of modulation, so they cannot be called FM. Therefore, it can be said that the stochastic screen is the optimal expression.

Conventionally, image processing methods as pseudo gradation expression methods are roughly classified into an error diffusion method and a mask method.

The mask method is basically a method of determining the output value of the pixel of the output image by making the pixels of the original image correspond to the elements of the threshold matrix on a one-to-one basis at the time of binarization. Generally, the dither method is known. Although the dither method has a high processing speed, it is not always possible to obtain an image of good quality.

On the other hand, the error diffusion method has better image quality than the dither method, but has a lower processing speed.

Therefore, in general, the dither method is used when the degree of demand for image quality is low, and the error diffusion method is used when the degree of image quality is high.

However, in the perturbation error diffusion method, which is one of the methods of diffusing an error in the error diffusion method, when the spatial frequency spectrum of a binarized pattern generated at each gradation has a blue noise characteristic, It was shown that it was excellent. Such a blue noise pattern has an aperiodic and uncorrelated structure, and is characterized by having no low-frequency granularity.

Here, when the characteristics of the blue noise in the spectrum space are made to correspond to the characteristics of the blue noise pattern in the real space, the granularity of the dot pattern is improved in the real space when the low frequency component is small in the spectrum space. If the spectrum is aperiodic in the spectrum space, a relationship can be derived in the real space in which a virtual pattern generated by interference with an input image or a periodic pattern having a mask size as a cycle as in the dither method is not generated. In other words, it can be said that both a low frequency component in the spectral space and a non-periodicity are necessary for a visually preferable dot pattern.

Based on this result, a method of realizing blue noise characteristics by a mask method having a high processing speed has been considered. The blue noise characteristic of the blue noise pattern means that an output pattern of dots when set to an arbitrary gradation is locally aperiodic and isotropic, and has a low low frequency component. In addition, the distribution of dots at an arbitrary gradation has no predetermined fixed distribution at all, and is non-deterministic in the sense that it is left to the randomness of the algorithm. Therefore, it can be said that the distribution of the dots having the blue noise characteristic is random, non-deterministic, and has the non-white noise characteristic.

Therefore, in order to produce a stochastic screen for realizing a blue noise pattern by a mask method, the frequency characteristics of the generated dot pattern are such that the power spectrum is small in the low frequency region, while the power spectrum is small in the high frequency region. In, the dither threshold is designed so that the power spectrum is concentrated.

[0011]

However, the conventional stochastic screen having a blue noise characteristic focuses only on concentrating a power spectrum in a high frequency region, and for example, a frequency by an output device such as a printer or a monitor. It did not take into account differences in response or spatial frequency response of human vision such as VTF. That is, for example, by adjusting the average gradation of the dot profile generated by the stochastic screen having the blue noise characteristic between the output data and the output device to the output device, the average is expressed in the linear graph shown in FIG. The contrast adjustment is only performed at a certain value, and it is not created in consideration of the spatial frequency characteristics of the output device.

As a result, for example, when the output device has a different frequency response depending on the direction of the XY space, such as when the sensitivity is high in the X direction and low in the Y direction in the XY space, However, when the frequency response is unstable in a high frequency region, the created stochastic screen may not be optimal for obtaining an appropriate and good output image.

That is, for example, an output pattern in which ON / OFF is repeated for each pixel corresponds to the highest frequency as data to be input to the output device, and when the frequency response in this frequency region becomes unstable. As shown in FIG. 7, the ON dots may be thick and crush the OFF dots as shown in FIG. 7, or the ON dots may not be transferred and may be lost to become OFF dots. In such a case, the frequency shifts to a frequency region much lower than the originally intended frequency, so that the granularity of the output image deteriorates and the output image of good quality cannot be obtained. Had.

The present invention has been made in view of these points, and has a case in which the frequency response is different depending on the direction of the XY space of the output device or the frequency response of the output device is unstable in a high frequency region. It is an object of the present invention to provide an image processing method using a stochastic screen that can obtain an output image with good image quality even when the image processing is performed.

[0015]

In order to achieve the above object, an image processing method according to the present invention expresses an intermediate gradation by performing dither processing using a stochastic screen on each pixel of an input image. An image processing method for obtaining an output image, wherein a stochastic screen is created in consideration of two frequency characteristics of a frequency characteristic of an output device and a blue noise characteristic.

According to the image processing method of the present invention,
When the frequency response is different depending on the direction of the XY space, or when the output device has an unstable frequency response in a high frequency region, it is preferable that the frequency response does not include a low or unstable frequency region. Output images can be created.

