JP6541215B2 - Image processing apparatus and method - Google Patents

Description

The present invention relates to a method of generating a halftone screen for high-quality image output without moiré, and an image processing apparatus thereof, in electrophotographic-based digital printing such as copiers, MFPs (Multi-Functional Printers), office printers, etc. It is a thing.

A recording apparatus based on electrophotography is characterized by being non-linear to the fluctuation of the spatial frequency of the recorded image due to the non-linear characteristic of the electrophotographic process. This is because the surface of the photosensitive drum surface is electrostatically uniformly charged, and the exposure process for removing the charge of the photosensitive drum by light beam scanning with a finite spread such as a laser spot, etc. It is caused by the presence of sexual characteristics and the complexity of the electrophotographic process such as development, transfer and fixing. For example, it is difficult to print small one-dot printing, and printing is only performed in a cluster state of several dots. Also, the dots that are slightly apart from one another are stuck or separated by the movement of the toner depending on the size of the intervals. On the other hand, in the process in which ink droplets adhere to paper and recorded like ink jet recording, although there is a micro phenomenon between the ink and the media (paper), the interference phenomenon between the printing dots is relatively difficult to occur, The ejected ink is reliably recorded on the paper.

Because of this phenomenon, halftone dot-based halftoning is the main way to reproduce tone in image formation based on electrophotographic recording. That is, in the dot representation, since the basic spatial frequency is fixed, the dot shape is not affected by the variation of the spatial frequency, and a stable output can be obtained. Assuming that the dot number of the dot is N lines / inch, the pitch P is 25.4 / N mm, and the basic spatial frequency is 1 / (2P) = N / (2 × 25.4), and the constant basic spatial frequency is maintained. In an electrophotographic recording apparatus, stable image reproduction can be obtained if it is designed to always stabilize the spatial frequency. For example, in order to form a halftone screen of 200 lines / inch with a printing apparatus of 1200 dpi, the spatial frequency response characteristic of 4 cycles / mm becomes the fundamental frequency, and by stabilizing this frequency, the response characteristic of the entire apparatus is improved.

On the other hand, while the AM modulation system by the dot system can obtain stable output, in color printing, a moiré phenomenon is likely to occur due to color overlapping of C, M, Y, K color toners. In order to avoid this, in general, in the printing technology, the screen angle is changed for each color, and the frequency of the moire beat generated between the colors is set to a high frequency side so as not to be a problem visually. For example, screen angles of 0 ° and 60 ° are given to halftone dots of each color centered on 30 °, thereby suppressing the occurrence of moiré due to color overlap. By thus making the screen angles different for each color, it is possible to reduce the occurrence of the moiré.

However, no matter how much screen angle is introduced and optimization is performed to suppress moiré, only by shifting the beat frequency to be generated to the high frequency side, a unique pattern caused by the overlap between colors remains. This is the so-called Rosetta pattern, which is an obstacle to obtaining high quality output. In particular, when aiming at high-quality photo output, the target image quality requires smooth image quality reproduction like photographs in a silver salt process. The Rosetta pattern is a big problem in this case.

Therefore, as an approach to avoid moiré and rosette patterns, there is a method of tone reproduction using an FM halftone screen by an error diffusion method or a blue noise mask method.

The Blue Noise Mask method, also referred to as the Stochastic Screen, forms a disperse-type FM screen, and has low dispersivity reduced using a relatively large dither threshold matrix. It exhibits good high frequency response characteristics. For this reason, it is widely used in halftone screens in digital printing, or printers of the ink jet system and the thermal transfer system.

However, in these disperse type FM screens, the spatial frequency distribution of the printing dots extends to the high frequency side, and the dot spacing of the printing dots changes according to the gradation level. For example, as the image density value increases, the distance between the print dots gradually decreases. For example, at an intermediate density, it looks like a random checkerboard pattern, and the spatial frequency characteristic is shifted to a high spatial frequency and is susceptible to the frequency characteristic of the printing apparatus. Since the electrophotographic recording method is susceptible to spatial frequency fluctuation as described above, such an FM modulation method can not be used at present.

Therefore, various methods have been proposed as a cluster-type FM screen in which spatial frequency characteristics are shifted to the low frequency side. This method can be matched to the MTF characteristics of the printer engine, and can output stable high-definition halftone images.

