JP3283966B2 - Improved frequency modulation halftoning method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

This invention relates to a method for halftoning continuous tone images. In particular, this method relates to an improved version of the frequency modulation screening method.

[0002]

BACKGROUND OF THE INVENTION Many reproduction methods can reproduce only a small number of image tones. For example, offset printing can only be printed with or without two tonal values, namely ink. A halftoning method and a screening method are used to reproduce an image having a continuous tone.

[0003] Halftoning converts the density values into a geometric distribution of printable binary dots. The eye cannot see the individual halftone dots, only the corresponding "spatially integrated" density values.

[0004] Used in the field of graphic arts,
A major class of halftoning methods have been reported.
These two methods are known as "amplitude modulation screening method" and "frequency modulation screening method".

[0005] According to the amplitude modulation screening method, halftone dots represent a particular color tone, but are arranged in a fixed geometric grid. By changing the size of the halftone dots, images of different tones can be simulated. Therefore, this method can also be called "dot size modulation screening method".

According to the frequency modulation screening method, not the size of the halftone dots but the distance between the halftone dots is modulated. Although this method is well known in the field of low resolution plain paper printers, much attention has been paid to high end printing methods such as offset printing. Perhaps it is due to the disadvantages discussed below.

[0007] Both classes of halftoning methods are used in combination with digital film recorders. A typical digital film recorder is composed of a scanning laser beam for exposing a photosensitive material with high resolution. A "grid" that can switch the laser beam on or off determines the resolution, and this grid is typically 1/1800 inch in pitch. The photosensitive material may be a photographic film, from which a printing plate is later produced by photolithography. The smallest addressable units on a recorder are often called "microdots", "recorder elements" or "rels".

Each of these two classes of halftoning methods has several variations, each of which is briefly described and its advantages and disadvantages are considered. The most important characteristics of the screening method or halftoning method for faithfully reproducing a continuous tone image are as follows. 1) Rendering characteristics: More specifically, the ability to represent the full range of tones, as well as the spatial details of the original image content without introducing artifacts such as moiré, texture and noise. Performance. 2) Photolithographic characteristics of algorithmically generated halftone dots: This characteristic is how the halftone dots are recorded, transcribed or replicated in a uniform manner in the different steps of photolithographic production of the printing plate. Or decide. 3) Halftone behavior in offset printing presses. 4) Speed performance achievable in this way.

The main advantages of the amplitude modulation screening method are that it can be easily executed as a high-speed algorithm for electronically screening images, and has excellent photolithographic reproduction characteristics. However, a significant disadvantage of the amplitude modulation screening method is that undesirable patterns occur in the halftoned image. By their cause, these patterns are called original moiré, color moiré and internal moiré.

[0010] Original moiré is due to the geometric interaction between the periodic elements in the original material and the halftone screen itself. To easily understand this problem, it is convenient to use the Fourier transform. One analysis is Jour
nal of the Optical Society
y of America, Vol. 65, No. 6, 716-72
Page 0, 1975 Kermisch and P.M.
G. FIG. Roetling's paper "Fourier Spe
ctrum of Halftone Images "
It has been done.

[0011] Solutions for reducing original moiré are described, for example, in US Pat. No. 5,130,821 and EP 3693.
No. 02 and No. 488324. However, these solutions do not completely solve the above problems.

Color moiré is caused by interference between halftones of different color separations of an image. To alleviate this problem, it has been proposed to use screen angles shifted from each other by 60 ° for different color separations. Some have been disclosed about the problem of making screens at or near such angles (eg, US Pat. Nos. 4,419,690 and 4,350).
No. 996, No. 4924301 and No. 51555
No. 99). It has also been described to use other combinations of halftone dot pattern angles, frequencies or relative phases for different color separations to overcome the problem of color moiré (see, for example, US Patent No. 44430).
Nos. 60, 4537470 and EP 501126).

[0013] Internal moiré is a pattern due to geometric interaction with the addressable grid in which the halftone is generated. Methods to reduce internal moiré typically eliminate or "smear" the phase error that is periodically maximized as a result of the frequency and angular relationship between the halftone screen where the moiré is represented and the addressable grid. Based on introducing random elements. An example of such a method is described in U.S. Pat.
No. 456924, No. 4499489, No. 4700
No. 235, No. 4918622, No. 5150428
And International Patent Application Publication No. WO 90/04898.