Furthermore, another image processing method according to the present invention is characterized in that the stochastic screen is anisotropic.

According to another image processing method according to the present invention, an image processing method can be performed using a stochastic screen filter generated so that an output pattern is anisotropic.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIGS.

In this embodiment, DP (x, y)
Represents a dot profile at coordinates (x, y), c
(X, y) is a cumulative matrix at coordinates (x, y), m
(X, y) represents the stochastic screen at coordinates (x, y), xsize represents the X size of the stochastic screen, ysize represents the Y size of the stochastic screen, and B represents the number of quantization bits. Shall be shown. Further, the gradation of the image processed in the present embodiment is set to 255 gradations, and noG has a value range of [0, 2 ^B −
1], and g represents the tone when noG is normalized to the range [0, 1].

When the image processing method according to the present embodiment is started, first, initial settings are performed (ST1).

In the initial setting, the dot profile and the accumulation matrix shown in FIG. 1 are initialized.

The initialization of the dot profile and the accumulation matrix is performed as shown in FIG.

More specifically, when noG = floor (0.5 · (2 ^B −1)); g = noG / (2 ^B −1); random numbers with a uniform distribution are generated (ST11). .
Here, rand (x, y) is a uniformly distributed random number in the range [0, 1] at the coordinates (x, y).

Next, filtering is performed (ST1).
2).

The filtering is as follows: FourierRand (u, v) = fft2 (ran
d (x, y)); FilteredFourierRand (u, v) =
Filter (u, v) ・ FourierRand
(U, v); FilteredRand (x, y) = ifft2 (F
ilteredFourierRand (u, v));

Next, binarization is performed (ST13).

In this binarization, the ideal number of black pixels is calculated as follows: floor ((xsize · ysize) · (1-no)
G / (2 ^B -1))), the filter is set to have a black point having the ideal number of black pixels.
eredRand (x, y) is binarized.

In the above-mentioned binarization, Filter
The redRand is ranked in descending order of the value, pixels are selected from the smaller values by the ideal number of black pixels, and the selected pixel is set to 0, and the others are set to 1.

Further, in the dot profile, 1 is set as an initial value, and in the temporary update of the dot profile, black pixels are replaced with white pixels by the number of deleted black pixels at random from the dot profile.
(X, y).

Here, Filter (u, v), Fil
ter2 (u, v) and Filter3 (u, v) are F
represents a filter on ourier space.
Filter3 (u, v) is a filter that emphasizes the low frequency region in the output characteristics of the output device.

Next, a difference matrix is obtained (ST14).

This difference matrix is: NormalDp (x, y) = WorkDp (x, y)
-G, NormalFourierDp (u, v) = fft2
(NormalDp (x, y)); when FilteredNormalFourierDp
(U, v) = Filter (u, v) · NormalF
ourierDp (u, v) FilteredNormalFourierDp2
(U, v) = Filter2 (u, v) · Normal
FourierDp (u, v) FilteredNormalFourierDp3
(U, v) = Filter3 (u, v) · Normal
FourierDp (u, v).

Also, FilteredNormalDp (u, v) = iff
t2 (FilteredNormalFourierD
p (u, v)); FilteredNormalDp2 (u, v) = if
ft2 (FilteredNormalFourier
Dp2 (u, v)); FilteredNormalDp3 (u, v) = if
ft2 (FilteredNormalFourier
Dp3 (u, v));

Then, NormalErrorDp (x, y) = Normal
Dp (x, y) -FilteredNormalDp
(X, y) NormalErrorDp2 (x, y) = Norma
IDp (x, y) -FilteredNormalDp
2 (x, y) NormalErrorDp3 (x, y) = Norma
IDp (x, y) -FilteredNormalDp
3 (x, y).

Next, the dot profile evaluation value is initialized (ST15).

The initialization of the dot profile evaluation value is as follows: ReverseFilter (u, v) = 1.0-Fi
lter (u, v), NormalFourierDp (u, v) = fft2
If (NormalDp (x, y)), ReverseFilteredNormalFour
ierDp (u, v) = ReverseFilter
(U, v) · NormalFourierDp (u, v)
v)

Therefore, minMSE = Σ (Σ (abs (ReverseFile)
terminatedNormalFourierDp (u,
v)))); BestDp (x, y) = WorkDp (x, y);

Next, a replacement order is calculated (ST16).

