JP2002281294A - Image processor and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the process time and the process cost for a gradation number recovery process and half-tone image enlarging process and prevent the image quality from deteriorating in these processes. SOLUTION: The image processor comprises a window 202 for inputting a dither image of a value N (>=2), a gradation number recovering/enlarging unit 203 for applying a gradation number recovery process and a half-tone image enlarging process to the dither image in the window 202 to recover and enlarge a half-tone image of a value M (>N) and a memory 205 for saving the enlarged half-tone image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image processing apparatus and method for a dither image.

[0002]

2. Description of the Related Art A conventional technique will be described below.

FIG. 8 is a block diagram showing conventional image processing. In the image processing shown in FIG. 8, the N value (N ≧
The dither image of 2) is restored to a halftone image by the number of gradations, and further enlarged.

In FIG. 8, when an N-value (N ≧ 2) dither image 101 is input to a window 102, a multi-value dither image in the window 102 is converted into an M-value (M> N) halftone image. The image restored to the halftone by the tone restoration unit 103 is input to the window 104.

The halftone image on the window 104 is enlarged by the enlargement unit 105 and input to the window 106, and the pixels in the window 106 are stored in the image memory 107 as the enlarged halftone image.

As described above, in the conventional image processing apparatus, two steps of a process of restoring the number of tones of a halftone image from a dither image and a process of enlarging the halftone image are executed.

In order to restore the number of tones of an N-value (N ≧ 2) dither image of an input image, there is a method of estimating continuous tones based on an average level in a window. Further, as the enlargement processing method, there is a commonly used interpolation method such as a bilinear method and a cubic method.

[0008]

However, in such a conventional technique, since the process of restoring the number of tones of a dither image and the process of enlarging a halftone image are performed in two processing steps, it takes time in software processing. .

Further, the hardware processing has a plurality of windows, so that the circuit scale becomes large, which leads to an increase in cost.

[0010] In terms of image quality, noise is added when the number of gradations is restored, and image quality deterioration such as blurring of an edge portion is increased due to characteristics of enlargement processing.

SUMMARY OF THE INVENTION An object of the present invention is to provide an image processing apparatus and method capable of reducing the processing time and processing cost in the gradation number restoration processing and the halftone image enlargement processing.

It is another object of the present invention to provide an image processing apparatus and method capable of preventing image quality deterioration in tone number restoration processing and halftone image enlargement processing.

[0013]

In order to solve this problem, an image processing apparatus according to the present invention comprises: a window into which an N-value (N ≧ 2) dither image is input; A configuration including a tone number restoring / enlarging unit that performs a tone number restoring process and enlargement process on a halftone image (M> N), and an image memory that stores the enlarged halftone image. It is.

As described above, since the number-of-tones restoring process of the dither image of N value (N ≧ 2) and the halftone image enlarging process for enlarging it are performed at the same time, the number of image memories and windows is reduced. Thus, the processing time and processing cost in the gradation number restoration processing and the halftone image enlargement processing can be reduced. Further, it is possible to prevent image quality deterioration in the gradation number restoration processing and the halftone image enlargement processing.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS According to the first aspect of the present invention, a window into which an N-value (N ≧ 2) dither image is input, and a dither image in the window having an M-value (M> N) An image processing apparatus comprising: a gradation number restoring / enlarging unit that performs a gradation number restoring process and an enlarging process on a halftone image; and an image memory in which the enlarged halftone image is stored. Is reduced, and the processing time and the processing cost in the gradation number restoration processing and the halftone image enlargement processing can be reduced.
Further, there is an effect that it is possible to prevent image quality deterioration in the gradation number restoration processing and the halftone image enlargement processing.

According to a second aspect of the present invention, a dither image having an N value (N ≧ 2) is input to a window, and the dither image in the window is transformed into a halftone image having an M value (M> N). This is an image processing method that performs tone restoration processing and enlargement processing, and stores an enlarged halftone image. The number of image memories and windows is reduced, and processing in tone number restoration processing and halftone image enlargement processing is performed. This has the effect that time and processing costs can be reduced. In addition, there is an effect that it is possible to prevent image quality deterioration in the gradation number restoration processing and the halftone image enlargement processing.

