JP3222570B2 - Image area identification device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital image processing apparatus for processing a digital image and recording / displaying the digital image. Thus, the present invention relates to an image area discriminating apparatus for discriminating an important area of gradation.

[0002]

2. Description of the Related Art In recent years, digital image processing apparatuses that handle document images as digital signals, such as facsimile machines, document files, and digital copying machines, have been increasing. When an image is treated as a digital signal, there are many advantages such as various editing / correction processes and electronic recording and transmission being easy. Conventionally, such digital image processing apparatuses mainly deal with monochrome binary images such as characters and line drawings, but recently there is an increasing demand to handle images having mixed gradation images.

[0003] One problem in recording an image in which characters and gradation images are mixed is hard copy. As a digital image recording method, a fusion heat transfer method and an electrophotographic method are often used. These methods usually only have the ability to represent on the order of two to several levels of density per recorded pixel. Therefore, a method such as an area modulation method is used to express a gradation image. The area modulation method is a method of expressing gradation by a combination of a plurality of pixels, and a systematic dither method or the like is famous.

The systematic dither method is a recording method in which gradation and resolution are compatible in principle. However, a character / line image having a sharp edge is reproduced by the original reading system, and moire noise is generated in a halftone original. There are drawbacks such as. In a document image, information of characters and line drawings is extremely important, and in a document image, halftone printing is often used for expressing a gradation image, so these drawbacks are fatal.

[0005] To eliminate these drawbacks, an image area discrimination method is used to perform image recording in which resolution and gradation are compatible. Image area identification is a method of identifying an input image to be recorded into an area where gradation is important, such as a photograph, and an area where resolution is important, such as characters and line drawings. By switching the recording process, an image can be recorded with good resolution and gradation.

As a conventional image area discriminating method, a method utilizing a difference in local density change between a gradation area and a character area,
There is known a method using a local pattern difference. As an example of the former, for example, JP-A-58-3374 discloses that
A method in which an image is divided into small blocks, the difference between the maximum density and the minimum density in each block is determined, and if the difference is larger than a threshold value, the block is identified as a character image area, and if the difference is smaller, the block is identified as a gradation image area. Is disclosed. With this method, it is possible to perform high-precision discrimination in the case of only continuous tone images and characters such as photographs, but the problem is that the discrimination accuracy is extremely poor in areas where there are many local density changes such as halftone dots. was there.

On the other hand, Japanese Patent Application Laid-Open No. 60-204 discloses an example of the latter.
No. 177 discloses a method of applying an Laplacian filter to an image, binarizing the image with an appropriate threshold value, and performing image area identification based on the shape of the binarized pattern, for example, a pattern of 4 × 4 pixels. I have. According to this method, halftone dots can also be identified. However, since the binarization pattern of the image is used, there is a problem that the effect is small unless a threshold value for binarization is properly selected, and a problem that the image is weak to noise exceeding the threshold value. Further, in the case of the table reference method, there is a problem that the reference area is limited to about 4 × 4 pixels due to the restriction of the table size. For this reason, sufficient identification accuracy could not be obtained in the case of low density characters and coarse halftone images.

Further, a method of combining the two image area identification methods described above or utilizing the characteristic that the image area is constant within a certain wide area to perform correction from the identification result of surrounding pixels is also used. Attempts have been made to improve the discrimination accuracy, but this has not been sufficient.

[0009]

As described above, of the conventional image area discrimination methods, the method using the grayscale distribution cannot discriminate the halftone image, and the method using the pattern shape does not allow the halftone image. However, there is a problem that it is susceptible to noise due to binarization, and there is a problem that it is not possible to refer to a wide area due to the limitation of the amount of calculation for pattern identification and the stability of identification is low. High identification accuracy could not be obtained.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an image area discriminating apparatus for discriminating, with high accuracy, an area where resolution is important, such as a character, from a gradation area by effectively utilizing pattern information of a wide area of an image. And

[0011]

In order to solve the above-mentioned problems, an image area identifying apparatus according to the present invention comprises a gradient vector calculating means for calculating a gradient vector of an input image, and a gradient calculated by the gradient vector calculating means. Declination doubling means for doubling the declination of a vector, averaging means for averaging the gradient vector obtained by the declination doubling means for each minute area of the input image, and averaging means And an image area identification signal generating means for generating an image area identification signal for the input image from the gradient vector averaged by the above.