As generation of such a cluster type FM screen, an error diffusion method by the output signal feedback, a VACS method, etc. are proposed. Although these can obtain high-quality FM screens, they require special-purpose hardware for high-speed processing with a large computational load.

There is also proposed a method of determining a dither threshold by forming a Voronoi polygon and creating a dither matrix. Although this method can provide well-distributed and high quality images, it is not easy to control cluster size.

On the other hand, a method using a dither mask called Green-noise mask has been reported. In this method, as in the blue noise mask method, the dither mask is created by the filter operation in the frequency space. The method of making a mask is simpler and easier than the former method, but because it is sensitive to the filter shape, a mask showing high quality output has not been obtained. In particular, the anisotropy and the roughness of the dots in the light and dark areas are noticeable, so that high quality image output can not be performed and it is not suitable for high-quality printing.

"Digital halftoning technique using a blue noise mask", J. Opt. Soc. Am. A / Vol. 9, No. 11, pp. 1920-1929 (1992). P. Li and J. P. Allebach: "Clustered-minority-pixel error diffusion", J. Opt. Soc. Am. A / Vol. 21, No. 7, pp. 1148-1160 (2004). Kawamoto Naoto: "Development of Variable Cluster Halftoning Technology", Journal of the Institute of Image Electronics, Vol.38, No.5, pp.735-745 (2009) Asai Hiroshi; "Non-periodically clustered color halftone screen", Journal of Image Electronics Society, Vol. 35, No. 5, pp. 566-575 (2006) D. Lau, G. R. Arce and N. C. Gallagher; "Digital Halftoning by means of green-noise masks", J. Opt. Soc. Am, Vol. 16, No. 7 / July, pp. 1575-1586 (1999).

Therefore, in order to perform stable output with an electrophotographic printer or the like, it is an issue to obtain high-quality halftoning in accordance with the response characteristics of the printer engine by a high-speed, simple method with a small calculation load. .

The present invention solves this problem and enables stable output at high speed with an electrophotographic printer or the like. That is, the frequency characteristic of the recording dither pattern can be matched to the spatial frequency optimum for the printer engine so as to match the frequency which can be recorded most stably in the electrophotographic method.

In addition, since the present invention binarizes in the dither system, the arithmetic processing is very simple and high-speed output is possible. The threshold matrix of the dither exhibits so-called green noise characteristics in which the frequency characteristics have band characteristics lowered on the high frequency side and the low frequency side in order to match the recording frequency optimum for the printer engine. As a result, it is possible to output a high quality and stable halftone image at high speed at a frequency that is the best at electrophotographic printers.

To this end, the invention comprises means for determining a dither matrix consisting of the following two steps:
1. First, the step of filtering by limiting the band so that the dot profile in the intermediate density state becomes an optimal spatial frequency for the printer engine 2. Next, the band-limited filtering is performed over the entire tone range The process of obtaining the dot profile and the process of obtaining two or more steps determine the threshold matrix of the dither.

More specifically,
1. Band pass filter D (u, v) where the blackening ratio g is a point profile with an intermediate value (g = 0.5) and the radial frequency f_r is 0 in the low frequency range and high frequency range, and has a finite value in the middle , 1/2) (where (u, v) is a two-dimensional spatial frequency), the first step of filtering
2. A second step of performing filtering with the band pass filter D (u, v, g) in which the blackening ratio g changed in gradation units changes the radial frequencies f_min and f_max according to g,
The threshold matrix of dither is determined by the two steps of

In the first step, g = 1/2 is used as the initial value, R ^ 2/2 black points are randomly given, and point profile p (i, j, 1/2) is given (initial state is white noise)
(STEP 1) Obtain a two-dimensional FFT of the point profile and obtain P (u, v, 1/2).
(STEP 2) A filter D (u, v, 1/2) is multiplied by P (u, v, 1/2) to obtain a new P '(u, v, 1/2). Here, a green noise filter is used in which D (u, v, 1/2) has a value only in the region where the radial frequency f_r is f_min ≦ f_r ≦ f_max.
(Step 3) Inverse Fourier transform is performed on P ′ (u, v, 1/2) to obtain a multi-valued point profile p ′ (i, j, 1/2).
(STEP 4) Error function: Calculate e (i, j, 1/2) = p '(i, j, 1/2)-p (i, j, 1/2) and calculate the error at each pixel position Arrange in descending order of positive and negative, and reverse white → black and black → white in descending order. At this time, the numbers to be reversed are equal.
(STEP 5) Repeat STEP 1 to STEP 5 until the error is within a certain tolerance. If all the pixels fall within the tolerance, a point profile of g = 1/2 is finally obtained.

