JPH0370433B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a preprocessing method for image binarization processing, and particularly to image processing for sharpening a multilevel quantized image.

In facsimile machines, digital copying machines, and the like, a document is scanned to sample the image density, which is further quantized into multi-value quantization to obtain a multi-value quantized image, which is then subjected to binarization and other necessary processing. MTF (Modulation) of the image reading section of such equipment
Transfer Function) is generally 30% to 40% when using a scanner such as a CCD image sensor.
%. Therefore, MTF correction is often applied to improve image quality.

In this MTF correction, if the pixel of interest is E in Figure 1, the surrounding pixels B, D,
This is done according to the following formula with reference to F and H.

E'=αE+β(B+D+F+H)...Equation (1) Here, E is the density level of the pixel of interest before correction,
E' is the density level after correction of the pixel of interest, B, D,
F and H are the density levels of surrounding pixels B, D, F, and H. α and β are coefficients that determine the degree of correction.

If the MTF is about 30% to 40%, the coefficients α and β in equation (1) are selected to be about α=3 and β=−0.5.

Applying this type of MTF correction has the effect of clearly sharpening line images such as thin lines and small characters. However, if the density level of the background is close to the intermediate level between white and black, and there is a small level difference, the level difference will be emphasized and cause grainy noise (black dots), which will cause a problem if the quality of the reproduced image deteriorates. be. For example, when sending or receiving a thin document with carbon on the back by facsimile, black spots may appear on the carbon part of the received image, but this is a side effect of MTF correction.

An object of the present invention is to provide a method for sharpening a multi-level quantized image, which improves the above-mentioned drawbacks.

According to the present invention, the difference in density level between a pixel of interest in a multivalued quantized image and each of its surrounding adjacent pixels on the upper, lower, left, and right sides is detected. If both of these level differences are less than a predetermined threshold, the density level before performing the MTF correction process is selected as the density level of the pixel of interest. On the other hand, if there is even one level difference that is greater than a predetermined threshold,
The density level obtained by performing MTF correction processing is selected as the density level of the pixel of interest. Binarization processing is performed on the thus selected density levels before or after MTF correction.

The present invention will be specifically described below with reference to the drawings.

FIG. 2 shows a functional block diagram of an example of an apparatus for performing image sharpening processing according to the present invention.

Pixel density signals read by a known scanner 100 are digitized by an A/D converter 101. In this example, the number of output bits of the A/D converter 101 is 4 bits, and the density of each pixel is set to level 0.
(white) to level 15 (black) shall be 16-value quantized.

The pixel density data output from the A/D converter 101 is stored in the register 10 of the register matrix 102.
At the same time as input to 2I, 1 line buffer (1)
103. 1 line batshua (1) 103
The output is input to the register 102F of the register matrix 102, and is also input to the 1-line buffer (2) 104 at the next stage. 1 line buffer (2) 1
The output of 04 is input to register 102C of register matrix 102. 1 line buffer(1)1
The pixel density data in 03 and (2) 104 are sequentially shifted to the right in synchronization with the pixel transfer timing of the scanner 100. Therefore, 1 line buffer(1)1
03, (2) Two lines of pixel density data are temporarily stored in 104. Also within the register matrix 102, registers 102I, 102F,
The contents of 102C are each shifted to the right register.

In the register matrix 102, the contents of the register 102E are the density data of the pixel of interest as the processing target. That is, in the case of correspondence with FIG. 1, the density data of pixel E is stored in register 10.
2E, density data of pixels D and F before and after on the same line are stored in registers 102 and 102F, density data of pixels B and C of the previous line are stored in registers 102B and 102C, and density data of pixels D and F of the next line are stored in registers 102B and 102C. The density data of pixels H and I are stored in registers 102H and 102.
Each exists in I.

Registers 102B, 102D, 102E, 10
The outputs of 2F and 102H, that is, the density data of the pixel of interest E and the pixels B, D, F, and H above, below, to the left, and to the right are input to the determination circuit 106. This judgment circuit 106 includes four comparators 106B, 106
D, 106F, 106H, and an OR gate 109 that takes the logical sum of the outputs of each comparator. The comparator 106B compares the density level difference (absolute value) between the peripheral pixel B and the pixel of interest E |B-E| with a certain determination threshold K, and if |B-E|K, outputs a "1" signal, and Otherwise, it outputs a “0” signal. Other comparators 106D, 106
Similarly, for F, 106H, the density levels of the surrounding pixels D, F, H and the pixel of interest E |D-E|, |F-E|,
|H-E| is compared with the determination threshold K, and if |D-E|K, a "1" signal is output from the comparator 106D, and if |F-E|K, a "1" signal is output from the comparator 106F, |H -E|K, a "1" signal is output from the comparator 106H. Otherwise, Companter 106D,
106F and 106H output "0" signals. Comparators 106B, 106D, 106F, 10
The logical sum signal of the output signal of 6H is the OR gate 109
Therefore, it is sent to the selector 108 as an output signal of the determination circuit 106. Note that the determination threshold K is, for example, K
=2 is selected.

Adder 105 includes registers 102B, 102
Receive the density data of four pixels B, D, F, and H on the upper, lower, left, and right sides of the pixel of interest E from D, 102F, and 102H, and calculate the sum of the density levels of these four pixels S=B+D
Calculate +F+H.