Here, the image area identification signal generating means calculates, for example, an absolute value of the averaged gradient vector, and further averages the absolute value in an area larger than the minute area to generate an image area identification signal. Configuration. In addition, the present invention is applied to the image processing apparatus for recording or displaying an image using the error diffusion method using the above-described image area identifying apparatus. in this case,
The image area identification signal generation means generates the intensity and direction of the averaged gradient vector as an image area identification signal. Then, according to this image area identification signal, an error diffusion coefficient for performing error diffusion processing on an input image is switched.

[0013]

By calculating the gradient vector of the input image, an edge-like area of the input image is detected. If the argument of the gradient vector is doubled, the gradient vectors on both sides of the line in the edge-like area in the input image become vectors in the same direction. If the gradient vector after declination doubling is averaged in a minute area, the noise is reduced by the averaging effect, and the direction of the edge is varied in a narrow area such as a halftone dot, so that the direction is changed. Are canceled, and the gradient vector of the linear portion having the same direction is a norm, that is, a vector having a large absolute value.

Therefore, by calculating the absolute value of the averaged gradient vector and comparing the absolute value with an appropriate threshold value, an area where the resolution is important, such as a character, which is constituted more by straight edges, is important. Or a gradation area including a halftone image or the like is accurately determined. Further, if the averaged gradient vector intensity and direction are generated as an image area identification signal, and the error diffusion coefficient in the error diffusion processing is switched according to the image area identification signal, gradation expression with less image quality degradation can be achieved. .

[0015]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a first embodiment in which the present invention is applied to a digital image copying apparatus. This image copying apparatus includes an image input unit 1, an image area identification unit 2, an adaptive processing unit 3, a binarization processing unit 4, and an image recording unit 5. The image area identification device according to the present invention includes an image area identification unit 2.
Applied to

The image input section 1 reads an original image as an electric signal corresponding to the brightness of each pixel. More specifically, the image input unit 1 scans a document image by, for example, a CCD array sensor, reads the image as a raster scan time-series signal, and converts the time-series signal into a digital signal by an A / D converter. After that, a shading correction process is performed. This process is a process for correcting unevenness in sensitivity of each element of the CCD line sensor and uneven illuminance on the original, thereby standardizing the image signals to be 0 and 1 when the original image is black and white, respectively. Then, an image signal is output. For the specific configuration of this part, for example,
This is described in detail in JP-A-61-71764.

An image signal obtained from the image input unit 1 is input to an image area identification unit 2 and an adaptive processing unit 3. The image area identification unit 2 detects the characteristics of the input image signal and determines the attributes of the local area, that is, whether the area is an area where the resolution is important, such as a character or a line drawing, or the gradation of the image, such as a gradation image. Is an important area. Then, the adaptive processing unit 3 performs processing according to the area identification result of the image area identification unit 2.

Next, the configuration of the image area identifying section 2 according to the present invention will be described in detail. The image area identification unit 2 identifies each part on the image as to whether or not it is a continuous edge. In the edge portion of a character / line image or gradation image, resolution is more important for image quality than gradation. In particular, edges that are continuous on a straight line appearing at the boundary of a character or a line or a contour portion of a gradation image have a visually high impression degree, and the image quality is largely degraded due to a decrease in resolution.

On the other hand, in a portion where the density change of the gradation image is gentle, the gradation is more important than the resolution. However, even at the edge portion, a finely formed edge such as a fine outline or a halftone dot on a gradation image has a small visual impression, and the image quality due to the reduction in resolution is small. Rather, the sense of obstruction is smaller when the edges of the halftone dots disappear. From this perspective,
The image area identification unit 2 identifies whether the resolution is important or the gradation is important, depending on whether the edges in the input image are continuous edges.