Subsequently, in the second step, at g = g ± Δg changed by one gradation unit,
(STEP 1) A two-dimensional FFT of p (i, j, g) is performed to obtain P (u, v, g).
(STEP 2) The filter D (u, v, g) is multiplied by P (u, v, g) to obtain a new P ′ (u, v, g). Here, D (u, v, g) is a filter in which f_min and f_max change with g.
(Step 3) Inverse Fourier transform is performed on P ′ (u, v, g) to obtain a multi-valued point profile p ′ (i, j, g).
(STEP 4) Error function: Calculate e (i, j, g) = p '(i, j, g)-p (i, j, g), arrange errors at each pixel position in the order of large positive and negative, A number N = R ^ 2/2 ^ n corresponding to the gradation value is reversed in descending order from white to black (when g → 1) or as black → white (when g → 0). At this time, the gradation value is written to the inverted pixel position of the dither matrix.
(STEP 5) Repeat STEP 1 to STEP 5 as g = g ± Δg. When g = 0, 1, this operation ends.
From the above operations, the dither threshold matrix is calculated.

According to the present invention, it is possible to determine the threshold value of the dither matrix showing the green noise characteristic in such a process, and to obtain a high-speed, stable and high-quality image for an electrophotographic printer.

Diagram showing a system of an image processing apparatus Block diagram for explaining image signal processing A diagram representing a recording device by laser beam scanning Diagram showing the processing flow of the first step Diagram showing the processing flow of the second step Diagram showing the output of this green noise mask method at g = 1/2 Grayscale output diagram when f_min and f_max with g are both fixed values (a), Grayscale output diagram when (b) only f_min is variable Diagram showing the states of f_min and f_max in FIG. 7 (b) Diagram showing the states of f_min and f_max in the present invention (a, b) = (0, -1/4) showing the obtained dither matrix; (a) a 128 x 128 dither pattern represented by a tone value; (b) 64 x 64 dither pattern (C) A diagram of the 32x32 dither pattern represented by a tone value The figure which connected the pattern of Fig.10 (b) four pieces of 2x2 Diagram showing the dither output and spectrum by this green noise method at each gradation (A) Original image, (b) Output image with blue noise (BN), (c) (a, b) = (1/8, -1/8) Output image at the time, (d) (a, b) = (0, -1/4) output image (A) Original image, (b) Output image with blue noise (BN), (c) (a, b) = (1/8, -1/8). Output image at time, (d) (a, b) = (0, -1/4) output image Image when combined with the output image of each color plate (A) C version: (1/4, 0) output image, (b) M version: (0, -1/8) output image, (c) ) Y version: (-1/8, -1/4) output image, (d) composite image output image (A) C version: output image by blue noise method, (b) M version: output image of (1/4, 0), (c) Y version: ((a) C version) Output image of 0, -1/8). Diagram showing dot pattern and spectrum by elliptic filter A diagram representing the output image and spectrum of various g dithers by an elliptic filter A diagram showing a method of data hiding, (a) A diagram showing a dither matrix (a) as a tone image, (b) a diagram showing a dither matrix (b) as a tone image, (c) a dither matrix Lena image output by repeating (a) and (b) mutually, (d) A figure showing the spectral characteristics.

Hereinafter, the present invention will be described with reference to the drawings. FIG. 1 shows the apparatus configuration of the present invention. The functions of a copier and an MFP (Multi-Functional Printer) including a scanner 2 and a printer 2 based on electrophotography are controlled by a built-in controller 3. The controller 3 has a ROM 4, a RAM 5, a program memory and an HDD 6 for storing image data, a display 8, a keyboard 9, and a communication function 10. A CPU 7 is responsible for overall control and software arithmetic processing of image processing.

FIG. 2 is a block diagram of a dot generator for generating a clustered dither halftone screen according to the present invention. The horizontal and vertical synchronization signals from the printer 2 are composed of H-Sync and V-Sync signals of an output raster image and a pixel clock. The image data from the image memory 11 stored in the RAM 5 or the HDD 6 is subjected to halftone processing in the image processing unit 12 and passes through a laser driver 13 to obtain a beam-modulated output signal 16 from the semiconductor laser 14. Although this modulated light beam is not shown, it is guided to an electrophotographic-based recording apparatus, and an image is outputted through an electrophotographic process of light exposure → development → transfer → fixing.

Halftone processing in the image processing unit 12 is performed by binarizing with a normal dither method, and the threshold value of the input image signal and the dither is compared, and if the image signal is larger than the threshold, it is set to “1”. Turn on / off the laser. In order to improve the image quality, there are many cases where multi-value output of three or four values is performed, but this has multiple thresholds for one pixel, and multi-value output is determined by comparison with these thresholds. . In any case, the comparison with the threshold value is performed, so in the present invention, explanation will be made with two values.

FIG. 3 shows a configuration diagram of an optical system of an image recording apparatus equipped with the present invention. A light beam emitted from a light source 17 by a semiconductor laser or the like is incident on the reflecting surface of the rotating polygon mirror 21 by a collimator lens 18 by a spherical system or an anamorphic optical system. Due to the rotation of the rotary polygon mirror, the reflected beam is deflected so as to be incident on an imaging lens 19 such as an fθ lens. The light spot imaged by the imaging lens 19 scans a light beam on the photosensitive drum 20.

Next, an algorithm for creating a dither mask showing green noise characteristics will be described.
A dot exhibiting green noise characteristics is filtered so that its spatial frequency spectrum is limited between f_min and f_max. Assuming that the dither matrix size to be obtained is R × R (R = 2 ^ m), its Nyquist frequency is f_N = R / 2. In the operation, when the blackening ratio of the printing dots (area ratio occupied by black dots in unit area) is g (0 ≦ g ≦ 1), g = 1 represents all black, and g = 0 represents all white. When the pixel data is n bits, the gradation level L is
L = 2 ^ n · (1-g)
It becomes.

In the step of obtaining a dither matrix, first, a first step of obtaining an intermediate (g = 1/2) point profile and a profile of the entire area of g = g ± Δg successively changed by one gradation unit are obtained; It comprises the second step of calculating the dither threshold matrix.

The first step will be described along the flow of FIG. Now, it is assumed that the dither matrix size is R × R, a black dot is to be struck in this R × R area, and the blackening ratio g, and the point profile at point (i, j) be p (i, j, g). First, an intermediate (g = 1/2) point profile is performed in the following steps.
The initial value is g = 1/2, R ^ 2/2 black points are randomly given, and a point profile p (i, j, 1/2) is created. (^ Represents a power. The initial state is white noise).
(STEP 1) Obtain a two-dimensional FFT of the point profile and obtain P (u, v, 1/2).
(STEP 2) A filter D (u, v, 1/2) is multiplied by P (u, v, 1/2) to obtain a new P '(u, v, 1/2). Here, a green noise filter is used in which D (u, v, 1/2) has a value only in the region where the radial frequency f_r is f_min ≦ f_r ≦ f_max.
(Step 3) Inverse Fourier transform is performed on P ′ (u, v, 1/2) to obtain a multi-valued point profile p ′ (i, j, 1/2).
(STEP 4) An error function: e (i, j, 1/2) = p '(i, j, 1/2)-p (i, j, 1/2) is obtained for each pixel. The white pixels and the black pixels are arranged in descending order of error, and inverted in the order of large to white → black and black → white. At this time, the numbers to be reversed are equal.
(STEP 5) Repeat STEP 1 to STEP 5 until the error is within a certain tolerance. If all the pixels fall within the tolerance, a point profile of g = 1/2 is finally obtained.

Subsequently, the point profile creation at g = g ± Δg will be described with reference to FIG. First, p (i, j, 1/2) of g = 0.5 obtained in the first step is set as an initial value, and is created in the following steps.
(STEP 1) A two-dimensional FFT of p (i, j, g) is performed as g = g ± Δg to obtain P (u, v, g).
(STEP 2) The filter D (u, v, g) is multiplied by P (u, v, g) to obtain a new P ′ (u, v, g). Here, D (u, v, g) is a filter in which f_min and f_max change with g.
(Step 3) Inverse Fourier transform is performed on P ′ (u, v, g) to obtain a multi-valued point profile p ′ (i, j, g).
(STEP 4) An error function: e (i, j, g) = p ′ (i, j, g) −p (i, j, g) is obtained for each pixel. Next, arrange the errors in descending order of white pixel and black pixel, and reverse white pixels to black pixels by N = R ^ 2/2 ^ n pieces in descending order (when g → 1), or black pixels are white pixels Invert to (when g → 0). At this time, the gradation value is written to the inverted pixel position of the dither matrix.
(STEP 5) Repeat STEP 1 to STEP 5 as g = g ± Δg, and when g = 0, 1, complete this operation.

Here, the radial frequencies f_max and f_min are set as follows.
When f_max and f_min of the filter D (u, v, g) are fixed values regardless of g, the uniformity of the point profile in the bright and dark areas is broken. For this reason, f_max and f_min are made variable by g, and f_max and f_min are set so as to approach the blue noise characteristic from the green noise characteristic in the bright and dark portions. By doing this, the roughness of the light and dark parts becomes somewhat fine and the graininess becomes inconspicuous.

Find the dither matrix along the above steps.
The frequency with the average dot spacing at the blackening rate g is
When f_0 = sqrt (g) · f_N: 0 ≦ g ≦ 1/2
f_0 = sqrt (1-g) · f_N: When 1/2 ≦ g ≦ 1
Given by Here, sqrt () represents a square root. When the filter D (u, v, g) is circular, as a band filter in the radial direction,
D (f_r, g) = finite value: when f_min ≦ f_r ≦ f_max
D (f_r, g) = 0: Suppose that it is another time. Because it is a finite value, it takes a decimal value between 0 and 1. f_max and f_min are the differences based on f_0,
f_max = f_max-f_0≡a · f_N
f_min = f_min-f_0 ≡ b · f_N
Express as In general, since the Nyquist frequency f_N differs depending on the size R of the matrix, by using values (a, b) normalized by f_N
(a, b) = (f_max / f_N, f_min / f_N)
By describing (a, b) as a parameter, it is possible to limit the cluster size regardless of the matrix size R.

Here, as a specific example, the case of n = 8 bits / pixel with a dither threshold matrix of R = 128 will be described. In this case, Δg = 1/256 and f_N = 64.
With the point generated with 8192 random numbers as a point profile of g = 0.5 as an initial value, the spectral characteristics of D (u, v, 1/2) are taken as parameters (a, b),
D (f_r, g) = 1: When f_0 + bfn ≦ fr ≦ f_0 + af n
D (f_r, g) = 0: other times
From the above, the point profile was obtained along the above-mentioned steps.

The results are shown in FIG. 6, and (a, b) = (. Infin., 0), (1/8, -1/8), (0, -1/4), (-1 / 16.-). 3/16), and the case of an elliptic filter are shown. As shown in the change from (a, b) = (1/8, -1/8) to (0, -1/4), the cluster size is coarse by making the band filter lower. It becomes a dot. Also, as shown in (a, b) = (0, -1/4) to (-1/16.-3/16), when the width of the band filter is narrowed, the periodicity in the radial direction is It becomes remarkable. In the case of an elliptic filter, when expressed by the values of principal axes in the x direction and y direction, x direction: (a, b) = (0, -1/4), y direction: 0.7 × (0, -1/4) Is shown in the figure. By making it elliptical, it becomes an anisotropic dot profile. Here, (a, b) = (∞, 0) has a blue noise characteristic and is not clustered. The RAPS (Radially Averaged Power Spectrum) value indicates a cyclically determined power spectrum distribution, and defines the peak spatial frequency as a frequency main value.

Next, a point profile at g = g ± Δg is obtained. The characteristics of the filter D (f_r, g) when g is changed broaden the bandwidth in the bright and dark areas so as to approach the blue noise characteristics. The reason is described below.