Since none of the variations of the dot size modulation screening method can completely eliminate the moiré problem, dot modulation screening methods have been proposed to further alleviate this problem.

Although various dot frequency modulation screening methods have been disclosed, they can be classified into the following subclasses. 1) A method based on thresholding between two points. 2) Diffusion of errors by scanning (and deformation) from row to row and column to column. 3) Error propagation due to Hilbert's scan (and deformation). 4) Special method.

A special method is disclosed in German Patent No. 293,092 and further described in US Pat. No. 4,485,397.

The most typical point-to-point thresholding method is "B
ayer "dither matrix (Bayer, B .;
E. FIG. , "An optimal method for
two level rendition of c
ontinous-tonepictures ", P
rc. IEEE International Co
nference on Communication
s, Conference Record, 26-
11, 26-15, 1973). This Bayer dither matrix is
Having a power of two size and thresholded to increasing levels of density, all halftone dots are "as far as possible" from the halftone used to represent the lower density levels. The threshold values are arranged in such a manner. There are a number of changes in the order of the thresholds in the dither matrix, "Lipp
known as the "Eland Kurland" dither matrix and the "Jarvis" dither matrix (Stoffels, JC, Mor).
eland J. F. , “A survey of
Electronic Technologies for
Pictorial Reproduction ”,
IEEE Transactions on Com
munications, COM-29, No. 12,
December 1981, p. 1898-1925).

Another point-to-point thresholding method uses "Blue Noise" instead of the Bayer dither matrix.
Mask ", which is disclosed in U.S. Pat. No. 5,111,310.
No. This Blue Noise Ma
sk is obtained by iterative optimization (filtering) (for the next threshold "layer") during the Fourier transform of the threshold matrix.

The halftone dot pattern produced by the Bayer dither matrix contains a strong periodic component that can be seen as a "texture", and its corresponding power spectrum is clearly "sharp". This can also cause moiré problems similar to the dot size modulation algorithm. However, compared to the dot size modulation method, the energy of the periodic dither component "spreads" much more evenly over the different harmonics and most of the components have relatively high frequencies, so the aliasing that occurs is less disturbing. Much smaller.

The "Blue Noise Mask" threshold matrix distributes halftone dots, the two-dimensional power spectrum of which is not "sharp" and "continuous". Therefore, this method does not have the problem of the periodic aliasing that occurs in the case of the dot size modulation method or the Bayer dither matrix.

The continuous character of the power spectrum of the Blue Noise Mask method already suggests that at least some energy is also present in very low frequency bands of the spectrum. Low-
The presence of this energy at the visible -frequency is the reason why the shades represented in this way appear grainy. The relationship between "granularity" and the shape of the frequency spectrum is discussed in detail by Urichney Robert (Urichney Robert, "Digit
al Halftoning ", MIT Press
Cambridge Massachusetts,
1987, ISBN 0-262-2109-
6).

Bayer dither matrix halftones do not behave very well on offset printing presses. The tone curve is "irregular" and strongly depends on the manner in which the printing press is operated, especially in the color range where halftone dots begin to connect. This problem is the same as dot size modulation halftone with extremely high ruling.

The "Blue Noise Mask" halftone performs much better. The range where the halftone dots connect extends over the entire tone scale, and the press gain is high but stable and easy to control. At the same time, the "Blue NoiseMask" method
Viewed as a method using a continuous range of screen ruling, it helps to understand the above.

All of the point-to-point thresholding methods are all such that once the threshold matrix is calculated, the actual halftoning process is fast, ie as fast as the fastest form of dot size modulation. It has the advantage that. However, the point-to-point thresholding method has the limitation that the distribution of higher density halftone dots must be made "on top" of the lower density value distribution. This makes it impossible to optimize the halftone dot distribution at each individual density level.

