DE29522383U1 - Image processing device - Google Patents

Description

·· &psgr;

Image processing device

Background of the invention

The present invention relates to an image processing apparatus having a function of transforming image data having multiple gradations into image data having two gradations (binary coding function) for the purpose of reducing the amount of data and outputting the image data to a CRT display or a printer or the like. More particularly, the present invention relates to an apparatus for binary coding image data using an error diffusion method, an image input device, an image processing device and an image output device, each of which includes the binary image coding device.

When image data with multiple gradation levels is output to a printer or display device that does not have a function for controlling the gradation on a pixel basis, or when the amount of data for storage and data transmission is reduced, a binary coding process for reducing the number of gradation levels of each pixel to two gradation levels is widely used. There are a variety of binary coding methods. Of these methods, the error diffusion method is popular because it can produce the image with the best quality.

The error diffusion method is based on "diffusion of errors" in which a quantization error caused during the process of binary coding of a pixel is distributed to the adjacent pixels that are not yet binary encoded in accordance with their weights without discarding the quantization error. Due to this, an average value of the binary coding errors in a local area of the image is extremely small. The

Quality of the image produced by the error diffusion method is much higher than that achieved by the systematic dither method in which the binary coding error is discarded. The error diffusion method produces an image with high resolution and is capable of reproducing a reproduction with a continuous gradation. Regarding the conventional method using the error diffusion method, reference is made to Published Unexamined Japanese Patent Application No. Hei. 1-284173 entitled "IMAGE PROCESSING METHOD AND APPARATUS FOR CARRYING OUT THE SAME".

In distributing the errors by the error diffusion method, a means which describes "how to specify the adjacent pixels to which the errors are distributed and how to weight these specified adjacent pixels" is called an "error diffusion matrix". The number of adjacent pixels to which the error is diffused is called a "matrix size". Some examples of the error diffusion matrices of the type in which the adjacent pixels are below a target pixel are shown in Fig. 5(a) to 5(e). The matrix sizes of these error diffusion matrices are 2, 4, 7, 10 and 13, respectively. When the error diffusion matrix shown in Fig. 5(b) is used, a binary coding error caused in the target pixel is quartered with an equal weight. These shared errors are each added to the gradation image data of their adjacent four pixels that are not yet binary encoded, namely one pixel on the right, one pixel on the bottom left, one pixel on the bottom and one pixel on the bottom right.

In the error diffusion method, it is necessary to distribute the binary coding error caused in the target pixel to the neighboring pixels not yet binary coded by using any of the error diffusion matrices as shown in Fig. 5(a) to 5(e). Therefore, the number of processing steps is noticeably increased when compared with the systematic dither method. The time for processing is also increased. The processing time and the number of processing steps are both increased when the size of the error diffusion matrix becomes large. Where a high

• » · t

processing speed is required, it is preferable to use the error diffusion matrix with the small size.

Let us consider the error diffusion method from the viewpoint of image quality. In a case where the error diffusion matrix of small size is used to reduce the processing time, the error diffusion area is small. In this case, if a density of the distributed dots formed by the binary coding process is low, "a distribution of dots will not be uniform, dots will be unevenly lined up, and the quality of the resulting image is mediocre." For example, when original image data of such low density that after being binary encoded, the ratio of black dots to all dots is 10% or less is binary encoded using the error diffusion matrix of small size (the error diffusion area is small) as shown in Fig. 5(a) or 5(b), the result of binary encoding the image data is as shown in Fig. 8(a). As shown, black dots are lined up while having uneven spacing. The resulting image quality is poor. When using the large error diffusion matrix with large dimensions (which provides a large error diffusion area) as shown in Fig. 5(e), the image quality is improved as shown in Fig. 8(b). However, when using the error diffusion matrix with large dimensions, the processing amount and the processing time are increased as described above. Thus, the conventional error diffusion method cannot satisfy both requirements of the image quality and the processing amount and time at the same time.

Summary of the invention

The present invention has been made in view of the above circumstances and has for its object to provide an image processing apparatus which can binary encode a multi-level image, whereby a good dot distribution, even at a low dot density, and a high image quality can be ensured without increasing the processing amount and the processing time.

This object is solved with the features of claim 1.

used for diffusion calculation, depending on the original image gradation data value of the target pixel and the binary coded result of the target pixel, wherein the binary coded result of the target pixel is the first gradation value, the error diffusion matrix selecting means selects an error diffusion matrix having a large size when the original image gradation data value differs greatly from the first gradation value and approaches the second gradation value, the binary coding error being limited to a larger value.