FIG. 7A is a gray scale output diagram when it is assumed that f_max and f_min are constant for all g and do not change. In the light and dark areas, the uniformity of the dots is impaired. Also, in the same figure (b),
f_max = f_max, 1/2
f_min = Min (f_min, 1/2, f_0)
And a gray scale output diagram. Here, f_max and 1/2 represent f_max and f_min when g = 1/2, and 1/2 represent f_min when g = 1/2. Also, Min (f_min, 1/2, f_0) represents the smaller value of f_min, 1/2 and f_0. The values of f_max and f_min for g are illustrated in FIG. Also in the case of FIG. 7 (b), the uniformity of the dots is similarly impaired in the bright and dark areas

The collapse of the uniformity of the point profile in the light and dark areas is mainly due to the inappropriate setting of f_max and f_min as the index band bands. In addition, since it is necessary to determine the dither threshold in accordance with the data of the previous gradation, optimization for each gradation can not be performed, and the degree of freedom is low. In particular, this effect is noticeable in the bright and dark areas of the image. In addition, since the dots become rough in the bright and dark areas, the graininess becomes particularly noticeable.

For these reasons, f_max and f_min are made variable according to g, and f_max and f_min are set so as to approach the blue noise characteristic from the green noise characteristic in the bright and dark areas. That is, in the bandpass filter D (u, v, g), the lower limit f_min of the spatial frequency of the band changes toward the low band toward g → 0 or g → 1 with g = 1/2, The upper limit f_max is set to increase toward the high frequency side, and the filter value in the region between the lower limit f_min and the upper limit f_max is set to take a finite value between 0 and 1. This makes it possible to avoid the graininess of the light and dark portions.

As an example, as shown in FIG. 9, one in which f_max and f_min are changed by the following three areas is used.
For 0 ≦ g <g_0: f_max = f_M; f_min = sqrt (g) · f_N
In g_0 ≦ g <g_1: f_max = f_M; f_min = f_min, 1/2
For g_1 ≦ g <1/2: f_max = f_max, 1/2; f_min = f_min, 1/2
Here, f_min, 1⁄2 and f_max, 1⁄2 is f_min and f_max at g = 1⁄2, g_0 is the intersection of f_0 and f_min, 1⁄2, g_1 is an arbitrary point, f_M is g_1 and Represents a straight line connecting R points. g is more than half folded symmetrically. The region (A) between f_max and f_min in each region is a multi-value in which the filter value gradually changes from 1 to 0 from f_min to f_max in the region where the filter value is 1 and (B).

In FIG. 10, the obtained threshold value matrix when (a, b) = (0, -1/4) in the case of R = 128, 64, 32 is represented as gradation data. Generally, in the case of a small size matrix, when performing large size image output, periodicity is likely to occur due to repetition of matrix size, but the green noise characteristic is exhibited, so the problem of periodicity is alleviated compared to the blue noise mask. FIG. 11 shows a dither matrix of R = 64 and outputs of an image of g = 1/2 are connected by 2 × 2 and output in a size of 128 × 128. The boundaries of the joints are not very noticeable.

In FIG. 12, according to the dither threshold value matrix when R = 128 and (a, b) = (0, -1/4), g = 1/8, 1/4, 1/2, 3/4, 7/8 The dot profile and the spectrum characteristic at the time of binarization output at the time of (tone value 224, 192, 128, 64, 32) are shown. It can be seen that the green noise characteristics in which the characteristics in the low frequency region and the high frequency region are degraded in each gradation are shown.

(A) Original image, (b) Output image with blue noise (BN), (c) (a, b) = (1/8, -1/8), (d) (g) Fig. 6 shows an output image of the green noise characteristic of the present invention in which a, b) = (0, -1/4).

Fig. 14 shows (a) an original, (b) an output image of blue noise (BN), and (c) (a, b) = (1/8, -1/8), (d) (a) of the Lena image. , b) = (0, -1 / 4) of the output image of the green noise characteristic of the present invention

Further, FIG. 15 shows color image output in three colors of cyan (C), magenta (M) and yellow (Y) in order from the top. Color moiré occurs by setting blue noise (BN) for C version, (a, b) = 1/8, -1/8 for M version, and (0, -1/4) for Y version It is difficult to do. Originally, in the FM screen, moire is less likely to occur in plate superposition due to the non-periodic structure, but it appears in the form of "granularity", resulting in image quality deterioration. This is because the periodicity of the micro domain of the FM halftone produces low frequency components by superposition and appears as graininess. As can be seen from the composed image, it can be seen that there are hardly any Rosetta patterns, graininess, color moiré, etc. that occur in the case of halftone printing.

