JPH01143564A - Picture processing unit - Google Patents

Abstract

PURPOSE:To improve the picture quality and processing efficiency by identifying a binary picture part and a non-binary picture part from picture information based on the information of picture and density and identifying the binary picture part and the non-binary picture part from the non-binary picture part with a different method. CONSTITUTION:A line buffer 1 stores tentatively a picture signal 11 in 8-bit per one picture element to be read, gives the result simultaneously to the 1st, 2nd and 3rd identification circuits 2, 3, 4 synchronizing with the clock and gives it to a comparator circuit 9 via a delay circuit 10. The circuits 2, 3, 4 apply identification in the unit of picture elements to the picture signal and give the resulting identification signals 12, 13, 14 to a selection circuit 8. The circuit 8 receives threshold value signals 15a, 15b, 15c from the 1st, 2nd and 3rd threshold value generating circuits 5, 6, 7, selects an optimum threshold value signal based on the identification signal and outputs it to the circuit 9. The circuit 9 compares it with a signal via the circuit 10 and outputs an output signal 18 with a level '1' when the selected signal is larger and outputs an output picture signal 18 with a level '0' when the selected signal is smaller.

Description

Detailed Description of the Invention [Objective of the Invention] (Industrial Field of Application) The present invention is directed to binary images such as character strokes Va mixed in document images and non-binarized images such as photographic images, that is, shading The present invention relates to an image processing device that processes images so as to simultaneously satisfy the resolution of characters and the gradation of photographs.

(Prior Art) In an image processing device capable of image processing not only text but also document images in which text and gray scale images such as photographs are mixed, text and line drawings read by a reading device such as a scanner are processed. Image information with contrast is simply binarized using a fixed threshold value, and image information with gradations such as photographs is binarized using pseudo gradation means such as a dither method.

More specifically, in this image processing device, when the read image information is simply binarized using a fixed threshold value, the resolution is preserved in the text and line image areas, so no image quality deterioration occurs, but in the photo image area, the image quality is not degraded. Because the gradation is not preserved, the resulting image is degraded in image quality.Furthermore, when the read image information is gradated using systematic dithering, etc., the gradation is preserved in the photographic image area, resulting in image quality degradation. However, in the area of characters and line images, the resolution decreases, so in order to prevent the image quality from deteriorating, binary conversion using a fixed darkness value is performed. Two types of binarization means are used.

That is, by performing a single binarization process on the read image information, it is impossible to simultaneously satisfy the image quality of each region having different characteristics, such as a character image and a photographic image. This also applies to various types of image processing; for example, if processing is not performed that matches the characteristics of the image, the image quality may deteriorate when enlarging or reducing a binarized image, or when encoding processing does not match the characteristics of the image. If processing is not performed using a compression method, the data will be compressed inefficiently. Therefore, it is essential to divide image information into regions according to the characteristics of the image and perform adaptive processing on each region. There is.

Therefore, conventionally, as a method that satisfies both the resolution of the character part and the gradation of the photograph part, the maximum density difference ΔD la It means the image No. 1 level and is different from the commonly used "density".Therefore, hereinafter, unless otherwise specified, density will be used in this meaning), and this maximum density difference △D max is compared with the judgment threshold 1゛h. do,
Divide into a text and line image area and a photo image area,
A method has been proposed in which the binarization process is switched according to the characteristics of each image region (for example, Japanese Patent Laid-Open No. 58-3374).

(Problems to be Solved by the Invention) - In the conventional image processing method mentioned above, low-contrast character areas, such as blurred characters such as pencil writing, are determined to be photographic areas because the maximum density difference is small. , the resolution of the character area is significantly impaired, and the large character area with a wide line width is determined to be a photographic area because the maximum density difference is small, resulting in the problem that the image density decreases due to the occurrence of gaps within the line width. be.

More specifically, as shown in FIG. 11, the document P has an area A of contrasting characters and line images, an area B of photographic images with gentle density changes, and a low-contrast character area such as faded characters. In the case where the image information is composed of an area C and an area with large line capital letters having a wide line width, a typical signal level of image information in each area of the document P is as shown in FIG. Now, the dynamic range of image density is set to 8.
Bit (0 to 255. 0 to ff in hexadecimal), each of the image areas A, B, C, and D has 16 pixels within a predetermined range within the image plane, for example, 16 pixels consisting of a 4 x 4 matrix. When determining the maximum density difference ΔDnax within the range of , the maximum density difference ΔDnax in area A = dd~
ΔDnax in f f (hex), Bm, Ll=
10-40 (hex), ΔDIIa in C region
X = 10~40 (hex), ΔD in D area
nax=O~5 (hex).

Therefore, the determination threshold Th for distinguishing between text and photographic areas is set to 8.
0 (hex), the following judgment condition: ΔDmax
>Th...Text portion ΔD l1ax≦Th...When each image area is identified by the photo portion, area A is identified as the text portion, and areas B, C, and D are identified as the photo portion. As a result, simple binarization processing using a fixed darkness value is performed in the A area, and dither processing is performed in the B, C, and Dlj'i areas, so that the A and B areas each adjust the resolution of the text and the gradation of the photo. However, in the C and Dfn regions, processing is performed that preserves the gradation level by aS, so the resolution of characters in the C region is significantly impaired, and in the D region, the image density decreases due to blanking. In addition, in order to accurately identify areas C and D, we set the determination threshold to 30 (hex) or 10 (hex).
x), the B area will be identified as a text area, resulting in a photographic image in which the gradation level will not be preserved. Since the B area, the C area of the low-contrast text area, and the D area of the line capital character area cannot be divided into accurate image areas, the image is divided into each area so that the resolution of the text area and the gradation of the photo area are simultaneously satisfied. There is a problem in that it is not possible to adaptively perform image processing according to the characteristics of the image.

The present invention has been made in view of the above, and its RL purpose is to combine image information in which a text portion and a photo portion are mixed, with the resolution of the text portion and the ll of the photo portion! An object of the present invention is to provide an image processing device that can adaptively perform image processing according to the characteristics of an image for each image area so as to simultaneously satisfy the following.

[Configuration 1 of the Invention (Means for Solving the Problems) In order to solve the above problems, the ii! of the present invention! The i-image processing device includes a first identification means for identifying a binary image part and a non-binary image part from the image information based on information related to the image density of the image information; a second identification means for identifying the binary image part and the non-binary image part in the image information using a method different from the first identification means for the image information of the identified non-binary image part; The gist is that.

(Function) The image processing device of the present invention identifies a binary image portion and a non-binary image portion from image information based on information related to image density, and identifies the binary image portion and non-binary image portion from the image information of the non-binary image portion. The binary image portion and the non-binary image portion are identified using a different method.

(Example) Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 1 is a block diagram showing the overall configuration of an image processing apparatus according to an embodiment of the present invention. The image processing device shown in the same figure is a scanner that is a reading means (not shown).
An image signal 11 of 8 bits per pixel is temporarily stored in a line buffer 1, and the image signal stored in the line buffer 1 is sent to a first identification circuit 2 and a second identification circuit 3 in synchronization with a clock signal (not shown). and a third identification circuit 4 at the same time, and a delay circuit 1 consisting of a delay memory etc.
0 to the comparator circuit 9. 1st, 2nd. The third identification circuits 2, 3.4 perform pixel-by-pixel identification on the image signal, and the first . Second.
The third identification signal 12°13.14 is supplied to the selection circuit 8 as a control signal. This selection circuit 8 is connected to the first threshold generation circuit 5. Second threshold generation circuit 6. The first threshold signal 15a and the second threshold signal 15 are output from the third threshold generation circuit 7, respectively.
b, the third threshold signal 15c is input, and the third threshold signal 15c is input. Second
．． Based on the third identification signal, the optimum threshold signal among the first, second, and third threshold signals 15a, 15b, and 15c is selected and output as the selected threshold signal 16 to the comparator circuit 9. The comparison circuit 9 uses this selection threshold signal 16 and the delay circuit 1.
The output image signal 18 of the image of the pixel (rl when the No. 5 level is greater than the selection threshold signal 16) is outputted, and the image signal level of the pixel is selected. When the threshold signal 16 or less, an output signal 18 of "0" is output.

Next, the first. Second. Details of the third identification circuit 2.3.4 will be explained with reference to the block diagram of FIG. 2.

The first identification circuit 2 is a circuit that identifies whether the image signal is a text portion or a non-text portion, that is, a photo portion. It has a value minimum value detection circuit 20, and the first maximum value minimum value detection circuit 20 detects 4X4=16 in the image signal.
The maximum image density D l1aX and the minimum image density D+xin are detected from among the image densities within a window made up of pixels. The detected maximum image density DlaX and minimum image density Dnin are supplied to the first subtraction circuit 21,
The first subtraction circuit 21 calculates the difference between the two image densities expressed by the following equation, that is, the maximum density difference ΔlaX within the window.

ΔDmax = Dnax −Dln -----...
-(1) This maximum density difference ΔD 1as is supplied to the first comparison circuit 22 and compared with a first determination threshold Thl stored in advance in a register or the like (not shown).

As a result of this comparison, the first comparison circuit 22 determines that the image signal is a character pixel and determines that the image signal is a character pixel when the maximum density difference ΔD la
” level, and when the maximum density difference ΔD lax is less than the first determination threshold Thl, it is determined that the pixel is a non-text pixel, that is, a photographic pixel, and the signal is set to “0”.
A first level identification signal 12 is output.

ΔDIlax ＞'l゛h 1 ・・・Character pixel △
r＞min≦Thl...photo pixel, that is,
The first identification circuit 2 identifies the pixel as a character pixel when the maximum density difference ΔD 1laX, which is the difference between the maximum image density DlaX and the minimum image density D1n, is larger than the first determination threshold 'r'ht, and identifies the pixel as a character pixel. When the determination amount of 1 is less than 1Th1, the pixel is identified as a photographic pixel.

Next, the second identification circuit v@3 is a circuit for identifying a line capital character portion, and has an average value calculation circuit 30 to which the image signal from the line buffer 1 is supplied, and this average value calculation circuit 30 is configured to: The average image density Da of the image density of a window consisting of 4×4=16 pixels is calculated. This average surface tIi concentration Da
is supplied to the second comparison circuit 31, and is compared with a second determination threshold h2 stored in advance in a register (not shown) or the like.

As a result of this comparison, when the average image density Da is larger than the second judgment threshold Th2 as shown below, the second comparison circuit 31 identifies the pixel as a bold character pixel and performs a second discrimination at the "1" level. A signal 13 is output, and when the average image density Da is less than the second determination threshold Th2, it is identified as a non-thick character pixel and r□.

A second level identification signal 13 is output.

Da>Th2... Line thick character pixel Da≦Th2... Non-line capital character pixel In other words, the second identification circuit 3 identifies it as a line thick character pixel when the average image density Da is larger than the second determination lA11ffTh2. However, when the second determination interval is less than fifnTh2, the pixel is identified as a non-linear capital character pixel.

Furthermore, the third identification circuit 4 is a circuit for identifying low-contrast character areas such as faded character areas, and is a circuit for identifying low-contrast character areas such as faded character areas.
It has a smoothing northern vP 140 to which an image signal is supplied from and an edge emphasis circuit 41 using a Laplacian filter. The smoothing circuit 40 supplies the image signal from which noise components have been removed to the edge emphasis circuit 41.
outputs an edge-enhanced image signal 47 in which edges are emphasized. This edge-enhanced image signal 47 is supplied to a second maximum value/minimum value detection circuit 42 . The second maximum fIifi small value detection circuit 42 detects the maximum image density DIIaX and the minimum image density Dnin within the window based on the edge-enhanced image signal 47. This maximum image density D wax
and the minimum image density D min are supplied to the second subtraction circuit 43, and the second subtraction circuit 43 calculates the maximum density difference ΔD l1aX within the window. The maximum density difference ΔD iax calculated by the second subtraction circuit 43 is supplied to the division circuit 44, and the average image density D within the window is supplied from the average value calculation circuit 30 as shown in the following equation.
The normalized maximum density difference ΔDn is calculated by dividing by a.

ΔDn=ΔDIIaX/Da ・・・・・・・・・
(2) This normalized maximum density difference ΔDn is supplied to the third comparison circuit 45 and compared with the third determination threshold Th3. As a result of this comparison, the third comparison circuit 45 identifies the pixel as a low contrast character pixel when the normalized maximum density difference ΔDn is larger than the third determination threshold value 1゛h3 as shown below, and identifies it as a low contrast character pixel. ” outputs the third identification signal 14 of the level,
When the normalized maximum density difference ΔDn is less than or equal to the third determination threshold Th3, the pixel is identified as a non-low contrast character pixel;
A third identification signal 14 of "0" level is output.

ΔD n >'I' h 3...Low contrast character pixel ΔDn≦Th3...Non-low contrast character pixel, that is, the third identification circuit 4 detects the maximum density difference ΔD 1laX
Normalized maximum density difference ΔD, which is normalized by the average image density Da
When n is larger than the third determination threshold 1Th3, the pixel is identified as a low-contrast character pixel, and when it is equal to or less than the third determination threshold Th3, it is identified as a non-low-contrast character pixel.

Next, a description will be given of the binarization processing of each pixel that has been identified as a character pixel, photographic pixel, thick line character pixel, low contrast character pixel, etc. by each of the identification means 2, 3.4 as described above.

In FIG. 2, the first maximum &f of the first discrimination circuit 2
A dynamic darkness value calculation circuit 3 to which the maximum image density DllaX and the minimum image density Dmin are input from the minimum value detection circuit 20.
3, and the output of this dynamic threshold calculation circuit 33 is connected to the first threshold generation circuit 5. When the pixel is determined to be a character pixel or a low-contrast character pixel, the dynamic darkness value calculation circuit 33 calculates a threshold for simple binarization of the pixel, that is, a binarization threshold nh.

That is, the dynamic darkness value calculation circuit 33 calculates the binarization threshold nh from the maximum image density D l1ax and the minimum image m density DIin within the window detected by the first maximum value minimum value detection circuit 20 according to the following formula. .

Bh = (I) nax + D1n)/2 ... self-
- - (3) This binarization threshold oh is supplied to the first threshold generation circuit 5, and the first dark value generation circuit 5 generates the first darkness value based on this binarization threshold Bh. The signal 15a is output to the selection circuit 8.

Further, the second threshold generation circuit 6 determines a binarization threshold for the photographic area and outputs it to the selection circuit 8 as a second threshold signal 15b. The third threshold generation circuit 7 determines the binarization threshold for the line capital character portion, and outputs the third threshold (ili 1B No. 15c) to the selection circuit 8.

The selection circuit 8 selects the first. Second. Third identification signal 12.13
．． 14 as the control signal. Second. Third threshold signal 1
From 5a, 15b, and 15c, the selection threshold t is determined under the following conditions.
Determine g no. 16.

Li Sarizawaokishōku Tatedan 1, first identification signal 12 When the parameter is 1゛° First dark value signal 15a (when distinguishing from text) (Gakujun binary dark value) 2, second identification signal 12 =''゛O゛°, and when the second identification signal 13 is "1", the third threshold signal 15c (identified as a line capital letter) (darkness value for line person characters) 3, the first identification signal 12.1lall Then, the second identification signal 13="O"
Then, when the third identification signal 14="1", the first dark value signal 15
a (Identified as low contrast character) (Binarization [) 4, first identification signal 12・'°0゛, second identification signal 13=
When “O” and third identification signal 14 = “O” Second threshold signal 15b (identification with photo) (Threshold for photo)
The selection circuit 8 selects the selected dark value signal 1 determined as described above.
6 is compared with the image signal 17 of the pixel supplied via the delay circuit 10, and an output image signal 18 is output. At the same time, it is possible to perform image processing that satisfies the requirements for photographic sections while preserving the resolution of characters including sections.

Next, the first. Second. The detailed circuit of the third identification circuit 2, 3.4 will be explained.

FIG. 3 is a detailed circuit block diagram of the first identification circuit 2, and FIG. 4 is its timing diagram.

In FIG. 3, selector 200. Counter 207, comparators 201-204. Comparators 205 and 206
The image data of 8 bits per pixel supplied from the line buffer 1 is read clock CLK in units of 4 pixels.
comparators 201 and 20 via selector 200 in synchronization with
2, 203, and 204. in particular,
For example, the 166 pixels shown in FIG.
Each of the four pixels in the four columns of column +1 and column J+2 is connected to comparator 2.
01,202,203.204.

A counter 207 connected to the selector 200 counts the read clock CLK and outputs an output signal SE shown in FIG.
0 and SE1 are supplied to the selector 200. selector 20
0 is the output port A, B for input data I supplied from the line buffer I in response to the output signals SEO, SEI of the counter 207.

Output to one of ports C and D. In FIG. 4, CLK is the read clock CLK, ■ is the input of the selector 200, A, B, C, and D are the outputs of the selector 200, and SEO and SEI are the outputs of the counter 207.

Comparators 201, 202, 203.204 detect maximum densities 211, 212, 213.214 and minimum densities 221, 222, 223.224 of the four pixels in each column, and each maximum density is supplied to a comparator 205; Each minimum concentration is a comparator 20
Supply to 6. Comparator 205 compares each maximum concentration,
The maximum image density Dnax among them is detected, and the comparator 206 compares each minimum density to detect the minimum image density Dmin among them.

The maximum image density D IIaX and the minimum image density 11in output from the comparators 205 and 206, respectively, are supplied to the first subtraction circuit 21 to calculate the maximum density difference ΔD.
l1aX is calculated, and this maximum density difference ΔDIlax is compared with the first judgment amount value 1゛h1 in the first comparator v+22, and the first identification signal 12 is output as described above.

FIG. 5 is a detailed circuit block diagram of the second identification circuit 3, and FIG. 6 is its timing diagram.

In FIG. 5, a selector 300, a force controller 307,
The adders 301 to 305 and the divider 306 constitute the average value calculation circuit 30, and the selector 30
0 and counter 307 have the same structure and function as selector 200 and counter 207 in FIG.

Adders 301, 302, 303, and 304 are four-manual adders that output the result of addition of four pieces of data, and these outputs are added by adder 305. Note that the adders 301 to 30
4 has an input of 8 bits each and an output of 10 bits, and the adder 305 has an input of 10 bits each and an output of 12 bits.

In the timing diagram of FIG. 6, CLK is a read clock, I, A, B, C, and D are input/output data of the selector 300, and SEO and SEL are the outputs of the counter 307.

Here, if the relevant pixel is (I, J) and the relevant window is as shown in FIG.
Image data of four pixels from the −1> row to the (I+2) row is supplied to the adder 301. Similarly, at the second clock (J)
The four pixels in the column are sent to the adder 302 at the third clock (J+1
) column are sent to the adder 303, and at the fourth clock (J+
2) The four pixels of the column are fed to an adder 304, and each adder calculates the sum of the four pixels. 4 4s on the 5th clock
Sum of pixels, i.e. output 3 of adders 301-304
11, 312, 313, and 314 are supplied to the adder 305, and the combination thereof is output from the adder 305. As a result, the adder 305 has calculated a total of 166 pixels in the window shown in FIG. This 166 pixel total is supplied to the divider 306 at the sixth clock;
In 06, the total of 166 pixels is divided by the total number of pixels, 16,
The average image density Da of the window is calculated. This average image density Da is determined by the second comparison circuit 3 at the seventh clock.
1 and is compared with the second determination value Th2.

FIG. 7 is a detailed circuit block diagram of the third identification circuit 4, and FIG. 8 is its timing diagram.

In FIG. 7, smoothing circuits 401 to 406 constitute the smoothing circuit 40, and edge emphasis circuits i? 8411-414
constitutes the edge emphasis circuit 41. Also, the 8th
In the timing diagram shown, CLK is the read clock C
It's LK. DIN is line buffer 1 to smoothing circuit 4
01 to 406 indicate the timing at which the image signal is input. (J-3> of the first clock is shown in FIG.
) column (1-3> row to (1+4) row) are supplied to the smoothing circuits 401 to 406.

SUM is the timing at which the image data smoothed by the smoothing circuits 401 to 406 is supplied to the edge enhancement circuits 411 to 414. (J-2) of the 4th clock is the 10th clock
From the (I-2) row of the (J-2) column shown in the figure, (1+3
) is the timing at which the smoothed six-pixel image data of the row is supplied to the edge emphasis circuits 411 to 414. ED
G is the timing at which the image data edge-enhanced by the edge emphasis circuit is supplied to the second maximum value/minimum value detection circuit 42. (J-1> of the 7th tarok is the edge-enhanced 4-pixel image data of rows (I-1) to (1-1-2) of column (J-1) shown in FIG. This is the timing at which the signal is supplied to the minimum crotch value detection circuit 42.