In the following description, a pixel to be binary encoded is referred to as a target pixel. The process for binary encoding the target pixel will be described below.

The binary coding means compares the correction data value of the target pixel with the threshold value, binary encodes the target pixel to the first gradation value or the second gradation value based on the comparison result, and produces the binary encoded result. The correction data is formed by adding binary coding errors provided from the binary encoded pixels adjacent to the target pixel to the original image gradation data of the target pixel. More specifically, the binary encoding is carried out in the following manner.

It is assumed that the first gradation value < second gradation value. Under this assumption,

if the correction value > threshold, the binary coding result is equal to the first gradation value, and

if the correction value > threshold, the binary coding result is equal to the second gradation value.

If the correction value = threshold, the binary coding result can be equal to the first gradation value or the second gradation value.

The error diffusion method calculates a binary coding error, that is, the difference between the correction data value and the binary coding result. This is

r ··

Binary coding error = correction data - binary coding result

The error diffusion means distributes the calculated error to the adjacent pixels not yet binary-coded in accordance with the weights defined by the error diffusion matrix. At this time, the error diffusion matrix selecting means selects an error diffusion matrix having a larger size when either of the following two conditions (a) and (b) holds while referring to the result of binary coding of the target pixel and the value of the original image gradation data, thus realizing error diffusion to a larger area.

(a) binary coding result = first gradation value, and original value of image gradation data = second gradation value

(b) binary coding result = second gradation value, and original value of image gradation data = first gradation value

A first point of the present invention is that a good dot dispersion property and the binary coding result with high image quality are ensured by using the error diffusion matrix with large magnitude only when the condition (a) or (b) holds. When the error diffusion matrix with large magnitude is selected under the condition (a), the dots binary coding to the first gradation value are evenly distributed. When the error diffusion matrix with large magnitude is selected under the condition (b), the dots binary coding to the second gradation value are evenly distributed.

A second point of the present invention is that the conditions (a) and (b) occur with an extremely small probability. For the condition (a), for example, if the original value of the image gradation data is approximately equal to the second gradation value, the binary coding result will also be the second gradation value with a high probability, while it will be the first gradation value with a low probability. Therefore, the use of the error diffusion matrix with large dimensions occurs with an extremely small probability. In most cases, the error diffusion

··

Matrix with small dimensions is used so that the processing amount and time are minimized.

Short description of the drawings

Fig. 1 is a block diagram showing a first embodiment of the present invention.

Fig. 2(a) and 2(b) are block diagrams showing an application of the present invention in which the present invention is applied to a system each including a scanner,

Fig. 3(a) and 3(b) are block diagrams showing another application of the present invention in which the present invention is applied to a system

which contains a printer,

Fig. 4(a) to 4(e) are tables showing examples of the selection conditions of the error diffusion matrix according to the present invention,

Fig. 5(a) to 5(K) are diagrams showing examples of the error diffusion matrices used in the invention,

Fig. 6 is a graph showing a variation of the selection probability for a large matrix with respect to a selection point for a matrix,

Fig. 7 is a diagram showing a target pixel and its neighboring pixels, namely binary coded pixels and not yet binary coded pixels,

Fig. 8(a) and 8(b) are diagrams each showing states of diffused black dots in a low density image area, and

Fig. 9 is a flow chart showing the flow of the binary coding process in the first embodiment of the invention.

Detailed description of the preferred embodiments

The preferred embodiments of an image processing apparatus according to the present invention will be described in detail with reference to the accompanying drawings.

Fig. 1 is a block diagram showing a first embodiment of the present invention. In the figure, an error diffusion binary coding means 20 receives and binary encodes the original image gradation data data (i, j) from an image gradation data output means 10, and outputs the result result (i,j) of the binary coding to an image binary data output means 30. In the above notation, / and j denote integers. The data described as data (i,j) denotes the data of the pixel located at the /-th position counted downward from an initial pixel and at the /-th position counted rightward from an initial pixel. In Fig. 1, the error diffusion binary coding means 20 is a key element of the image processing apparatus of the present invention. The embodiment of the present invention to be described hereinafter is specified as follows.

a) The original image gradation data used in the embodiment is data with 256 gradation levels in the range of 0 (white) to 255 (black). The pixel density becomes higher as the gradation value becomes higher.

b) The original image gradation data is binary encoded to 0 or 255 using the error diffusion method.

c) For the scanning sequence for binary coding, the main scanning direction means the direction of the binary coding scanning in which the binary coding proceeds from left to right. The sub-scanning direction means the direction of the binary coding scanning in which the binary coding proceeds from top to bottom.