Probably the best known of all "dot frequency modulation" methods is the error diffusion algorithm. The image is processed from row to row and column to column, and the errors resulting from binarization (or quantization under more general circumstances) of the image data during rendering may be one or more. "Diffused" to unprocessed pixels. Depending on how this error is diffused (how many pixels are weighted), one can distinguish between several algorithms, usually named by the inventor. The best known is the algorithm of Floyd and Steinberg (Floyd, RW and L. Steinb).
erg, "An adaptiveivegorith
m for spatial grayscale ",
Proc. (SID, 17/2, 75-77)
, But additionally, Jarvis, Judith and Ninck filters (Jarvis, JF, C.I.
N. Judeice, and W.S. H. Nink
e, 1976, "Anew technology.
for displaying continuous
tone pictures on bilevel
displays ", IEEE Tran. on
Commun. , COM-24, 891-898
P.); Stucky's filter (Stucki, P.,
1979, "MECCA, -a multiple"
error correcting computer
Tion algorithm for bileve
lhardcopy reproduction ",
Research Report RZ1060, I
BM Research Laboratory, Zu
Rich, Switzerland); and Stephenson and RL (Stevenson, RL).
and G. R. Arce, 1985, "Bi
nary Display of hexagonal
ly sampled continuous ton
e images ", J. Opt. Soc. A
m. 2, No. 7, pages 1009-1013).

There are many variations on the basic error diffusion algorithm described above. A simple variant consists of processing the pixels according to a "zigzag" sequence. That is, the directions of error diffusion are from left to right of "even" rows and right to left of "odd" rows (or vice versa). Urichney (Urichney Robert
t, “Digital Halftoning”,
MIT PressCambridge Massac
Husets, 1987, ISBN 0-262.
2109-6) describes a random perturbed weight as a method of reducing "worm-like" textures that occur at about 50% tint levels.
(Eight) is disclosed. One important variant related to the present invention is the diffusion of errors with one randomly selected weight. In this variant, the error is distributed to exactly one unprocessed pixel, which is randomly selected in the vicinity of the pixel to be processed.

Instead of perturbing the weights, it is also possible to perturb the threshold at which the pixel values are compared for binarization. Another method of reducing worming is disclosed in US Pat.
Nos. 130823 and 5150429.

In US Pat. No. 5,130,819, the diffused error is the error averaged over a small area of pixels that has already been processed, instead of the local error at the location of only one pixel.

Since the error diffusion method is designed to minimize both local and global errors between the original image content and the halftones, a frequency modulation method based on thresholding and It has better rendering characteristics when compared. It is excellent at expressing details, and the particles that appear pale are small. However, in some cases, a form of "edge enhancement" is introduced, which is not always desirable.

The printing characteristics of the standard error diffusion method are less desirable. This can be explained by the relationship between the total dot perimeter versus the integral dot area for the Floyd and Steinberg algorithms. At low densities, individual halftone dots are not connected and the perimeter of the dot increases in proportion to the number of dots and the integrated dot area. At a given tint, halftone dots begin to connect, and the entire dot perimeter increases less than the integrated dot area. However, at about 50%, the algorithm produces a checkerboard-like pattern whose perimeter is very high relative to the integrated dot area. Above 50%, the same behavior repeats symmetrically. The inventors of the present invention have discovered that the transition in distribution of halftone dots ordered from less than 50% to about 50% or more causes irregularities in the press gain, which irregularities are greatly affected by the press setting. They found that the gradation and neutral balance became unstable. This situation is superior in this respect to variations of the standard error diffusion algorithm, which suppresses the above tendency and produces a texture at about 50% tint level.

The class of error diffusion methods is inherently slower than the dot size modulation method and the point-to-point thresholding method based on the above method because of the need to calculate and diffuse the error. Error diffusion with larger weights is slower than error diffusion with smaller weights.

All of the above error diffusion algorithms have in common that the order in which the pixels are processed is linear, ie, from left to right or vice versa and from top to bottom or vice versa. Witten and Neal describe in their literature (WittenIan H., and R
adford M.A. Neal, “Using Pean
o Curves for Bilevel Disp
layer of Continuous-Tone Im
Ages ", IEEE CG & A, May 4, 1982
7-52) employs another method. The problem of their invention was to change the order of processing of the pixels themselves. That is, in their method, the error is always propagated from the previous pixel to the next pixel in one order and follows a path known as "Peano Curve". “Digital Halftoni
ng with Space Filling Cur
ves ", Luiz Velho, Jonas de
Miranda Gomes, ACM Compu
As suggested in ter Graphics, Vol. 25, No. 4, 1991, other curves can be used, such as the "Hilbert Curve" shown in FIG. All of these curves
It has the property that it is a “space filling defined fractal curve”.

The rendering characteristics of the Hilbert and Peano scanning method are comparable to those of the Floyd and Steinberg algorithm, but the achievable speed is faster than the fastest error diffusion algorithm (error diffusion algorithm with one weight). This is because it is not necessary to store the sum of the errors propagated by the values of the pixels in the next order. Alternatively, this sum can be used immediately to determine the halftone value, after which a new error is calculated. This achievable high speed makes the error propagation algorithm very attractive for processing high resolution images.