FIG. 3 shows a specific example of the configuration of the image area identification unit 2. The image area identification unit includes gradient filters 11 and 12 for calculating a gradient vector, a look-up table 13 for doubling the argument of the gradient vector, and a first table for averaging the gradient vector in a minute area. Averaging circuit 14,
15. Lookup table 16 for calculating absolute value and second averaging circuit 1 for finally obtaining image area identification signal
Consists of seven.

An image signal from the image input unit 1 shown in FIG. 1 is input to gradient filters 11 and 12, and a gradient vector Vx in the x direction and a gradient vector Vy in the y direction are calculated by the following equations (1), (2) and (3).
Is calculated.

Vx (x, y) = P (x + 1, y + 1) + P (x + 1, y) -P (x, y + 1) -P (x, y) (1) Vy (x , y) = P (x + 1, y + 1) + P (x, y + 1) -P (x + 1, y) -P (x, y) (2) | V | = (Vx 2 + Vy 2 ) (3)

FIGS. 2A and 2B show examples of kernels of the gradient filters 11 and 12. FIG. Where P (x, y) and Vx (x, y)
And Vy (x, y) represent the image signal at point (x, y) and the x and y components of the gradient vector, respectively.

Next, using the first look-up table 13, the vectors Ux, Uy obtained by doubling the argument θ from the gradient vectors Vx, Vy while maintaining their absolute values are given by the following equations.
(4) Determined according to (5) (6).

Ux = | V | · cos2θ (4) Uy = | V | · sin2θ (5) θ = arctan (Vy / Vx) (6)

The look-up table 13 is, for example, a formula
(4) ROM storing values calculated in advance according to (5) and (6)
It is realized by.

Next, the argument doubling vector obtained by the look-up table 13 is used for the first averaging circuits 14 and 15.
Is averaged in the minute area S1 near the pixel of interest,
The vectors shown in the following equations (7) and (8) are U'x and U'y. As the minute region S1, for example, 5 around the target pixel to be referred to
A rectangular area of about × 5 pixels is used.

U′x = ΣUx / Σ (7) U′y = ΣUy / Σ (8)

Next, the vectors U'x and U'y averaged by the averaging circuits 14 and 15 are inputted to a second look-up table 16, and the absolute value Q is calculated as shown in the following equation (9). Is required. This lookup table 16 is also realized by a ROM, like the lookup table 13.

Q = | U ′ | (9)

Further, the absolute value Q is averaged by the second averaging circuit 17 in the area S2 near the pixel of interest to obtain Q '. As the area S2, an area wider than the minute area S1 in which the gradient vectors are averaged, for example, a rectangular area of about 9 × 9 pixels is used.

Q ′ = ΣQ / Σ (10)

The Q 'obtained by the averaging circuit 17 in this manner
(Referred to as edge degree) is output as an image area identification signal.
Further, the image area identification signal is subjected to a threshold processing with a further appropriate threshold as needed.

Next, a description will be given of how characters and edges of an input image are identified by the image area identifying section 2 with reference to FIGS. An example is given in which a line portion as shown in FIG. 4A and a halftone dot portion as shown in FIG. The gradient vector V, which is the gradient of the input image, has a magnitude corresponding to the inclination of the edge of the image.
The direction is parallel to the normal direction of the edge.

FIGS. 5 (a) and 5 (b) schematically show the gradient vectors for the images of FIGS. 4 (a) and 4 (b). The direction and length of the arrow in FIG. 5 indicate the direction and magnitude of the gradient vector, respectively. As shown in FIG. 5, the direction of the gradient vector is uniform at the edge of the character portion in FIG. 5A, and the direction varies in the halftone dot portion in FIG.

Only the declination is 2 in the gradient vector V of FIGS.
The doubled vector U is shown in FIGS. 6 (a) and 6 (b). Here, the line to the right in the horizontal direction is used as the reference for the declination. Due to the doubling of the deflection angle, the direction is shifted by 180 ° in FIGS. 4 (a) and 4 (b).
That is, the gradient vectors whose directions are opposite to each other are vectors in the same direction in FIGS. 5 (a) and 5 (b), and the direction is 9 in FIGS.
The gradient vectors shifted by 0 ° are vectors in directions opposite to each other in FIGS. 5 (a) and 5 (b). As a result, for example, the gradient vectors at both edges of the line in FIG. 6A are vectors in the same direction.

FIG. 7 shows the result of averaging the gradient vector U obtained by doubling the declination in the neighboring region. By this averaging, the gradient vectors whose directions are not aligned as shown in FIG. 6B are canceled each other, and become a vector having a small absolute value as shown in FIG. 7B. On the other hand, the gradient vectors having the same orientation as shown in FIG. 6A are not canceled out with each other, and become vectors having a large absolute value.

That is, the edge degree Q is large in an area composed of edges having the same gradient vector direction, such as a character or a line drawing, while the gradient vector direction is large in a halftone dot or a complicated contour of a gradation image. Is irregular, Q is 0 or a small value. In addition, the edge degree Q has a small value even in a portion where the gradient is originally small. Here, FIGS. 6 (a) and 6 (b)
If the declination doubling process is not performed as described above, the gradient vectors at the edges on both sides of the thin line whose gradient directions are opposite to each other are canceled out by averaging. The value of Q becomes small, making it difficult to be identified as an edge. By averaging after doubling the deflection angle, such a problem is eliminated.

As described above, the edge degree Q takes a high value at successive edges. However, at the end points and intersections of the lines as shown in FIGS. 8A and 8B, the edges are composed of lines having various directions. Is a low value. It is necessary to identify such a part as a character and express it with high resolution. By performing averaging by the second averaging circuit 17, even such a portion can be identified as a character. One character is composed of a line element part with a high edge degree Q and an intersection or an end point with a low edge degree Q, but the intersection or the end point is isolated, and a line element with a high edge degree always exists around it. I do.
Therefore, the averaging circuit 17 averages the edge degree Q in a region larger than a line element constituting a character, so that an intersection, an end point, and the like also have a large value under the influence of a surrounding region having a high edge degree Q.

In this embodiment, the simple averaging is used as the averaging. However, a weighted averaging that changes the weight coefficient depending on the pixel may be used. Further, the first averaging circuits 14, 1
When the required image area identification accuracy can be obtained with only 5, the second averaging circuit 17 may be omitted, and the edge degree Q may be used as the image area identification result. Returning to FIG. 1, the adaptive processing unit 3 performs a process suitable for each area according to the image area identification signal Q ′ output from the image area identification unit 2. FIG. 9 shows a specific configuration of the adaptive processing unit 3. First, an image signal from the image input unit 1 in FIG. The high-frequency extraction filter 21 has a kernel as shown in FIG. 10, for example. The output of the high-pass extraction filter 21 is
A large positive value is taken at the convex portion of the image density, and a large negative value is taken at the concave portion, and a value close to 0 is taken at the portion where the density change is small and the gradient is uniform.

The output signal of the high-frequency extraction filter 21 is input to the modulation circuit 22 and multiplied by a coefficient corresponding to the edge degree Q ', which is the image area identification result from the image area identification unit 2, and then added to the adder. At 23, it is added to the original image signal. This coefficient takes a large value, for example, about 4 when the edge degree Q 'is high, and takes a small value, for example, about 0 to -1 when the edge degree Q' is low. As a result, the edge is more emphasized in a portion having a high edge degree Q ', such as an edge portion of a character line drawing or a gradation image, and the edge is enhanced as in a halftone image or a gradation image where a density change is gentle. In the portion where the degree Q 'is low, smoothing is performed and edges are suppressed.

The image signal thus adaptively processed by the adaptive processing unit 3 is converted into a binary image signal by the binarization processing unit 4, and then the image is recorded on the paper by the image recording unit 5. In this embodiment, for example, an electrophotographic method is used for this image recording.
In the electrophotographic method, an electrostatic latent image corresponding to an image signal is formed on a photoconductor by modulating a laser beam applied to a photoconductor in accordance with an image signal, and the electrostatic latent image is developed with toner. An image is formed.

In the present embodiment, the binary control of the laser beam on the photosensitive member is switched between 0 and 1 to record the binary level per pixel. For this reason, since multi-valued density cannot be expressed in pixel units, the binarization processing unit 4 performs gradation expression using the tissue dither method. The tissue dither method is a recording signal modulation method that is compatible with both resolution and gradation for ideal character images and gradation images.

However, the tissue dither method has a drawback that if the original image has a halftone structure, moire noise is generated and the image quality is remarkably deteriorated. Although a sharp edge can be expressed with a high resolution, the resolution is reduced when the edge is rounded due to blurring of an image input system or the like. In order to prevent such image quality deterioration, it is necessary to remove a dot structure from a dot image by using a low-pass filter or the like. For characters and edges, high-frequency emphasis processing etc.
Needs to be corrected.

Therefore, in the present embodiment, as shown in FIG. 1, the adaptive processing section 3 performs the above-described adaptive processing before performing the binarization processing in the binarization processing section 4, so that a halftone image Structure is removed and edges are sharpened. Therefore, characters and edges are sharply reproduced, and gradation portions are reproduced smoothly without moiré, so that high-quality image recording can be performed.

In the present embodiment, the linear filter processing is performed by the adaptive processing unit 3 and the characteristics of the filter are switched according to the image area identification result from the image area identification unit 2. However, the present invention is not limited to this. It is not limited. For example, a logical filter such as a median filter may be used, and the size of the filter may be switched according to the identification result. Further, a contrast enhancement function may be applied, and the degree of the enhancement may be switched according to the image area identification result.

Next, a second embodiment in which the present invention is applied to a digital copying machine will be described. FIG. 11 shows the configuration of a digital copying machine according to this embodiment. In this embodiment, as in the first embodiment, after an image is read as a digital signal by the image input unit 31, the edge degree of the digital image signal is detected by the image area identification unit 32, and the edge degree is detected. Output as result.

That is, unlike the first embodiment, the image area discriminating section 32 has the averaging circuits 14, 15 as shown in FIG.
, The strength and direction of the averaged gradient vector from the edge strength signal U'a
And an edge direction signal U′θ. This is shown in the following equation. Here, U'x and U'y are the averages of the declination doubling gradient vectors, respectively, as shown in equations (7) and (8).

U′a = (U′x ² + U′y ² ) (11) U′θ = arctan (U′y / U′x) (12)

The image signal read by the image input unit 31 is
After being also input to the binarization processing unit 33 and converted into a binarized signal by an error diffusion method, it is sent to the image recording unit 34 and recorded on the paper.

In this embodiment, a binary electrophotographic system is used as in the first embodiment, but the binarization processing section 33 performs gradation expression using an error diffusion system. The error diffusion method is a method of distributing and adding a quantization error at the time of binarization to nearby pixels before binarization processing in accordance with a diffusion matrix coefficient. This is known as a gradation recording method that achieves both characteristics.

FIG. 12 shows a typical example of the diffusion matrix. This method preserves the average density in a local area, so moire noise is unlikely to occur, but on the other hand, wavy edges are reproduced in “jaggies” or unique irregular textures in areas with gradual density changes. And other problems.

In this embodiment, the binarization processing section 33 switches the diffusion matrix (error diffusion coefficient) according to the edge direction signal and the edge intensity signal output from the image area identification section 32, thereby deteriorating the image quality, especially The former prevents edge deterioration. FIG. 14 shows the relationship between the edge strength signal, the edge direction signal, and the selected diffusion matrix. If the edge strength is less than a certain threshold,
A standard diffusion matrix is used as shown in FIG. If the edge strength is larger than the threshold value, one is selected from the four diffusion matrices shown in FIGS. 14 (a), (b), (c), (d) and (e) according to the normal direction of the edge.

By selecting such a diffusion matrix, no error is diffused in the normal direction of the edge, and the edge can be reproduced sharply. That is, as in the present embodiment, the direction and the intensity of the continuous edge are detected by the image area identification unit 32, and the binarization processing unit 33 is accordingly detected.
By switching the error diffusion processing in the above, it is possible to realize high-quality recording with particularly sharp edges.

In this embodiment, the example shown in FIG. 12 is used as the diffusion matrix. However, the present invention is not limited to this, and a matrix having a larger size may be used. Further, in this embodiment, a binary electrophotographic system is used for the image recording unit 34, but another recording system such as a thermal transfer system may be used. Further, a recording method that takes, for example, four levels of density per pixel may be used. In this case, the binarization processing unit 3
Although it is necessary to extend the multi-level processing to perform the multi-level processing instead of 3, the multi-level processing in this case may be a multi-level error diffusion method which is an extension of the error diffusion method.