More specifically, the process of binary coding the data of all the pixels of a frame starts from an initial pixel at the upper left corner of the screen and proceeds in the main scanning direction, one element at a time, in sequential order. When the binary coding process reaches the pixel positioned at the right end, the binary coding of all the pixels of a line is completed. Then, the process of binary coding shifts to the line below by one pixel in the sub-scanning direction, and the binary coding is carried out from the pixel at the left end to the pixel at the right end to complete the

» &diams; · bold · - .

to complete the binary coding of all the pixels of the second line. In this way, the binary coding is done for all pixels of a full image.

The pixel located at the /-th position from the starting pixel counted downwards and at the y-th position from the starting pixel counted to the left is called the pixel of the row /and column/ and is written in the form Pp, j] . Here the starting pixel is expressed as

P [0, O]. It is assumed that the pixel P [i,j] is the target pixel to be binary coded. For the normal pixels except the pixels located at the extreme ends of the image, as shown in Fig. 7, the pixel (on the upper line) just above the target pixel and the pixel to the left of the target pixel on the same line are the binary coded pixels. The pixel (on the lower line) below the target pixel and the pixel to the right of the target pixel on the same line are the pixels not yet binary coded.

In this embodiment, the "first gradation value" is 255 and the "second gradation value" is 0. Furthermore, in this embodiment, only when the result of the binary coding is the first gradation value, the error diffusion matrix is switched to another error diffusion matrix.

Now, the procedure of binary coding the original image gradation data data (i,j) of the pixel p, j] carried out by the error diffusion binary coding means 20 (Fig. 1) will be described with reference to Fig. 9 which shows a flow chart. The binary coding procedure consists of five steps 1 to 5.

[Step 1]

The data correction means 101 reads an accumulated diffusion error value total_err (i, j) from the diffusion error storage means 105 and adds it to the original image gradation data data (i,j) obtained from the gradation image data output means 10. As a result, the data correction means 101 produces correction data data_c (i,j).

data_c(U)'rgeta(jyj).+'1fteJ_en.ßnl ··;:.;;.";: jj -j

The accumulated diffusion error value total_err (i,j) stored in the diffusion error storage means 105 indicates the total sum of the errors that relate to the target pixel

P//,i/in a step 5 which will be described later, during the process of binary coding those pixels O'ene are already binary coded and referred to as P [i-1,j] and P [i, j-1] ) which are adjacent to the target pixel P [i, j] are diffused.

[Step 2]

The binary coding means 102 compares the correction data data_c (i,j) of the target pixel, which is obtained from the data correction means 102, with the threshold value thresh and generates the binary coding result result (i, j). Here:

If data_C (i, j) > thresh, result (I, j) = 255. If data_C (i, j) < thresh, result (i, j) = 0.

In this version thresh = 128

The binary coding result result (i,j) output from the error diffusion binary coding means 20 is also supplied to the binary image data output means 30. In this case, it is sometimes output in the form of binary data with 0 or 1. For example, when result (i, j) = 255, it is replaced by "1".

Step 3]

The error diffusion means 103 first calculates the binary coding error err (I, j) using the following equation:

err (Y, j) = data_c (i, j) - result (i, j)
wherein

data_c (i, j): correction data

result (i, j): binary coding result

[Step 4]

The error diffusion matrix selecting means 104 sets a first gradation selection point t_chg1 to 32 and selects an error diffusion matrix matrix in the following manner.

a) If result (i, j) = 0, the small matrix Mat_S is selected.

b) If result (i, j) = 255,

if data (i, J) < t_chg1, the large matrix Mat_L is selected, and if data (i, J) > t_chg1, the small matrix Mat_S is selected.

[Step 5]

The error diffusion means 103 distributes the binary coding error err(i,j) to the adjacent pixels such as the pixels P[i,j+1], P[i ⁺ 1,j], and P [i+1,j+1] in accordance with the weights of the error diffusion matrix. More specifically, the diffusion error derived from the target pixel P[i,j] is added to the accumulated values of the diffusion errors total_err(i,j+1), total_err(i+1,j+1) and the like for each pixel stored in the diffusion error storage means 105. Before starting the sampling for binary coding the pixels of the current image, it is necessary to set the initial values of the previous accumulated values of the diffusion errors to 0.