Unfortunately, the Hilbert and Peano scanning method has the same problem as the Floyd and Steinberg algorithm in that it tends to produce strong patterns and textures at a given integrated dot area.
As shown in FIG. 2, when these textures appear,
Change the dot circumference irregularly along the color scale,
In addition, when these shades are printed on an offset printing machine, irregularities occur in the gradation and neutral balance.

[0035]

SUMMARY OF THE INVENTION It is an object of the present invention to provide an improved method for screening continuous tone images using a frequency modulation screening method.

Further objects of the present invention will become clear from the description hereinafter.

According to the present invention, a continuous tone image (20)
Selecting an unprocessed image pixel (20) according to the space-filling two-dimensional curve (10) and then converting the unprocessed image pixel (20) as follows, ie, the unprocessed image From the tonal value of the pixel (20) , the reproduction value (3) used to record the image pixel on a recording medium (80) , for example a photographic film or display.
0) and determines an error value (5 ) based on the difference between the tone value of the unprocessed image pixel (20) and the reproduction value (30).
0) , so that the unprocessed image pixel (20)
Is the image pixel to be processed (42) , and the error value (50)
Is added to the tonal value of the unprocessed image pixel (20) and then the resulting tonal value (42) is substituted for the tonal value or the unprocessed image pixel to which the error value is distributed Distributing the error value over two or more unprocessed image pixels by replacing the tonal value of the unprocessed image pixel with the sum of the tonal value of the unprocessed image pixel and a portion of the error; and A method for screening a continuous tone image, comprising the step of repeating and processing until processing , wherein said curve is randomized.
A method is provided.

The present invention will be described by way of example, but the present invention is not limited by the following figures.

According to the method according to the invention, the pixels of the continuous tone image are halftoned in the order described in the randomized space-filling two-dimensional curve. Halftoning the pixels in this order determines the halftone value closest to the continuous tone pixel value, calculates the difference between the continuous tone pixel value and the halftone value, and then at least some of the error This is achieved by adding one or more pixels in the following order:

Diffusion or propagation of errors gives the algorithm rendering properties comparable to known error diffusion or propagation algorithms. When the types of pixel processing paths are randomized, the generation of patterns and textures that degrade halftone printability is avoided. If the error is propagated only to the next pixel in halftoning order, performance equivalent to error propagation along Peano scan can be achieved.

The term "randomized space-filling two-dimensional curve" means a two-dimensional curve connecting all image pixels, which is in principle deterministic, but the deterministic character of this curve is a large number. Are subdivided by randomizing the curve at A particularly advantageous randomized space filling two-dimensional curve is a randomized space filling defined fractal curve.

A randomized space-filling two-dimensional curve can be obtained in a number of ways. For example, image pixels are processed by randomly selecting each pixel each time a new unprocessed pixel needs to be selected.

According to another method, the image is divided into several matrices of pixels of the raw image. Each image pixel in the matrix is then processed by random selection until all are selected. When one matrix is processed, the next matrix is processed until all the matrices have been processed. The order in which the matrices are processed is also random or in a particular order, for example, left to right and top to bottom.

Another variation on the above method is to recursively redivide the image into sub-matrices until the size of the matrix matches the pixels. For each subdivision,
The order in which the matrices are processed can be determined at random.

According to yet another method, the order in which the pixels are selected is described in a two-dimensional space-filling deterministic fractal curve, for example, a Hilbert curve or a Peano curve, which has no specific points at predetermined points to subdivide the deterministic structure. Be randomized.

[0046]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings, but the present invention is not limited thereto.

FIG. 4 shows a circuit for implementing a new halftoning method in combination with a binary recording device. First, the different building blocks of this circuit will be described, and then its operation will be described.