FIG. 16 is a color output of a natural image, where (a, b) of each color plate is (a) C plate: (1/4, 0), (b) M plate: (0, -1/8) ), (C) Y version: (-1/8, -1/4), (d) Composite image is shown. The composite image is an overlay of color plates, and in the case of halftone printing, a rosette pattern occurs, but this case does not occur.

FIG. 11 shows combinations of cluster sizes: (a) C version: blue noise (BN), (b) M version: (1/4, 0), (c) Y version: (0, −1 / 8) and (d) show composite images. Compared to the case of FIG. 10, the cluster size is smaller in this case.

In the above, it has been described that the cluster size can be controlled by changing the spectral bands (a, b). By using this dither matrix, high-speed output of cluster halftone dots suitable for a printer engine is possible. This makes it possible to set an optimum cluster size for a recording engine having a low high frequency response characteristic, such as an electrophotographic printer, and enables stable image output.

As another application, an example is shown in which the dot pattern is controlled by changing the shape of the filter D. FIG. 18 is a diagram in which the output dots have anisotropy by using an elliptical band pass filter. (Left): x direction: (0, -1/4) y direction: 1.3 x (0, -1/4), (center): x direction: (0, -1/4) y direction In the case of 0.7 × (0, -1/4), (right) is obtained by rotating the pattern of (b) by 90 degrees.

FIG. 19 shows the output of each gradation using the dither matrix in FIG. 18 (center), where g = 1/8, 1/4, 1/2, 3/4, 7/8 (gradation It shows the binarized output at the values of 224, 192, 128, 64, 32) and their spectral characteristics. It can be seen that anisotropic green noise characteristics are maintained at each gradation.

FIG. 20 shows an image output using two types of dither matrices having different anisotropy. (a) is a 64x64 elliptic filter with x direction: (0, -1 / 8), y direction: 1.4 x (0, -1 / 8), and (b) is a 90 ° rotation of it. is there. (c) shows an image binarized alternately using such two dither matrices. (d) is obtained by applying FFT to an image cut out in units of 64 × 64 pixels to obtain spectral characteristics. It can be seen that corresponding elliptical shaped spectral characteristics are obtained.

Here, it is possible to conceal information in a binary image by associating "0" with the dither matrix of (a) and "1" with the dither matrix of (b). In order to detect the direction of the ellipse from the spectral characteristics, the x-axis cross section and the y-axis cross section of the spectral characteristics in the case of the elliptic filter of FIG. 4 can be determined easily by comparing the rising positions.

When embedding information in a 512x512 image with a 64x64 dither matrix, only 64-bit information can be embedded. In order to increase the amount of information to be embedded, it is possible to embed as much as 512 bits (64 bytes) of information by adding an orientation of ± 45 ° to the ellipse and using a 32 × 32 dither matrix.

The present invention has developed an image processing apparatus and method using a dither halftone mask exhibiting new green noise characteristics of cluster type as described above. This can be realized by confining the spectrum in a specific band (f_max, f_min) in frequency space. Smooth and high by adjusting the band of the band filter with each tone, in order to maintain uniformity and uniformity in the entire tone range, and in particular, to make the dot roughness in the light and dark region inconspicuous. It is possible to obtain an image quality output pattern.
As applications, acquisition of a dither matrix of cluster size according to the MTF characteristics of the recording engine, embedding of information by a dither matrix incorporating an anisotropy, etc. can be considered.

The present invention is useful for digital recording devices, particularly for printers with unstable high frequency characteristics such as electrophotographic systems. It is possible to easily select one that matches the printer's MTF characteristics, because it is easy to obtain a dither mask that changes the band pass filter parameters. Stable high image quality can be maintained by such cluster type halftone. In addition, it is possible to conceal information in the image and use it in various ways.