The edge-enhanced image data is supplied to the second maximum/minimum value detection circuit 42, and the second maximum/minimum value detection circuit 42 calculates the maximum image density Dl1aX and the minimum image density D1n within the window. , second subtraction circuit 4
3, the maximum concentration difference ΔD l1aX is calculated. This maximum concentration difference ΔD max is divided by the average tIA concentration Da in a division circuit 44 to calculate a normalized maximum concentration difference ΔD n. This normalized maximum density difference ΔDn is determined by the third comparison circuit 45.
It is compared with the third determination value Th3, and the third identification signal 14 is output.

In the above embodiment, the reference image range is described as a predetermined 4×4 image area, but the present invention is not limited to this, and can be any image area. Further, the binarization threshold value Oh in the dynamic darkness value processing is not limited to the above-mentioned threshold value, and any value, for example, the average image density Da within a predetermined range can also be used. Furthermore, a positive or negative tolerance value α is added to each value, for example, the binarization threshold Bh is set to Bh
It may be set to 10α.

Further, in this embodiment, the value of the feature amount and determination l for identifying a character image and a photographic image, that is, a binary image and a non-binary image from image information: [Image signal read by a reading means as an I value, In other words, the expression corresponding to the reflectance of the image information is used, but this value is converted to the image density, that is, the logarithm of the reciprocal of the reflectance, or a converted signal that takes into account human visual characteristics is used. may be identified.

[Effects of the Invention] As explained above, according to the present invention, low:1 contrast character parts such as blurred characters such as pencil writing, which were conventionally determined to be photographic areas and whose resolution was significantly impaired, and line width Appropriate image processing can be performed while preserving the resolution of line character portions in which white spots have occurred, thereby improving image quality and processing efficiency in various image processing.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an image processing apparatus according to an embodiment of the present invention, FIG. 2 is a detailed block diagram of the image processing apparatus of FIG. 1, and FIG. 3 is a block diagram of an image processing apparatus used in the image processing apparatus of FIG. A detailed block diagram of the first identification circuit, FIG. 4 is a timing diagram of the circuit of FIG. 3, and FIG. 5 is a detailed block diagram of the second identification circuit used in the image processing device of FIG. 1. , FIG. 6 is a timing diagram of the circuit of FIG. 5, FIG. 7 is a detailed block diagram of the third identification circuit used in the image processing device of FIG. 1, and FIG. 8 is a timing diagram of the circuit of FIG.
The timing diagram of the circuit shown in FIG. FIG. 10 is a diagram showing the window area referred to by the third identification circuit; FIG. 10 is a diagram showing the identification area used by the third identification circuit;
1 is a diagram showing each image area, and FIG. 12 is a schematic diagram showing the image (IA signal level) in each image area. 2... first identification circuit, 3... second identification circuit, 4
... Third identification circuit, 5... First threshold ffi generation circuit, 6... Second threshold generation circuit, 7... Third threshold generation circuit, 8... Selection circuit, 9... Comparison circuit. Agent f+-r11! ±Miyoshi 1 dumbass

Claims (4)

[Claims] (1) a first identification means for identifying a binary image part and a non-binary image part from the image information based on information related to the image density of the image information; A second identification means for identifying the image information of the non-binary image part between the binary image part and the non-binary image part in the image information using a method different from the first identification means. Image processing device. (2) The first identification means includes a character that identifies a text image portion and a photographic image portion based on a comparison between a maximum image density difference within a predetermined range of the image information and a predetermined determination image density difference. 2. The image processing apparatus according to claim 1, further comprising a means for identifying both image portions. (3) The second identification means compares the average image density of the image information identified as a photographic image part by the character image part identification means with a predetermined determination average image density, and 2. The image processing apparatus according to claim 1, further comprising a thick line character image portion identifying means for identifying a bold character image portion. (4) The second identification means includes a smoothing means for smoothing the image information identified as a non-thick character image portion by the thick line character image means, and a smoothed image signal from the smoothing means. an edge emphasis means for enhancing edges; a standardization means for normalizing the maximum image density difference within a predetermined range of the edge-enhanced image signal from the edge emphasis means by a predetermined average dual image density; and a low contrast character image portion identifying means for identifying a low contrast character image portion and a non-low contrast character image portion based on a comparison between the maximum image density difference and a predetermined image density difference. An image processing device according to claim 3.