Block 20 is a memory block having continuous tone pixel values of the image. Generally, these values are 8-bit values and are composed of N rows × M columns. Block 3
Reference numeral 0 denotes a memory block having the same layout as the block 20, in which halftoned pixel values are stored. In the case of a binary recording device, all halftoned pixel words are one bit long. The block 80 is a device that can record the information of the block 30 on a substrate. Block 70 is an arithmetic unit that can calculate the sum of the pixel value P (i, j) and the error E in the output of the delay register 60. The conversion of continuous tone pixel values to halftoned pixel values is performed at block 40. This conversion is based on a thresholding operation.
That is, if the continuous tone value at point (i, j) is smaller than the value of 128, the value of "0" is stored in the halftone memory, otherwise, the value of "1" is stored.
Block 50 includes an arithmetic unit capable of calculating the continuous tone value at point 42, the sum of the original continuous tone value and the error, and the error between the halftoned pixel values and storing it in the delay register 60. Contains. Block 8 is a counter that orders the processing of the N ^* M pixels of the image. Block 10 is an LUT with N ^* M entries (one for each image pixel) and a unique combination of row and column addresses that match one pixel location in the image. Two methods for calculating such a table will be described later. Block 5 is a clock.

The operation of this diagram will be described. At each clock pulse, the counter 8 is incremented and a new pair of coordinates (i (n), j (n)) is obtained from the block 10. These coordinates are used as address values for the pixel memory 20, and the continuous tone pixel value P (i (n), j
(N)) is obtained. This pixel value immediately has the error E (i
(N-1), j (n-1)). This pixel value has been stored in the register 60 after having undergone the half-toning step first, and the total value of both is stored in the block. It is compared in 40 with a threshold value 41. As a result of performing the thresholding operation, the value H (i (n), H (i (n),) written to the halftone pixel memory at the position (i (n), j (n))
j (n)) is determined. At the same time, a new error E (i
(N), j (n)) is P (i (n), j (n)) and H
Calculated from the difference of (i (n), j (n)) and then stored in delay register 60. The circuit sets the counter 8 to 1
Initially, by setting the error to 128, operation is stopped when the counter reaches the N ^* M level. The halftone memory 30 is then read from row to row and column to column, and its contents are stored in the recorder 8
0 is recorded on the substrate.

The circuit of FIG. 4 requires a large amount of memory.
That is, sufficient memory is required to accommodate the LUT with N ^* M entries containing the sequence of address values to control the sequence of all continuous tone images, all halftoned images, and pixel processing. Therefore, another method that requires only a small memory is disclosed. By subdividing the image into bands that are processed and recorded one by one, the amount of memory required can be reduced. FIG. 5 shows a diagram for a band with four lines.

The circuit shown in FIG. 5 is an extension of the circuit shown in FIG. However, the circuit of FIG.
The 0 and halftone memory blocks 30 contain only four lines of image data. Therefore, LUT10 has 4
^* Contains only coordinate pairs for M data entries.
Therefore, LUT 10 contains only coordinate pairs for 4 ^* M data entries. Pixel counter 8 and LUT1
A modulo 4 ^* M arithmetic unit 9 is arranged between the addressing of 0. Block 11 moves image data from mass storage device 12, for example, a hard disk, to pixel memory 20. The block 11 also has a detector 13 that sends a signal when the output of the arithmetic unit 9 is equal to zero.

The operation of the circuit of FIG. 5 is as follows. If the output of the arithmetic unit 9 is zero, the block 11 stores the four lines of image data from the mass storage
Move to zero. Next, these 4 ^* M pixels are halftoned according to the method of the present invention. After a 4 ^* M clock pulse, a new set of four lines of image pixels are moved from mass storage 12 to pixel memory 30. The 4 ^* M halftoned pixels contained in the memory 30 are also recorded on the substrate by the recording device (80).

Two methods are provided to obtain the contents of block 10 of FIG.

The first method is based on the Hilbert path randomization. The Hilbert path is a two-dimensional "fractal" curve, which can be obtained by a recursive program. FIG. 1 shows some recursions of the process of generating a 32 × 32 Hilbert path. Table 1 at the end shows some C programs that can be used to obtain such a path. The randomization of this path is linear. That is, the path goes from point to point,
At every point in the path, it is randomly determined whether to swap the current and proceed to the next point. This randomization process can be performed one or more times. FIG. 6 shows the Hilbert curve after randomization. FIG. 7 shows a graph of total perimeter versus integrated dot area from a Hilbert scan with two randomizations. Compared to FIG. 2, the graph of FIG. 7 is very smooth, with no peak at about 50% light level. FIG. 8 shows that no texture or patterning has occurred along the tone scale.