In this embodiment, the filter
The replacement order is calculated by focusing on the difference from the result filtered by (u, v). At this time, sortWhit
eInd is a table in which pixel coordinate values (x, y) are stored so as to give NormalErrorDp values in descending order, and sortBlackInd is Normal
5 is a table in which pixel coordinate values (x, y) are stored so as to give lErrorDp values in descending order. At this time, the calculation of the replacement order is as follows: sortWhiteInd = sort (NormalE
rrDp (x, y)). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 1 & NormalErrorDp (x, y)> 0
Are satisfied. Also, sortBlackInd = sort (NormalE
rrDp (x, y)). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 0 & NormalErrorDp (x, y) <0
Are satisfied.

On the other hand, in another embodiment, the Filter 3 which places importance on the low frequency region among the output characteristics of the device is used.
The replacement order is calculated by focusing on the difference from the result filtered by (u, v). The calculation of the replacement order in this case is as follows: sortWhiteInd3 = sort (Normal
ErrorDp3 (x, y)). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 1 & NormalErrorDp3 (x, y)>
It is assumed that 0 is satisfied. Also, sortBlackInd3 = sort (Normal
ErrorDp3 (x, y)). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 0 & NormalErrorDp3 (x, y) <
It is assumed that 0 is satisfied.

Next, a dot profile that minimizes the dot profile evaluation value is obtained (ST17).

For the number of substitutions, the following processing is repeated from the minimum number of substitutions to the maximum number of substitutions. At this time, the minimum replacement number is set to one. In the method of focusing on the difference from the result of filtering with Filter (u, v) in the calculation of the replacement order, the maximum replacement number is sortedWhite.
The table size of Ind and the table size of sortBlackInd are smaller values. On the other hand, in the method of focusing on the difference from the one filtered by Filter3 (u, v) in the calculation of the replacement order, the sortW
hitInd3 table size and sortBlac
The value of the table size of kInd3 is the smaller value.

In obtaining a dot profile that minimizes the dot profile evaluation value in ST17, first, pixels are selected and replaced according to the replacement order shown in FIG. 3 (ST21).

With reference to the sortWhiteInd3, the replacement number of pixels is selected. Then, 1 of the selected pixel is replaced with 0. Also, referring to the sortBlackInd3, the pixel of the replacement number is selected, and 0 of the selected pixel is set to 1
Replace with Then, the replacement result is represented by WorkDp2 (x,
y).

Next, an evaluation value of the replaced dot profile is calculated (ST22).

The evaluation value of the replaced dot profile is calculated as follows: NormalDp2 (x, y) = WorkDp2 (x,
y) -g; NormalFourierDp2 (u, v) = fft
2 (NormalDp2 (x, y)); ReverseFilteredNormalFour
ierDp2 (u, v) = ReverseFilter
(U, v) · NormalFourierDp2 (u, v)
v) MSE = @ abs (ReverseFiltered
NormalFourierDp2 (u, v);

Then, one having a small dot profile evaluation value is selected (ST23).

When the dot profile evaluation value is small, when MSE <minMSE, minMSE = MSE; BestDp (x, y) = WorkDp2 (x, y);

Thus, a dot profile that minimizes the dot profile evaluation value is obtained.

Finally, the dot profile and the accumulation matrix are updated (ST18).

In the present embodiment, noG = flow
r (0.5 · (2 ^B −1)).

In this case, the update of the dot profile and the accumulation matrix is as follows: Dp (x, y) = bestDp (x, y), and C (x, y) is as follows: Dp (x, y) = 1 C (x, y) = noG-
1: When Dp (x, y) = 0, C (x, y) = noG;

This completes the initial setting of the dot profile and the accumulation matrix.

Next, the following processing ST2 to S shown in FIG.
T9 is calculated from the gradation noG = startNoG + 1 (2 ^B
-1) Perform until -1.

First, the parameters are updated (ST2).

At this time, the normalized gradation g = noG / (2 ^B
-1) ;, the ideal number of black pixels is floor ((x
size · ysize) · (1- noG / (2 B -
1))), and the number of deleted black pixels is obtained by subtracting the ideal number of black pixels from the current number of black pixels.

Next, a filter is created (ST3).

Here, fg represents a cutoff frequency.

First, the filter setting processing section 31 shown in FIG.
The stochastic screen stop area setting processing unit 32 performs stochastic screen stop area setting processing. The stochastic screen stop area setting processing unit 32 is a stop area in a known blue noise mask. When the cut-off frequency is g ≦ ｆ, fg = floor ((K · √g) ·
xsize) When g> 1/2, fg = floor ((K · √ (1-
g)). xsize).

At this time, K is Scaling Fact.
or, in the present embodiment, K = 1 / √2. Floor () represents a function for rounding to the nearest positive number in the negative direction.

On the other hand, the output device rejection zone setting processing section 33 performs output device rejection zone setting processing for setting the frequency characteristics of the output device.

In the present embodiment, fgMax = 0.
25 × xsize, and the highest frequency in the X direction is の of the highest frequency in the Y direction.

The output device rejection zone setting processor 33 sets 0 when u ≧ fgMax to u ≦ −fgMax.
When u is a value other than the above, processing is performed so that it becomes 1.

Then, in the stop band adjustment processing section 34,
Adjust the stochastic screen rejection area while considering the output device rejection area, and create a filter. here,
The filters are set to three types.

Filter (u, v) is defined as u ≧ fgMa
When x or u ≦ −fgMax, it becomes 0, and √ ((2.
0 · u) ² + v ² ) ≦ 0 when fg, and 1 when the value is not the above value.

Filter2 (u, v) is defined as u ≧ fgM
The value is set to be 0 when ax or u ≦ −fgMax, and to be 1 when the value is not the above value.

Filter3 (u, v) is √ ((2.
0 · u) ² + v ² ) ≦ 0 when fg, and 1 when the value is other than the above value.

Next, the dot profile is provisionally updated (ST4).

In this temporary update of the dot profile, black pixels are replaced with white pixels by the number of black pixels to be deleted at random from the dot profile, and are set to WorkDp (x, y).

Next, a difference matrix is obtained (ST5).

This difference matrix is: NormalDp (x, y) = WorkDp (x, y)
-G, NormalFourierDp (u, v) = fft2
(NormalDp (x, y)); when FilteredNormalFourierDp
(U, v) = Filter (u, v) · NormalF
ourierDp (u, v) FilteredNormalFourierDp2
(U, v) = Filter2 (u, v) · Normal
FourierDp (u, v) FilteredNormalFourierDp3
(U, v) = Filter3 (u, v) · Normal
FourierDp (u, v).

Also, FilteredNormalDp (u, v) = iff
t2 (FilteredNormalFourierD
p (u, v)); FilteredNormalDp2 (u, v) = if
ft2 (FilteredNormalFourier
Dp2 (u, v)); FilteredNormalDp3 (u, v) = if
ft2 (FilteredNormalFourier
Dp3 (u, v));

Then, NormalErrorDp (x, y) = Normal
Dp (x, y) -FilteredNormalDp
(X, y) NormalErrorDp2 (x, y) = Norma
IDp (x, y) -FilteredNormalDp
2 (x, y) NormalErrorDp3 (x, y) = Norma
IDp (x, y) -FilteredNormalDp
3 (x, y).

Next, the dot profile evaluation value is initialized (ST6).

The initialization of the dot profile evaluation value is as follows: ReverseFilter (u, v) = 1.0−Fi
lter (u, v), NormalFourierDp (u, v) = fft2
If (NormalDp (x, y)), ReverseFilteredNormalFour
ierDp (u, v) = ReverseFilter
(U, v) · NormalFourierDp (u, v)
v)

Therefore, minMSE = Σ (Σ (abs (ReverseFile)
terminatedNormalFourierDp (u,
v)))); BestDp (x, y) = WorkDp (x, y);

Next, a replacement order is calculated (ST7).

The calculation of the replacement order in the present embodiment focuses on the difference from the one filtered by Filter (u, v), and sortWhiteInd = sort (NormalE
rrDp (x, y). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 1 & Dp (x, y) = 0 & NormalEro
It is assumed that rDp (x, y)> 0. Also, sortBlackInd = sort (NormalE
rrDp (x, y). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 0 & NormalErrorDp (x, y) <0
Are satisfied.

In another embodiment, the replacement order is calculated by focusing on the difference between the output characteristic of the output device and the output characteristic filtered by Filter 3 (u, v), which emphasizes the low frequency region, and sortWhiteInd3 = sort ( Normal
ErrorDp3 (x, y). Here, the coordinates (x, y) are represented by WorkDp (x, y).
y) = 1 & Dp (x, y) = 0 & NormalEro
It is assumed that rDp3 (x, y)> 0. Also, sortBlackInd3 = sort (Normal
ErrorDp3 (x, y). However, coordinates (x, y) WorkDp (x, y)
= 0 & NormalErrorDp3 (x, y) <0

Next, a dot profile that minimizes the dot profile evaluation value is obtained (ST8).

For the number of substitutions, the following processing is repeated from the minimum number of substitutions to the maximum number of substitutions. At this time, the minimum replacement number is set to one. In the method of focusing on the difference from the result of filtering with Filter (u, v) in the calculation of the replacement order, the maximum replacement number is sortedWhite.
The table size of Ind and the table size of sortBlackInd are smaller values. On the other hand, in the method of focusing on the difference from the one filtered by Filter3 (u, v) in the calculation of the replacement order, the sortW
hitInd3 table size and sortBlac
The value of the table size of kInd3 is the smaller value.