The density data of the pixel of interest in the register 102E and the output data S of the adder 105 are input to a conversion ROM (read only memory) 107.
This conversion ROM 107 stores the algorithm of equation (1) above in advance, and outputs the MTF-corrected density level E' of the pixel of interest. Then, the data of the corrected density level E′ and the density level E before correction
data is input to the selector 108.

The selector 108 receives “1” from the determination circuit 106.
When the signal is given, MTF-corrected density data E' output from the conversion ROM 107 is selected and output. Conversely, when the output of the determination circuit 106 is "0", the density data before MTF correction is selected by the selector 108 and output.

Note that the function of the selector 108 is
It is also easy to implement with 07. In other words, if the output of the determination circuit 106 is "1", E'=αE+β(B
+D+F+H), and if the output of the determination circuit 106 is "0", the internal pattern of the conversion ROM 107 can be created so that E is output as is. Of course, it is also possible to implement the functions of the conversion ROM 107 using hardware.

Binarization processing is normally performed after the selector 108, but this binarization processing is also performed in the conversion ROM 107.
It is also possible to have them executed at the same time.

Furthermore, the determination circuit 106 may be implemented as a ROM.

Here, the effects of the present invention will be specifically explained in comparison with the conventional method.

Figure 3 shows the processing results according to the conventional method, and shows the density level E'=αE+β(B
+D+F+H) is shown in the middle column, and its binarized output is shown in the right column. However, α=3, β=−0.5,
Pixel density level is level 0 (white) to level 15 (black)
16 values, and the binarization threshold is set to 7.5, which is the 50% level.

In pixel patterns a and b, the density level of the pixel of interest is lower than the binarization threshold before MTF correction, but the MTF
After correction, it exceeds the binarization threshold. Therefore, both binary outputs become "1". Such pixel patterns often occur on the white part of thin slips where the carbon on the back side is transparent, and originally two
This is the case when a pixel whose value output should be "0" becomes "1". In other words, this is an example where MTF correction causes unnecessary black spots.

Conversely, the binarized output of the target pixels in pixel patterns e and f should be "1", but since the level difference after MTF correction is below the binarization threshold, the binarized output is "0". It gets old and so-called white spots occur. In general, even in image areas with the same density, variations in density levels (density unevenness) of about ±1 level, including quantization noise, are unavoidable. many. This is also a drawback of conventional processing methods.

FIG. 4 shows the processing results according to the method of the present invention, where the density level E'=αE+βS of the target pixel after MTF correction for the pixel pattern shown in the left column, the level difference |H-E|, |F- E |, |D-E|, |B-E
The magnitude relationship between | and the determination threshold K and the level output as a processing result are shown in the middle column, and the binarized output for that output level is shown in the right column.
However, S=B+D+F+H, α=3, β=-
0.5, K=2, and the density level of the pixel is quantized into 16 values from level 0 (white) to level 15 (black). Furthermore, the binarization threshold is selected to be 7.5, which is the 50% level. Note that the pixel patterns a to f are the same as those in FIG. 3.

Now, in pixel patterns a and b, the level difference between the density level (before correction) E of the pixel of interest and the density levels H, F, D, and B of each surrounding pixel |H-E|, |F-
E|, |D−E|, and |B−E| are all smaller than the determination threshold K. Therefore, since the density level E before MTF correction is selected as the output level, its binarized output becomes "0". In the conventional method, the binarized output becomes "1" for these pixel patterns, and black dots occur as noise, but according to the present invention, such noise can be prevented from occurring. The same can be said about pixel pattern k.

On the other hand, for pixel patterns e and f where white spots were a problem, the level difference |H-E|, |F-
Since E|, |D-E|, and |B-E| are all smaller than the determination threshold K, the level E before MTF correction is selected as the output level, so its binarized output becomes "1". Therefore, white spots can be prevented.

As detailed above, according to the present invention, unnecessary black spots that could not be removed conventionally can be removed, and
On the other hand, it is also possible to prevent white spots. From this, in images between narrowly spaced lines, it is possible to prevent unnecessary points from occurring between the lines.
A clear image between lines can be obtained, and a character image with good contrast can be obtained. Also, there will be no breaks in the lines. That is, there are no unnecessary dots between the lines that make them look like a single thick line, and there are no lines that are broken and the characters are difficult to read, so a very high quality image can be obtained.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of pixel positions, FIG. 2 is a functional block diagram showing an example of an apparatus that executes image sharpening processing according to the present invention, FIG. 3 is a diagram for explaining processing results according to the prior art, and FIG. FIG. 4 is a diagram for explaining the processing results according to the present invention. 100...Scanner, 101...A/D converter, 102...Register matrix, 102A~
102I... Register, 103, 104... 1 line buffer, 105... Adder, 106... Judgment circuit, 106B, 106D, 106F, 106
H...Comparator, 107...Conversion ROM, 1
08...Selector, 109...OR gate.

Claims (1)

[Scope of Claims] 1. As a pre-processing for image binarization processing, regarding the density level E of the pixel of interest, referring to the density levels B, H, D, and F of each adjacent pixel on the upper, lower, left, and right sides of the pixel of interest, E' = αE + β (B + H + D + F) However, α and β are correction degree coefficients. Calculate the difference from each of the density levels B, H, D, and F, and if all of the differences are less than a predetermined threshold, select the density level E before correction, and if even one of the differences is greater than the predetermined threshold, select the density level E before correction. Concentration level after correction, if present
An image sharpening processing method characterized in that E′ is selected and subjected to binarization processing.