[0056]

As described above, according to the present invention, for an image in which a binary image such as a character or a line image and a gradation image are mixed, a high resolution such as a character region or an edge portion of a gradation image is obtained. A required portion and a region where smooth gradation is required can be accurately distinguished. In particular, halftone dots, thin characters, extremely thin lines, etc., which were low in identification accuracy in the conventional method, can be identified with high accuracy. Further, the sharpness at the edge is improved by the error diffusion method for the part where high resolution is required.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an image copying apparatus according to a first embodiment.

FIG. 2 is a diagram illustrating a kernel of a gradient filter used in an image area identification unit in FIG. 1;

FIG. 3 is a block diagram showing a configuration of an image area identification unit in FIG. 2;

FIG. 4 is a diagram illustrating an example of an image input to an image area identification unit.

FIG. 5 is a diagram showing a gradient vector of the input image of FIG. 4;

FIG. 6 is a diagram showing a vector obtained by doubling the declination of the gradient vector of the input image of FIG. 4;

FIG. 7 is a diagram showing a result of averaging the vectors of FIG. 6;

FIG. 8 is a diagram showing intersections and end points of an image;

FIG. 9 is a block diagram showing a configuration of an adaptive processing unit in FIG. 1;

FIG. 10 is a diagram illustrating a kernel of a high-pass extraction filter in the adaptive processing unit in FIG. 9;

FIG. 11 is a block diagram showing a configuration of a digital copying machine according to a second embodiment.

FIG. 12 is a diagram showing an example of a typical error diffusion matrix.

FIG. 13 is a block diagram showing a configuration of an image area identification unit in FIG. 11;

FIG. 14 is a diagram showing a relationship between an edge strength signal, a direction signal, and a spreading matrix.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Image input part 2 ... Image area discriminating part 3 ... Adaptive processing part 4 ... Binarization processing part 5 ... Image recording part 11, 12 ... Gradient filter 13 ... Look-up table 14, 15 ... Averaging circuit 16 ... Look-up Table 17 ... Averaging circuit 18 ... Lookup table 21 ... High-frequency extraction filter 22 ... Modulation circuit 23 ... Adder 31 ... Image input unit 32 ... Image area identification unit 33 ... Binarization processing unit 34 ... Image recording unit

────────────────────────────────────────────────── ─── Continued from the front page Examiner Minoru Matsunaga (56) References JP-A-50-40039 (JP, A) JP-A 1-271887 (JP, A) JP-A 4-576 (JP, A) ( 58) Surveyed field (Int.Cl. ⁷ , DB name) H04N 1/40-1/409 G06T 7/60

Claims (3)

(57) [Claims] A gradient vector calculating means for calculating a gradient vector of an input image; a declination doubling means for doubling a declination of a gradient vector calculated by the gradient vector calculating means; Averaging means for averaging the gradient vector obtained by the doubling means for each minute area of the input image; and generating an image area identification signal for the input image from the gradient vector averaged by the averaging means. An image area identification device, comprising: an image area identification signal generation unit. 2. The image area identification signal generating means calculates an absolute value of a gradient vector averaged by the averaging means, and further averages the absolute value in an area larger than the minute area to obtain the image. The image area identification apparatus according to claim 1, wherein the apparatus generates an area identification signal. 3. A gradient vector calculating means for calculating a gradient vector of an input image; a declination doubling means for doubling the declination of the gradient vector calculated by the gradient vector calculating means; Averaging means for averaging the gradient vector obtained by the doubling means for each minute area of the input image, and identifying the intensity and direction of the gradient vector averaged by the averaging means for an image area of the input image. An image area identification signal generating means for generating a signal, an error diffusion processing means for performing error diffusion processing on the input image according to a predetermined error diffusion coefficient, and an image area identification signal generated by the image area identification signal generating means. An image processing apparatus comprising: a switching unit that switches an error diffusion coefficient to an error diffusion coefficient.