In determining a dot profile that minimizes the dot profile evaluation value, a dot profile that minimizes the evaluation value of the dot profile and the dot profile evaluated in the initialization of the accumulation matrix is determined (ST17). Similarly, a dot profile that minimizes the dot profile evaluation value is obtained.

Next, the dot profile and the accumulation matrix are updated (ST9).

The dot profile and the accumulation matrix are updated as follows: Dp (x, y) = BestDp (x, y) C (x, y) = C (x, y) + (1−Dp (x, y)
y)).

Then, updating of the parameters (ST2)
Update of the dot profile and the accumulation matrix from (S
T9) is calculated from the gradation noG = startNoG + 1 (2
Perform ^B- 1) -1.

Then, the dot profile and the accumulation matrix are initialized (ST10).

The dot profile is set by inverting the gradation of the initially set (ST1) dot profile, that is, by inverting 0 and 1. In addition, the accumulation matrix inherits the above final result as it is.

Then, updating of the parameters (ST2)
Update of the dot profile and the accumulation matrix from (S
The processing up to T9) is performed with gradation noG = (255-star)
(tNoG) +1 to (2 ^B -1). However, when updating the dot profile and the accumulation matrix, Dp (x, y) = BestDp (x, y) C (x, y) = C (x, y)-(1-Dp (x, y) ).

In the present embodiment, when a filter is created (ST3), the filter setting processing section 31 adjusts the stop zone of the stochastic screen in consideration of the stop zone of the output device to create a stochastic screen. I do.

Therefore, if the output data and the output device have different frequency responses, or if the frequency response of the output device in the high frequency region is unstable, the stochastic device considering the spatial frequency of the output device is used. Since recording is performed using a stick screen filter, the output result is as shown in FIG. 5, where the ON dots are thickened and the OFF dots are crushed, or the ON dots are not transferred and are lost as OFF dots. It is possible to obtain an output image with good image quality and high repetition reproducibility without causing deterioration and without deteriorating the granularity of the image. This is shown in FIG.
It can be seen that there is a remarkable difference as compared with the conventional example shown in FIG.

In this embodiment, a stochastic screen filter is used in consideration of the spatial frequency characteristics of the output device in the real space. However, the present invention is not limited to this. Characteristics may be considered.

Further, the present invention can be variously modified as needed, which is limited to the above embodiment.

[0094]

As described above, according to the image processing method of the present invention, an image is formed using a stochastic screen created in consideration of the two frequency characteristics of an output device and the frequency characteristic of blue noise. Perform processing method.
For this reason, even when the frequency response differs depending on the direction of the XY space or when the output device has an unstable frequency response in a high frequency region, an output image with good image quality can be obtained. It has the effect that can be done.

According to another image processing method according to the present invention, an image processing method can be performed using a stochastic screen filter generated such that an output pattern is anisotropic. For this reason, even when the frequency response differs depending on the direction of the XY space, or when the output device has an unstable frequency response in a high frequency region, it is possible to flexibly cope with the problem and obtain better image quality. This has the effect that an output image of can be obtained.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating an image processing method according to the present invention.

FIG. 2 is a flowchart for explaining initialization of a dot profile and an accumulation matrix in the image processing method according to the present invention shown in FIG. 1;

FIG. 3 is a flowchart for explaining a case where a dot profile that minimizes the evaluation value of the dot profile in the initialization of the dot profile shown in FIG. 2 and the accumulation matrix is obtained;

FIG. 4 is a conceptual diagram showing a filter setting processing unit used when creating a filter in the image processing method according to the present invention.

FIG. 5 is an enlarged view showing a dot profile created by the image processing method according to the present invention and a printing result thereof;

FIG. 6 is a graph showing contrast adjustment values in a conventional image processing method.

FIG. 7 is an enlarged view showing a dot profile created by a conventional image processing method and its printing result.

[Explanation of symbols]

 31 filter setting processing unit 32 stochastic screen stop zone setting processing unit 33 output device stop zone setting processing unit 34 stop zone adjustment processing unit

Claims (2)

[Claims] 1. An image processing method for obtaining an output image expressing an intermediate gradation by performing dither processing using a stochastic screen on each pixel of an input image, comprising: a frequency characteristic of an output device and a blue noise. Characteristic 2
An image processing method characterized in that a stochastic screen is created in consideration of two frequency characteristics. 2. The image processing method according to claim 1, wherein the stochastic screen is anisotropic.