In this embodiment, the small matrix Mat_S is the error diffusion matrix shown in Fig. 5(b) and the large matrix Mat_L is the error diffusion matrix shown in Fig. 5(d). If the small matrix Mat_S is selected as the error diffusion matrix, the total sum of the weights is 4. Then, the result of multiplying err(i,j) by the weight and dividing the resulting product by 4 is the amount of diffusion error to be distributed to the adjacent pixels. An actual error diffusion process is carried out as given below.

[Table 1]

total_err (i, j+1) = total_err (i, j+1) + err (i, j) * (1/4)

total_err (M, j) = total_err (M, j+1) + err ft /> * (1/4) total_err (M, j+1) = total_err (M, j+1) + err ft j) * ( 1/4)

When the large matrix Mat_L is selected as the error diffusion matrix, the total sum of the weights of the matrix of Fig. 5(d) is 16. The error diffusion process is carried out as given below.

[Table 2]

total_err ft j+1) = total_err ft /+ 1) + err ft j) * (3/16)

total_err ft j+2) = total_err ft j+2) + err ft j) * (1 /16)

total_err (M, j-2) = total_err (M, j-2) + err ft /) * (1/16)

total_err (7+ f, y- 1) = total_err (i+1, j-1) + err ft j) * (2/16)

total_err (i+1, j) = total_err (i+1, j) + err ft j) * (3/16)

total_err (i+1, j+1) = total_err (i+1, j+1) + err ft j) * (2/16)

total_err (1+1, j+2) = total_err (i+1, j+2) + err ft j) * (1 /16)

total_err (i+2, j-1) = total_err (1+2, j-1) + err ft j) * (1 /16)

total_err (i+2, j) = total_err (i+2, j) + err ft /) * (1 /16)

total_err (i+2, j+1) = toial_err (i+2, j+1) + err ft j) * (1 /16)

The process of binary encoding the target pixel P [i,j] and distributing the resulting errors to its adjacent pixels is completed when steps [step 1] through [step 5] of the procedure have been performed. Then the value; is incremented by one, and the binary encoding process begins again, with a new target pixel as the pixel located to the right of the previous target pixel. This binary encoding process is repeated for all the pixels on a row. After that, the value; is set to zero, and the value / is incremented by one, so that the pixel at the left end on the next row becomes the new target pixel. The binary encoding process is again repeated for all the pixels on the next row. In this way, the data of a full image is binary encoded.

In the binary coding procedure performed by the error diffusion binary coding means 20, steps [Step 4] and [Step 5] are unique, while the remaining procedure essentially corresponds to the

corresponding Pr$4&dur 'dee^QnventiarilJen FöhteF-piffusäoias-S/erfahrens.

In those steps, one of two types of error diffusion matrices is selected under specific conditions, and error diffusion is performed using the selected matrix. The original image data and the results of binary coding define the matrices as shown in Fig. 4(a).

The reason why this embodiment provides good dot diffusion will be described. The inventor of the present patent application considered the "mechanism for uniformly distributing black dots" in the error diffusion method and the reason why the mechanism is disturbed and the black dots are lined up as in Fig. 8(a). The "mechanism which "uniformly distributes black dots" in the error diffusion method can be considered as follows.

1) In the error diffusion method, when a pixel is binary encoded as 255 (black dot), the binary coding error err (i, j) generated in the step [step 3] is usually negative, and the negative error is diffused to the adjacent pixels not yet binary encoded in the step [step 5] .

2) The corrected data of the adjacent pixels that received the negative error are smaller than the data of those pixels before they were corrected. This fact makes it difficult to binary encode the data of those pixels as 255 (black dot).

3) As a result, the black dots are evenly distributed in a separate manner while not being lined up.

In order to distribute the black dots evenly, the distance between the adjacent black dots must be increased as the density becomes lower. For example, when the gradation is 16, the ratio of the black dots generated is 16/255 = _1/16 . In other words, one black pixel is generated every 16 pixels. The distance between the black dots is approximately four pixels ⁱⁿ the two-dimensional plane. When the error diffusion matrix shown in Fig. 5(b) is used, an error generated when a black dot is not directly distributed _among the pixels is calculated.

from the target pixel at a distance of two or more pixels. For this reason, the black dots are lined up while being about two pixels away from each other. There is such an indirect diffusion effect that "from a pixel to which the error has been diffused, the error is further diffused to its adjacent pixels". Due to this indirect diffusion effect, there is a possibility that an error is transferred to pixels outside the matrix. However, when the matrix is small, this effect is also weak. In the case of the error diffusion matrix of Fig. 5(b), no error is transferred from the target pixel P [i,j] to the pixel P [i+1,j-2] even if the indirect diffusion effect exists.

As a consequence, it is thought that the "reason why the black dots are lined up in low density image regions as shown in Fig. 8(a)" is that "in low density image regions, although the distance between the adjacent black dots must be increased, the error cannot be diffused to the range of the necessary point-to-point distance by the small size error diffusion matrix".

Based on the above considerations, it is concluded that from the point of view of "improving the diffusion of black dots", it is not always necessary to use the error diffusion matrix with large dimensions, and that

1) only the binary coding error caused when the pixel data is binary encoded into the black dot is distributed over a larger area, while the binary coding error when the pixel data is binary encoded into the white dot is distributed over an area that is not so wide, and that

2) in accordance with the point-to-point distance necessary to ensure uniform distribution of points, the size of the selected matrix is larger as the density of the image area becomes lower. In the image area with medium and high density, the error diffusion matrix with small size can be used.

The image processing apparatus of the present embodiment is designed on the basis of the above conclusions, and proves successful in remarkably improving the pixel distribution in low density image areas. In the image processing apparatus, in order to select the error diffusion matrix, the matrix selection point t_chg1 is set to 32. The large matrix Mat_L is selected only when the original image gradation is set to a value less than 32 and the result of binary coding is 255 (black point). Under other conditions, the small matrix Mat_S is selected.

In the present embodiment, as described above, a probability that the large matrix Mat_L is selected is very small, and in most cases, the small matrix Mat_S is selected. The reason for this will be described. Usually, a probability that the result of binary coding is 0 and a probability that the result of binary coding is 255 depend on the original value of the image gradation data (i,j) , and they are as shown below.

Probability that result (i, j) = 0: (255 - data (i, j) I 255)

Probability that result (i, j) = 255: data (i, j) 1 255

Since the large matrix Mat_L is selected under the conditions that result (i,j) = 255 and

data (i,j) < 32, is a probability that the large matrix Mat_L is selected

data (i,j) 1 255 if data (i,j) is within the range 0 to 32, and 0 if data (i,j) is within the range 33 to 255.

Accordingly, when data (i,j) = 32, the large matrix Mat_L with the highest probability is selected. This probability is 32 / 255 (« 0.13). Thus, even in the worst case where the original image data data (i,j) are all, the probability that the large matrix is selected is 13% or less. In a case where the original values of the image data are uniformly distributed in the range of 0 to 255, a probability that the large matrix

is selected is 1% or less. The number of processing steps and the processing time are almost the same as those in the case where the small matrix is used alone. A variation of the probability that the large matrix is selected with respect to different values of the first gradation selection point t_chg1 is shown in a graph of Fig. 6. The formula for calculating the probability is given by

Lchg1 Σ (k/255) k = 0

As the first gradation selection point approaches the second gradation value (0 in this embodiment), the probability that the large matrix is selected becomes small. This is advantageous in improving the processing speed. As can be seen from the graph of Fig. 6, in a range where the first gradation selection point t_chg1 is 127.5 (an intermediate value between the first and second gradations) or less, the probability is about ten and several % or less, and satisfactorily small. In a range where the first gradation selection point t_chg1 goes beyond 127.5, an increasing part of the probability that the large matrix is selected gradually increases. Therefore, in a case where only a first gradation selection point is used as in this embodiment, it is preferably set to 128 or less.

As is apparent from the foregoing description, the image processing apparatus of the first embodiment outputs the result of binary coding with high image quality and good distribution of black dots, which are comparable to those achieved by the image processing using only the large matrix. The processing time for the binary coding is almost equal to that with the image processing using only the small matrix. The present embodiment brings no improvement in the dispersion characteristics of the white dots in a high density image area in particular. In the printer system in which the output apparatus prints dots with black ink on white paper, black dots will spread in a low density image area, enlarging their areas, and it is

likely that they will be conspicuous. On the other hand, white &bull; · ··· · · · ··· · .· ·. .**. *. · · ·

&iacgr; 4 # * I

Dots in the high density image areas tend to reduce their areas as a result of the dispersion of their adjacent black dots. Accordingly, the white dots become hardly conspicuous. From this fact, it is apparent that the image quality can be remarkably improved by improving only the dispersion property of the black dots as in the present embodiment.