1 Scanner 2 electrophotographic printer 3 controllers 4, 5, 6, 7, 8, 9, 10 are respectively ROM, RAM, program memory, CPU, Display, keyboard, communication function
Reference numerals 11, 12, 13, 14, 15, 16 respectively indicate image memory, threshold data, comparator, laser driver, output signal 17 semiconductor laser light source 18 collimator lens 19 imaging lens 20 photosensitive drum 21 scanner

Claims (4)

In a digital image processing apparatus that generates a dither threshold matrix that outputs input image data in a cluster type halftone, the dither threshold matrix is calculated from two steps,
The first step is a two-dimensional FFT on the profile p (i, j, 1/2) of the point (i, j) at the intermediate value (g = 1/2) of the blackening ratio g (0 ≦ g ≦ 1) To convert P (u, v, 1/2) (where (u, v) is a two-dimensional space frequency), and for this P (u, v, 1/2), the radial frequency f_r is Perform a filter operation with the green noise filter D (u, v, 1/2) that has a value less than the lower limit f_min of the frequency and 0 or more and is greater than the upper limit f_max and has a value only in the range of f_min f f_r f f_max,
In the second step, from the point profile obtained in the first step, with respect to the blackening ratio g changed in Δg units, toward g → 0 or g → 1 centering on g = 1/2 Perform the filter operation with the green noise filter D (u, v, g) set so that the lower limit f_min of the spatial frequency of the band changes to the lower band and the upper limit f_max increases to the high band,
An image processing method for generating a dither threshold matrix characterized in that spatial frequency characteristics of an output image according to the dither threshold matrix calculated in the above steps have green noise characteristics lowered in low and high bands.
In the second step, the band pass filter D (u, v, g) has a lower limit value f_min of the spatial frequency of the band toward g → 0 or g → 1 with g = 1/2 as the center. The upper limit f_max is set to increase to the high band side, and the filter value in the region between the lower limit f_min and the upper limit f_max is set to take a finite value between 0 and 1 The image processing method of claim 1, wherein the dither threshold matrix is calculated.
In the first step, G = 1/2 is given as an initial value, and R ^ 2/2 black points are randomly given,
(STEP 1) Perform two-dimensional FFT of point profile p (i, j, 1/2) to obtain P (u, v, 1/2).
(STEP 2) A filter D (u, v, 1/2) is multiplied by P (u, v, 1/2) to obtain a new P '(u, v, 1/2). Here, D (u, v, 1/2) is a green noise filter having a value only in a region where the frequency f_r is f_min ≦ f_r ≦ f_max, and the filter value is 1.
(Step 3) Inverse Fourier transform is performed on P ′ (u, v, 1/2) to obtain a multi-valued point profile p ′ (i, j, 1/2).
(STEP 4) An error function: e (i, j, 1/2) = p '(i, j, 1/2)-p (i, j, 1/2) is obtained for each pixel. Next, the white pixels and the black pixels are arranged in descending order of error, and the white pixels → black pixels and black pixels → white pixels are inverted in descending order. At this time, the numbers to be reversed are equal.
(STEP 5) Repeat STEP 1 to STEP 5 until the error is within a certain tolerance. If all the pixels fall within the tolerance, a point profile of g = 1/2 is finally obtained.
Processing is performed, and then, in the second step, g ± Δg changed by one gradation unit,

(STEP 1) A two-dimensional FFT of the point profile p (i, j, g) is performed to obtain P (u, v, g).
(STEP 2) The filter D (u, v, g) is multiplied by P (u, v, g) to obtain a new P ′ (u, v, g). Here, D (u, v, g) is a filter in which f_min and f_max change depending on g, f_min decreases toward the low frequency side toward g → 0 or g → 1, and f_max increases toward the high frequency side.
(Step 3) Inverse Fourier transform is performed on P ′ (u, v, g) to obtain a multi-valued point profile p ′ (i, j, g).
(STEP 4) An error function: e (i, j, g) = p ′ (i, j, g) −p (i, j, g) is obtained for each pixel. Next, arrange the errors in descending order of white pixel and black pixel, and reverse white pixels to black pixels by N = R ^ 2/2 ^ n pieces in descending order (when g → 1), or black pixels are white pixels Invert to (when g → 0). At this time, the gradation value is written to the inverted pixel position of the dither matrix.
(STEP 5) Repeat STEP 1 to STEP 5 as g = g ± Δg. When g = 0, 1, this operation ends.
3. The image processing method for generating a dither threshold matrix according to claim 2, wherein the dither threshold matrix is calculated by performing the processing consisting of:
The dither threshold value matrix according to claim 3, wherein the band pass filter D (u, v, g) is an elliptical shaped band pass filter for calculating an anisotropic point profile. Image processing method to generate.