In the second method, the rectangular matrix is recursively subdivided into smaller sub-matrices. In every subdivision, a random permutation order number is assigned to every submatrix. This process is stopped when subdivision is at the level of individual matrix elements. An example is shown in FIG. In FIG. 9, 3
A 2x32 matrix has 2x with 16x16 elements
Recursively subdivided into two matrices, after which all sub-matrices are again subdivided into 2 × 2 sub-matrices of 8 × 8 elements. Next, all of these matrices are again subdivided into 2 × 2 individual matrix elements. Table 2 at the end shows the C program that implements this process, while FIG. 9 shows the paths obtained by that program. The graph of total dot perimeter versus dot area according to this method, shown in FIG. 10, is very smooth and shows no apparent texture. Testing confirmed that the halftone exhibited the above attributes and that the printing was very stable within a wide range of press parameters. The printman reported that the neutral balance and gradation were stable and the gradation was significantly smoother.
At certain tonal levels there were no tonal jumps and excellent shadow detail was easily achieved. There was no moiré and the details were much better than the dot size modulation method.

The above method of obtaining a path is absolutely configured to create a matrix of square images or image parts. The actual state is hardly such a state. Three solutions are available to adjust for this problem: In a first solution, the original continuous tone image is
It is "padded" at zero value along its longest side until it becomes a square. In the second method, the path is calculated based on the size of the longest side of the rectangular image. Points on the path that do not match the image pixel position simply skip over when passing through that path. In a third solution, the image is, for example, 16 × 1
It is divided into square parts such as six elements. A randomized path is obtained for one of these squares using one of the methods described above. After that, the route is connected. This method still requires that the dimensions of the image be a multiple of the dimensions of the squares, but these squares are relatively small (16 × 16 in our example).
So, the above does not actually bring any restrictions.

Obviously, many variations exist on the embodiments provided herein. In one variation, devices that record multiple levels instead of two can be used (eg, devices such as thermal sublimation printers, slide recorders, electrophotographic printers, etc.). The thresholding operation shown in FIG. 4 is then replaced by thresholding with a number of thresholds. The halftone value depends on which of the two thresholds the continuous tone pixel value lies between. For performance and adaptability, the address value matches the sum of the error and pixel value,
On the other hand, an LU whose content contains the corresponding halftone value
With T, it is best to perform a thresholding operation in block 40.

In another variation, the error resulting from the difference between the continuous tone pixel and the halftoned pixel value is propagated to one or more unprocessed pixels instead of being propagated only to the next pixel in processing order. Is done.

Still another variation is that the error E in FIG. 4 is based on the current pixel P (i (n), j (n)) and the halftone value H (i (n), j) of the current pixel. (N)), but not when obtained as a weighted average from the error of one or more processed pixels. The circuit that accomplishes this is disclosed in FIG.

FIG. 11 shows a more sophisticated variant of the circuit of FIG. The differences are a first set of k shift registers 43 where the unprocessed pixels are delayed k times, a second set of 1 shift registers 63 where the difference between the continuous tone value and the halftoned value is delayed once, and a half set. The third set of the K shift register 33 in which the coordinate value for addressing the tone memory 30 is delayed k times is added. Further, E (i) of the weighted average value derived from the output of the shift register 63
(N−k), j (n−k)), and the average value is calculated as k
An arithmetic unit 65 for distributing to the output of the shift register 43 is provided.

In each cycle of clock 5, the data in shift registers 33, 43 and 63 is shifted by one position, and counter 8 is incremented by one. This is done in three general operations which the inventors of the present invention refer to as "halftoning", "error combining" and "error distribution".

First, the halftoning will be described. The new value of the counter 8 is the coordinate pair i (n),
j (n), and then the pixel P in the pixel memory 20
(I (n), j (n)). The latter value is
Used for the input of the shift register 33. The output of the same shift register 33 has a value Q (i) based on (but not necessarily equal to) this pixel value P (i (nk), j (nk)) delayed k times. (N-
k), j (nk)). This value Q (i (n-
k), j (nk)) are compared with a threshold value 41, and based on this comparison, the halftone value H (i (nk), j
(Nk)) is determined, and is written to the cell 31 at the address (i (nk), j (nk)) of the halftone memory 30. The latter address is obtained by delaying the address value coming from the memory block 10 k times by the shift register 33.