A second embodiment of the present invention will be described. A block diagram showing the arrangement of the second embodiment is similar to that of the first embodiment shown in Fig. 1. Consequently, reference will be made to the block diagram of Fig. 1 in the description of the second embodiment. The first embodiment is designed to improve the dispersion property of the black dots in a low-density image area. The second embodiment is designed to improve the dispersion property of the white dots in the high-density image area, in addition to the dispersion property of the black dots in the low-density area. For this purpose, a second gradation selection point 2 t_chg2 is used in addition to the first gradation selection point 1 t_chg1 . The large matrix is also used when the result of the binary coding is 0 (white dot) in the high-density image area where the original image data data (i,j) > t_chg2 . In the second embodiment, t_chg2 = 233. The mechanism for improving the dispersion property of white dots in high density areas is exactly the same as that for improving the dispersion property of black dots if the logic for the dispersion property of black dots is reversed.

The second embodiment differs from the first embodiment only in the step [Step 4] . The step [Step 4] in the first embodiment is replaced by the step [Step 4'] in the second embodiment.

[Step 4']

The error diffusion matrix selecting means 104 selects the error diffusion matrix in the following manner by selecting the first gradation selection

Point t_chg1 (set to 32 in this embodiment) and the second gradation selection point t_chg2 (set to 223 in this embodiment) are used.

a) If result (i,j) = 255,

if data (i,j) <t_chg1, the large matrix Mat_L is selected, and if data (i,j) >t_chg1, the small matrix Mat_S is selected.

b) If result (i, j) = 0,

if data (i,j) >t_chg2, the large matrix Mat_L is selected, and if data (i,j) <t_chg2, the small matrix Mat_S is selected.

In this embodiment, the original image data and the result of binary coding define the matrices as shown in Fig. 4(b). The reason why the number of processing steps and the processing time slightly increase in this embodiment is as described in the first embodiment. Thus, the second embodiment improves the dispersion property of the black dots in the low density areas and the dispersion property of the white dots in the high density areas. Accordingly, the second embodiment further improves the image quality when compared with the first embodiment.

A third embodiment of the present invention will be described next. In the third embodiment, the large matrix for improving the dispersion property of the black dots is different in construction from that for improving the dispersion property of the white dots, while in the second embodiment, the former is the same as the latter. As already stated, in the printer system, the black dots are usually noticed more readily than the white dots. In the present embodiment, the error diffusion matrix shown in Fig. 5(e), which is larger than the large matrix Mat_L, is used as the large matrix for the black dots. This matrix is called the "giant" matrix Mat_H . With the use of the giant matrix Mat_H, the dispersion property of the black dots in the low density areas is further improved. The first point for gradation selection t_chg1 for black dots is 32, the same as that in the second embodiment, but the second point for gradation selection t_chg2 for white dots is greatly increased from 223 to 239. This reduces the frequency of the Verweodurjajdec gcaßeta matrix to

to increase the processing speed. In the present embodiment, the step [Step 4'] in the second embodiment is replaced by the following step [Step 4 "7.

[Step 4"]

The error diffusion matrix selecting means 104 sets the first gradation selection point t_chg1 to 32 and the second gradation selection point t_chg2 to 239 and selects an error diffusion matrix matrix in the following manner.

a) If result (i, j) = 255,

if data (i,j) <t_chg1, the huge matrix Mat_H is selected, and if data (i,j) >t_chg1, the small matrix Mat_S is selected.

b) If result (i, j) = 0,

if data (i,j)>t_chg2, the large matrix Mat_L is selected, and if data (i,j) <t_chg2, the small matrix Mat_S is selected.

In this embodiment, the original image data and the results of binary coding define the matrices as shown in Fig. 4(c). When the huge matrix is selected, the error diffusion process in the next step [Step 5] is naturally different from that in the second embodiment. However, the basic idea thereof is the same as in the second embodiment, and hence the detailed description thereof is omitted here.

A fourth embodiment of the present invention will now be described. In the fourth embodiment, the number of matrix sizes is increased to three sizes, a small size, a large size and a giant size, in accordance with the gradation of the original image to improve the dispersion of black dots. Further, the processing time and image quality are optimized. The construction of each of the error diffusion matrices with small, large and giant sizes is the same as that of the third embodiment. In the fourth embodiment, the selection of the matrix is made in the step [Step 4/as shown in Fig. 4(d).

A fifth embodiment of the present invention will be described. In this embodiment, a mean matrix Mat_M is additionally used as one of the error diffusion matrices. Thus, a more fine selection of the matrix is performed. The error diffusion matrix shown in Fig. 5(c) is used for the mean matrix Mat_M . The selection of the matrix in step [Step 4] is made as shown in Fig. 4(e) in accordance with the data value of the original image and the binary coding result.

In the above-mentioned embodiments, the error diffusion matrices of Figs. 6(b), 5(c), 5(d) and 5(e) are assigned to the small, medium, large and giant error diffusion matrices, respectively. However, any error diffusion matrix may be used if the order of the matrix size is appropriate.

The above-mentioned embodiments are designed with more emphasis on improving the dispersion property of the black dots than that of the white dots. When the binary coding result is the black dot, the error diffusion matrix is finely switched. However, when the output device for the binary image data is a CRT display device, white dots are more emphasized than black dots. In this case, the black and white dots may be treated equally. Alternatively, the image processing device may be designed to place emphasis on the white dots. In this case, when the binary coding result is the white dot, the error diffusion matrix is finely switched.

In the above-mentioned embodiments, the gradation data corresponds to such a density that the value 0 of the gradation is assigned to the white point and 255 to the black point. However, even for gradation data corresponding to such a brightness that the value 0 is assigned to the black point and 255 to the white point, the basic idea is substantially the same as that of the previous embodiments, except that the logic with respect to black and white is reversed.

The image processing apparatuses of the above-mentioned embodiments deal with monochromatic data. It is evident that the technical idea of the present invention is applicable to color data consisting of various color components. In this case, the invention is applied to each color component. For complete color data consisting of three color components of colors R (red), G (green) and B (blue), the image processing apparatus of the present invention is applied to each of these color components. The method of matrix selection may be different for each color component. For example, the matrix is switched at many steps for the color components of R and G as shown in Fig.4(d). For the color component of B which is less noticeable than the color components of R and G ₁ , a pattern of matrix selection as shown in Fig.4(a) is used.

In the above-mentioned embodiments , the threshold value thresh used in step [step 27] is a fixed value. Alternatively, the threshold value thresh may be varied depending on the pixel by adding systematic dither noise which varies in fixed periods depending on the position of the pixel or an appropriate amount of random noise to the threshold value thresh .

In the binary coding sampling for binary coding the image data of a frame with multiple gradation levels, the starting pixel can be set to any position other than the upper left corner of the screen. The scanning direction can be changed after each line so that the pixels of the odd-numbered lines are scanned from the left end to the right end, and the pixels of the even-numbered lines are scanned in the opposite direction, from the right end to the left end. In this case, the error diffusion matrix must be reversed after each line when used.

Any kind of error diffusion method, except the process of step [Step 4] which is essential to the present invention, can be used for the present invention.

The error diffusion binary coding means 20 may be implemented by software or hardware designed exclusively for the error diffusion binary coding method.

2 and 3 show specific illustrations of parts of the image processing apparatus other than the error-diffusion binary coding means 20 shown in Fig. 1. Fig. 2(a) shows an image input apparatus including error-diffusion binary coding means 20. Fig. 3(a) shows an image output apparatus including error-diffusion binary coding means 20. Figs. 2(b) and 3(b) show computer devices each including the error-diffusion binary coding means 20.

Fig. 2(a) is an application of the present invention in which the present invention is applied to an image input device. The error diffusion binary coding means 20 is included in the scanner as an image input device. In the arrangement of the image input device, a CCD sensor 202 corresponds to the gradation image data output means 10 in Fig. 1, and a computer 203 corresponds to the gradation image data output means 30 in Fig. 1. In the image input device, the CCD sensor 202 within the scanner 201 reads a reflection density of each pixel in an original image, processes the read image data for analog-to-digital conversion and sensor characteristic correction, and outputs the thus processed image data as gradation image data data (i,j) to the error diffusion binary coding means 20. The error diffusion binary coding means 20 binary encodes the gradation image data. The result of the binary coding result (i,j) is transmitted as the output of the scanner 201 to the computer 203 through an interface such as SCSI. The binary data received from the computer is displayed with a CRT display device, processed with image processing software such as photo retouching software, or stored on an external storage such as a hard disk. The error diffusion binary coding means 20 may be realized by storing an error diffusion binary coding program in a ROM, and a CPU included in the scanner reads it from the ROM and executes it. Alternatively, it may be implemented by

a hardware with high processing speed can be realized, which is only used for the process of error diffusion.

Fig. 2(b) shows another application of the present invention in which a scanner driver included in a computer includes the error diffusion binary coding means 20. A scanner 211 corresponds to the gradation image data output means in Fig. 1. A photo retouching software 214 corresponds to the binary image data output means 30 in Fig. 1. In the application, a command for reading an image is executed in, for example, the photo retouching software 214 running in a computer 212 so that the scanner 212 reads an image in response to the command. The scanner 211 outputs the read gradation image data data (i,j) directly to the computer 212 without binary coding the gradation image data. In the computer 21, a scanner driver 213 including the error diffusion binary coding means 20 binary encodes the received gradation data. The result (i,j) of the binary coding is transferred to the photo retouching software 214 running in the computer 212. Under the control of the photo retouching software 214, the result of the binary coding is displayed by the CRT display or further processed. Usually, a printer driver is a kind of program executed by the CPU of the computer, and the error diffusion binary coding means 20 is described in the form of software. The computer may include high-speed processing hardware used exclusively for error diffusion, and the printer driver may execute the binary coding processing at high speed by using the hardware.

Fig. 3(a) shows still another application of the present invention in which the present invention is applied to an image output device. The error diffusion binary coding means 20 is included in a printer as an image output device. In this application, a computer 301 corresponds to the gradation image data output means 10 in Fig. 1, and an ink jet head 303 corresponds to the gradation image data output means 30 in Fig. 1. The gradation image data data (i,j) output from the computer 301 is processed by the error diffusion binary coding means 20 ⁱⁿ a manner consistent with :·: : ·· .:« »

the result (i,j) of the binary coding, the inkjet head 303 ejects ink droplets onto a printing sheet, thereby forming dots on the sheet.

Fig. 3(b) shows still another application of the present invention in which a printer driver in a computer includes the error diffusion binary coding means 20. In this application, a photo retouching software 312 corresponds to the gradation image data output means 10 in Fig. 1, and a printer 314 corresponds to the gradation image data output means 30 in Fig. 1. For example, a printer output command is executed in the photo retouching software 311 running in a computer so that the photo retouching software 311 outputs the gradation image data data (i,j) for transmission to a printer driver 313. The printer driver 313 binary encodes the data data (i,j) using the error diffusion binary encoding means 20 included therein and outputs the result (i,j) of the binary encoding together with a printer control command to a printer 314. In turn, the printer 312 subsequently prints a binary image on a print sheet.

As is apparent from the foregoing description, in the present invention, an error diffusion matrix selecting means selects an error diffusion matrix of an appropriate matrix size in accordance with the original image data value of a pixel to be binary-coded and the result of binary coding. An error diffusion matrix of large size is selected only when necessary. Therefore, a binary coding apparatus is realized which is good in dot dispersion characteristics, has high processing speed, and is good in image quality. This good dot dispersion characteristic produced by the binary coding apparatus is comparable to that obtained when only the error diffusion matrix of large size is used. The processing speed for producing the good point dispersion property is almost the same as that obtained when only the small-sized error diffusion matrix is used.

Claims (2)

1. An image processing apparatus comprising a binary coding processing device for converting multi-level image data into two-level image data representing the binary coding result formed from only two gradation levels having a first gradation value and a second gradation value, the binary coding processing device comprising:
Binary coding means ( 102 ) for binary coding a multi-level image data value of a target pixel into the first gradation value or the second gradation value and
Error diffusion means ( 103 ) for distributing a binary coding error caused during the process of binary coding the multi-level image data value of the target pixel to its neighboring pixels which have not yet been binary coded,
marked by
a plurality of error diffusion matrices for distributing a binary coding error to the adjacent, not yet binary coded pixels, wherein the error diffusion matrices have different sizes for carrying out the distribution, and
Error diffusion matrix selection means ( 104 ) for selecting one of the plurality of error diffusion matrices used by the error diffusion means ( 103 ), the selection being dependent on a) the multi-level image data value of the target pixel and b) the binary coding result of the target pixel, wherein the error diffusion matrix selection means ( 104 ) are arranged to select an error diffusion matrix of large size for distributing the binary coding error over a large area when a) the binary coding result of the target pixel is the first gradation value and b) the multi-level image data value of the target pixel is significantly different from the first gradation value and approaches the second gradation value. 2. Image processing device according to claim 1, characterized in that the error diffusion matrices also have different shapes for distributing the binary coding error in the horizontal direction.