According to a third aspect of the present invention, there is provided an image processing method for performing an enlargement process using a neural network according to the second aspect, wherein the number of image memories and windows is reduced, and This has the effect that the processing time and processing cost in the tone restoration processing and the halftone image enlargement processing can be reduced. Also,
This has an effect that it is possible to prevent image quality deterioration in the gradation number restoration processing and the halftone image enlargement processing.

According to a fourth aspect of the present invention, in the second or third aspect, the neural network is an image processing method in which the neural network is configured in a layered manner, and the number of image memories and windows is reduced. This has the effect that it is possible to reduce the processing time and processing cost in the gradation number restoration processing and the halftone image enlargement processing.
In addition, there is an effect that it is possible to prevent image quality deterioration in the gradation number restoration processing and the halftone image enlargement processing.

Hereinafter, an embodiment of the present invention will be described with reference to FIG.
This will be described with reference to FIG.

FIG. 1 is a block diagram showing an image processing apparatus according to an embodiment of the present invention, and FIG. 2 shows a correspondence relationship between positions of an input window and an output window in image processing according to an embodiment of the present invention. FIG. 3 is an explanatory view showing a state of scanning an input window in image processing according to an embodiment of the present invention. FIG. 4 is a diagram showing a structure of a neural network in image processing according to an embodiment of the present invention. FIG. 5 is a block diagram illustrating a learning process for restoring and enlarging the number of tones of a dither image in image processing according to an embodiment of the present invention, and FIG. 6 is an image according to an embodiment of the present invention. Restoration of dither image in processing
FIG. 7 is a block diagram showing a process for restoring and enlarging the number of tones of a dither image in image processing according to an embodiment of the present invention.

As shown in FIG. 1, in the image processing according to the present embodiment, the dither image of the N value (N ≧ 2) of the input image is restored to a halftone image by the number of tones, and further enlarged. Things.

As shown in FIG. 1, the image processing apparatus according to this embodiment includes a window 202 to which a dither image 201 having N values (N ≧ 2) is input and a dither image 201 in the window 202 having an M value ( (M> N) A gradation number restoration / enlargement unit (gradation number restoration / enlargement unit) 203 that performs the gradation number restoration processing and enlargement processing on the halftone image of M> N) and the result of the restoration / enlargement processing is input. It comprises a window 204 and an image memory 205 for storing pixels in the window 204 as a halftone image having an M value (M> N) after enlargement. Although the window 202 and the window 204 are shown in the present embodiment, they may be common or separate. The processing image of the window and the image will be described later.

As described above, the gradation number restoration processing and the halftone image enlargement processing for enlarging the dither image of the N value (N ≧ 2) are collectively performed in the gradation number restoration / enlargement unit 203. Thus, the number of image memories and windows 204 can be reduced, and the processing time and processing cost in the gradation number restoration processing and the halftone image enlargement processing can be reduced.

Further, it is possible to prevent the image quality from deteriorating in the gradation number restoring process and the halftone image enlarging process.

The correspondence between windows will be described below. In the present embodiment, the magnification is 2 times, but may be 3 times or more.

FIG. 2A shows a state in which a dither image having an input N value (N ≧ 2) is divided by a window. Here, the window is divided into 5 × 5 pixel windows. FIG. 2B shows output pixels of an output halftone image having an M value (M> N) after the processing. Here, 10 × 10 pixels correspond to the window of the input image.

That is, the window of 5 × 5 pixels centered on the center pixel 301 in the window before the processing (FIG. 2A) is the same as the center pixel 302 in the window after the processing (FIG. 2B). This corresponds to four pixels. When the position is shifted to process the next pixel, the center pixel 303 in the window before processing corresponds to the four pixels of the center pixel 304 in the window after processing.

In FIG. 3 showing the scanning of the input window, the pixel 401 is the first window center pixel of the entire input image 402, and the windows move in the order of the input window 403 and the input window 404.

When taking a window on the next line,
The pixel 405 is a central pixel. Hereinafter, one screen scan is repeated for both rows and columns until a window can be taken.

FIG. 4 shows the structure of the neural network.

Here, one pixel of the input / output image corresponds to one unit. Therefore, the first layer 501 is an input layer (5 × 5 = 25 units), the second layer 502 is an intermediate layer (25 units or more), and the third layer 503 is an output layer (2 units).
× 2 = 4 units).

The first layer 501 as an input layer and the third layer 503 as an output layer are normalized so as to take values between 0 and 1, and each unit is a perceptron. Connections between units do not exist between units in the same layer,
01-all units between the second layer 502, the second layer 50
The connection is made by a combination of all units between the second and third layers 503. As an example of the learning algorithm, a back propagation method is used. Since the description of this method is publicly known, the description is omitted here.

Next, a method of learning using a learning algorithm this time (hereinafter, referred to as a “learning method”) will be described.

N-value (N ≧ 2) dither image
FIG. 5 shows a learning process for the enlargement process.

As shown in the figure, a teacher image 601 having at least 256 gradations in gray scale, color, etc.
N value (N ≧ 2) obtained by reducing the teacher image 601 by the nearest neighbor method or the like and further dithering the N value (N ≧ 2)
2) Prepare a dither image 602. That is, the N value (N
≧ 2) The target output of the dither image is the teacher image.

In FIG. 5, a dither image 60 having N values (N ≧ 2) is input from a teacher image 601 to a window 603.
2 to the window 604. Therefore, the window 603 and the window 604 are input to the neural network unit (before learning) 605. Here, the correspondence between the input and output patterns of the window 603 and the window 604 is learned by the back propagation method.

Restoring the N-value (N ≧ 2) dither image to the number of gradations
FIG. 6 shows a learning process for performing the enlargement process.

First, the maximum learning number m is substituted (S70).
1), a learning number counter n is initialized to 0 (S70).
2).

Next, it is checked whether n exceeds m (S
703) If n exceeds m, the learning is terminated.

If n does not exceed m, the total number of window patterns to be learned is input to a variable k in advance (S
704). Note that an example of the variable k is the total number of pixels of the input image.

Next, the window pattern counter 1 is initialized (S705), and it is confirmed whether the window pattern counter 1 has not reached the total number k of window patterns (S706). If the window pattern counter 1 exceeds the total number k of window patterns, it is considered that the input / output relationship of all pixels between the teacher image 601 and the dither image 601 of N value (N ≧ 2) has been learned once, and will be described later. S71
Go to 0.

If the window pattern counter 1 does not exceed the total number k of window patterns, an input is made from the teacher image 601 to the window 603, and the N value (N ≧ 2)
Is input to the window 604 from the dither image 602 (S707). Therefore, the window 603 and the window 6 are added to the neural network unit (before learning) 605.
Enter 04. Here, the correspondence between the input and output patterns of the window 603 and the window 604 is learned by the back propagation method (S708).

After the learning operation, the window pattern counter 1 is incremented (S709), and S
Returning to 706, it is confirmed whether the window pattern counter 1 has not reached the total number k of window patterns. If not reached, the position of the center pixel of the window is advanced by one, and the operations from S706 to S709 are performed. This operation is performed on the entire surface of the image, and the learning is performed once.

When one learning operation is completed, n is incremented (S710), and the process returns to S703 to check whether the number of learning operations has reached the maximum (S703).

Thereafter, if the number of times of learning is not the maximum, the window reference position is reset to the initial position, and the operations from S704 to S710 are repeatedly executed. When the number of times of learning is the maximum, the learning operation is terminated and the process ends.

FIG. 7 is a block diagram for restoring and enlarging the dither image of N value (N ≧ 2).

Here, a neural network unit (learned) 801 is used. The neural network unit (learned) 801 corresponds to the gradation number restoring / enlarging unit 203 by the learning operation described above.

An image is extracted from a dither image 802 having N values (N ≧ 2) to a window 803 and input to a neural network unit (learned) 801. Then, the pixels subjected to the gradation number restoration / enlargement processing are output to the window 804, and the window 804 is rearranged and integrated into the output image 805. Then, the output image 805 can be made closer to the characteristics of the teacher image 601.

As described above, by using a sufficiently learned neural network for the algorithm for restoring the number of tones of the dither image 602 of N value (N ≧ 2) to a halftone image with higher resolution, the gradation of the dither image The software processing time and hardware cost when performing number restoration and halftone image enlargement can be reduced, and the image quality of each algorithm for dither image gradation number restoration and halftone image enlargement Deterioration can be prevented.

Preliminary experiments have shown that the above processing using a neural network can sharpen the edge.

Also, when the teacher image 601 is used as the original image and an image obtained by applying only the reduction algorithm to the input image and learning is performed, the neural network can be similarly used as the enlargement algorithm.

[0052]

As described above, according to the present invention, the process of restoring the number of tones of a dither image having an N value (N ≧ 2) and the process of enlarging the halftone image for enlarging it are performed at once. Thus, the number of image memories and windows is reduced, and the processing time and the processing cost in the gradation number restoration processing and the halftone image enlargement processing can be effectively reduced.

Further, since the tone number restoration processing of the N value (N ≧ 2) dither image and the halftone image enlargement processing for enlarging the dither image are performed at once, the tone number restoration processing and the halftone image enlargement processing are performed. An effective effect is obtained that it is possible to prevent image quality deterioration in processing.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an image processing apparatus according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing the correspondence between the positions of an input window and an output window in image processing according to an embodiment of the present invention;

FIG. 3 is an explanatory diagram showing a state of scanning an input window in image processing according to one embodiment of the present invention;

FIG. 4 is an explanatory diagram showing a structure of a neural network in image processing according to one embodiment of the present invention;

FIG. 5 is a block diagram showing a learning process for restoring and enlarging the number of gradations of a dither image in image processing according to an embodiment of the present invention.

FIG. 6 is a flowchart illustrating a learning process for restoring and enlarging the number of gradations of a dither image in image processing according to an embodiment of the present invention.

FIG. 7 is a block diagram showing a process for restoring and enlarging the number of gradations of a dither image in image processing according to an embodiment of the present invention;

FIG. 8 is a block diagram showing conventional image processing.

[Explanation of symbols]

101 N value (N ≧ 2) dither image 102 window 103 gradation number restoration unit 104 window 105 enlargement unit 106 window 107 image memory 201 N value (N ≧ 2) dither image 202 window 203 gradation number restoration / enlargement unit (floor (Tone restoration / enlargement unit) 204 window 205 image memory 301 central pixel in window before processing 302 central pixel in window after processing 303 central pixel in window before processing 304 central pixel in window after processing 401 pixel 402 Whole input image 403 Input window 404 Input window 405 Pixel 501 First layer 502 Second layer 503 Third layer 601 Teacher image 602 N-value (N ≧ 2) dither image 603 Window 604 Window 605 Neural network Knit (before learning) 801 neural network unit (learned) 802 N values (N ≧ 2) dithered image 803 window 804 window 805 output image

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification code FI Theme coat ゛ (Reference) H04N 1/405 G09G 5/36 520F 1/40 H04N 1/40 C 103B F-term (Reference) 5B057 CA02 CA07 CA12 CA16 CB02 CB08 CB12 CB16 CD05 CE11 CH08 5C076 AA21 AA27 BA03 BA06 BB40 BB44 5C077 LL17 LL18 MP01 NN08 PP20 PP68 PQ15 PQ22 RR06 5C082 AA01 BA35 BB15 BB44 CA11 CA33 DA54 DA55 MM02

Claims (4)

[Claims] An image processing apparatus includes: a window into which a dither image having an N value (N ≧ 2) is input; and performing a tone number restoration process on the dither image in the window to a halftone image having an M value (M> N). An image processing apparatus comprising: a gradation number restoring / enlarging unit for performing an enlarging process; and an image memory for storing the halftone image after the enlarging. 2. A dither image having an N value (N ≧ 2) is input to a window, and the dither image in the window is subjected to a gradation number restoration process to a halftone image having an M value (M> N) and enlarged. Performing an image processing and storing the enlarged halftone image. 3. The image processing method according to claim 2, wherein the enlargement processing is performed using a neural network. 4. The image processing method according to claim 3, wherein said neural network is configured in a layered manner.