Next, the "error combining" portion will be described. Value Q
(I (nk), j (nk)) and H (i (nk),
j (nk) difference d (i (nk), j (nk))
Is calculated by the arithmetic unit 50 and the shift register 6
3 is added to the input. Each output of one cell of the shift register 63 has a difference value d (i (nk-1),
j (nk-1)), d (i (nk-2), j (n-
k-2)). . . d (i (nk-1), j (nk-
1) is given. From these 1 values, the multiplier w
(1), w (2). . . The error value E (i (nk), j (nk)) is synthesized by w (1) and the arithmetic unit 65. This error value E (i (nk), j (nk))
Indicates a weighted average value of an error introduced as a result of performing quantization over a region of one pixel that has been halftoned earlier.

The last step is to calculate the error E (i (n−
k), j (nk)) are distributed to unprocessed pixel values. This is because the error E (i (n−
k), j (nk)) with weights v (1), v (2). . .
This is done by multiplying by v (k) and then adding the resulting value to the respective output of k in the cell of shift register 43. Thus, the error E (i (n−
k), j (nk)) are spread over the region of the value of k in the shift register 43.

The circuit of FIG. 11 can be modified to process images of bands in a manner similar to that described in FIG.

It will be apparent to those skilled in the art that the halftoning method can be implemented in hardware or software.

[0067]

[Table 1]

[Table 2]

[Table 3]

[Table 4]

[Table 5]

[Table 6]

[Table 7]

[Brief description of the drawings]

FIG. 1 shows a Hilbert curve.

FIG. 2 shows the dot perimeter versus tonal value (0 = minimum density, 256 = maximum density) for halftoning, where the pixels are processed in the order of a Hilbert curve.

FIG. 3 shows a halftoned tonal scale obtained by processing image pixels in the order in which the Hilbert curve is drawn.

FIG. 4 is a schematic explanatory diagram of a circuit for implementing the halftoning method of the present invention.

FIG. 5 is a schematic explanatory diagram of a circuit for processing an image of a band.

FIG. 6 shows a randomized Hilbert curve.

FIG. 7 shows dot perimeter vs. tone value (0 = minimum density, 256 = maximum density) for halftoning, where pixels are processed in the order of a randomized Hilbert curve.

FIG. 8 shows a halftoned tonal scale obtained by processing image pixels in the order in which a randomized Hilbert curve is drawn.

FIG. 9 illustrates the order in which image pixels are processed when the image is recursively subdivided into a matrix.

FIG. 10 shows dot perimeter vs. tone value (0 = minimum density, 256 = maximum density) for halftoning, where the pixels are in the order obtained by recursively subdividing the image into a matrix. Has been processed.

FIG. 11 is a schematic explanatory diagram of a circuit for implementing the halftoning method of the present invention, wherein a weighted average value of a difference between a tone value of at least two image pixels and their corresponding reproduction values is obtained. It is spread to many pixels.

[Explanation of symbols]

Reference Signs List 5 clock 8 pixel counter 9 modulo 4 ^* M arithmetic unit 10 LUT 12 mass storage device 13 detector 20 pixel value memory 30 halftone pixel value memory 31 cell 40 halftone device 41 threshold value 33,43k shift register 63 1 shift register 60 delay register 50, 65, 70 arithmetic unit 80 recorder

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Paul Delabastita Septestart, Mortzeel, Belgium 27 Inside Agfa Geverth na Mroseze Bennacht Chap (56) References JP-A-3-151762 (JP, A) JP-A-3-243063 (JP, A) JP-A-4-37256 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) H04N 1/40-1/409

Claims (1)

(57) [Claims] 1. A method for screening a continuous tone image (20) , comprising selecting an unprocessed image pixel (20) according to a space-filling two-dimensional curve (10) and then selecting the unprocessed image pixel (2).
0) from the tonal value of the unprocessed image pixel (20) as follows: the reproduction value used to record the image pixel on the recording medium (80).
(30) , wherein the tone value and the reproduction value of the unprocessed image pixel (20) are determined.
An error value (50) is calculated based on the difference between the unprocessed image pixel (20) and the unprocessed image pixel (20).
(42) , adding the error value (50) to the tone value of the unprocessed image pixel (20) , and then replacing the tone value with the obtained total value (42) , or By replacing the tonal value of each unprocessed image pixel whose value is distributed with the sum of the tonal value of the unprocessed image pixel and a portion of the error, the error value is spread over two or more unprocessed image pixels. Distributing; repeating the above steps until all image pixels have been processed , wherein the curve is randomized in a method for screening continuous tone pixels comprising the step of processing.
A method characterized in that: