JPH08163436A - Image generator - Google Patents

Abstract

PURPOSE: To generate an image with a desired pattern and a desired texture from a computer graphic image. CONSTITUTION: A classification circuit 3 generates a class code depending on a pattern of level distribution of plural picture elements in a block of computer graphics signals. A class versus representative picture element value table obtained by learning is stored in a ROM table 4. A representative picture element value is decided by using a real image of a pattern and a texture desired to be added in the case of learning. An area signal generating circuit 5 generates an area signal used to designate the area to which a texture is desired to be added. A selector 2 selects a representative picture element value in place of picture elements of the computer graphics signal in the area. A desired pattern and a desired texture are expressed by the representative picture element value.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image generating apparatus for generating an image in which a texture such as a lawn surface is added to an image generated by computer graphics.

[0002]

2. Description of the Related Art An image is made more realistic by adding a specific pattern or a texture (referred to as a texture) such as a surface of grass, a grain of wood or a glossy surface to an image created by using computer graphics. It can be anything. The conventional method is to make a computer graphics creation tool (a tool for creating computer graphics, application software that runs on a computer) a pattern prepared in advance according to the shape of an object of computer graphics, It was a piece of glue.

[0003]

However, it is virtually impossible for the application software to cover all the patterns and textures intended by the computer graphics creator. Unless the computer graphics creator wants to use a completely original pattern or texture that they have never seen in the world, there are video samples such as those seen in a scene of a movie or in a photograph. If they exist, by utilizing them, it is possible to reduce the load on the creation tool and easily obtain what the creator intends.

Therefore, an object of the present invention is to learn a desired pattern or texture contained in an image signal and add it to an image created by computer graphics, thereby facilitating an intended pattern or texture. An object is to provide an image generation device that can be added.

[0005]

According to the present invention, in an image generating apparatus for adding a pattern or texture to an image created by computer graphics, a plurality of pixels in a computer graphics signal around a pixel of interest are used. Means for determining the class of the pixel of interest, a table in which representative pixel values related to the desired pattern or texture obtained by learning in advance are stored for each class, and attention in the area where the pattern or texture is to be added Regarding the pixel, the image generating apparatus is characterized by comprising a means for selectively outputting the representative pixel value read from the table corresponding to the class from the class determining means, instead of the pixel of the computer graphics signal. is there. Further, the coefficient of the linear linear combination can be stored in the table instead of the representative pixel value.

[0006]

Using the signal of the actual image of the pattern or texture, the level distribution pattern of a plurality of pixels around the pixel of interest, that is, the representative pixel value for each class is obtained. This representative pixel value is stored in the table. The pixel value of the computer graphics signal in the area where the pattern or texture is desired to be added is replaced with the representative pixel value output from the table. As a result, an image signal with a pattern or texture added is output. The user can add a desired pattern or texture without increasing the load on the computer graphics creation tool.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is an overall block diagram of an embodiment of the present invention. In FIG. 1, reference numeral 1 is an input terminal for a computer graphics signal (GC signal). This G
The C signal is supplied to one input terminal A of the selector 2, the class classification circuit 3, and the area signal generation circuit 5.

As will be described later, the class classification circuit 3 uses the CG around the pixel to which a desired pattern, texture or the like is to be added.
A class code is generated according to the level distribution of a plurality of pixels of the signal. This class code is supplied to the ROM table 4 as an address signal. The ROM table 4 stores a table of class-to-representative pixel values acquired by learning in advance. The representative pixel value read from the ROM table 4 is supplied to the input terminal B of the selector 2.

The selector 2 is controlled by the area signal from the area signal generating circuit 5. That is, in the CG signal,
In the area where the original pixel is output as it is, the selector 2 selects the input terminal A. On the other hand, in the CG signal, for the pixel to which the texture is added, the input terminal B, that is, the representative pixel value from the ROM table 4 is selected.
The output signal of the selector 2 is taken out to the output terminal 6. The area signal generation circuit 5 has, for example, a memory in which a user reproduces CG on a monitor in advance, specifies an area to which a texture is to be added, and stores coordinate values of the specified area. The area designation is not limited to this, and each pixel in the CG signal can be added with a flag that distinguishes an area to which a texture is added from an area to which a texture is not added. The configuration of FIG. 1 can also be incorporated in a computer graphics system.

FIG. 2 is a block diagram for explaining the processing at the time of learning. An actual image of the texture to be added, for example, an image of the surface of the lawn is photographed by the television camera.
This image is a still image. Not only a television camera, but an image source such as a VTR or a disc reproducing device can be used. Further, in reality, the image signal supplied to the input terminal 11 is a signal read from the image memory, and its signal rate depends on the processing speed. The input texture signal is converted into a digital signal and supplied from the input terminal 11 of FIG. 2 to the two-dimensional low pass filter 12 and the table calculation circuit 13. The number of still images is increased within the capability of the table calculation circuit 13.

The two-dimensional low pass filter 12 is an example of a circuit for removing a texture component. In addition to the low-pass filter, a circuit that removes the texture component according to the texture can be used. Low pass filter 1
The two output signals are supplied to the class classification circuit 14. The class classification circuit 14 determines the class of the pixel to which the texture is added, like the class classification circuit 3 in FIG.

The class code from the class classification circuit 14 is supplied to the table calculation circuit 13, and the class code is supplied to the ROM table 15 as an address. The table calculation circuit 13 calculates the representative pixel value of each class. The calculated representative pixel value is written in the ROM table 15. This ROM table 15 is used as the ROM table 4 in FIG.

An example of the table calculation circuit 13 is shown in FIG. Input signal from input terminal 11 and class classification circuit 1
The class code from 4 is supplied to the averaging circuit 21.
Further, the class code is supplied to the data number counter 22. The averaging circuit 21 classifies all pixels of a still image of one frame into classes and accumulates pixel values of each class. On the other hand, the number of data of each class is accumulated by the data number counter 22. Then, the average pixel value of each class is obtained by dividing the cumulative value of pixel values of each class by the cumulative value of the number of data. This average pixel value is the representative pixel value and is stored in the ROM table 15.

When obtaining the representative pixel value, a normalization process may be introduced in order to prevent the word length when accumulated from becoming long. For example, the dynamic range DR (= maximum value M
AX-minimum value MIN) is calculated, and the pixel value accumulated for averaging is normalized by subtracting the minimum value MIN and dividing by the dynamic range DR.

An example of the class classification circuit 14 is shown in FIG.
The output of the two-dimensional low-pass filter 12 is supplied to the blocking circuit indicated by 23. The blocking circuit 23 converts the time series. The output signal of the blocking circuit 23 is supplied to the binarizing circuit 24 and the threshold value determining circuit 25. The threshold value determination circuit 25 determines the threshold value TH for each block. This threshold value TH is compared with the output signal of the blocking circuit 23, and each pixel is binary (`0 'or
Is converted to `1 '). That is, when the pixel value from the blocking circuit 23 is represented by L, L is converted to "1" if L≥TH, and conversely, L is converted to "0" if L <TH.

The output of the binarization circuit 24 is supplied to the class code generation circuit 26. The class code generation circuit 26
Generate a class code with the same number of bits as the number of pixels in the block. In the class classification circuit 3, the input signal is CG.
The configuration is the same as that of FIG. 4 except that it is a signal.

FIG. 5 shows an example of blocks generated by the time series conversion of the blocking circuit 23. 3 × 3 pixels (×
The output signals of the low-pass filter 12 of 9 pixels of a to i) indicated by 1 form one block. The central pixel (indicated by ◯) of this one block is the target pixel X. The pixel of interest is classified into classes based on the pattern of the level distribution of the 9 pixels a to i.

The threshold value determining circuit 25 generates an average value of the values of 9 pixels in a block as a threshold value. Then, as shown in FIG. 6, the average value TH is compared with the value of each pixel of a to i, and as described above, `0` or` 1`.
Is binarized to. In the example of FIG. 6, (00011110
9 bits of 0) occur as a class code. Therefore, in this case, there are 2 ⁹ classes.

ADRC can be used as a class classification. First, the maximum value MAX and the minimum value MIN of a plurality of pixel values in a block are detected. The difference (dynamic range DR) is the nth power of 2 (in the case of n-bit quantization)
Divide by to obtain the quantization step size. The value obtained by subtracting the minimum value MIN from the pixel value is divided by the obtained quantization step width, and rounded to an integer to perform adaptive quantization. This is called the ADRC coding method and is expressed by the following equation. Where Q is the quantized value and x ^
Indicates the decrypted value, and [] indicates the truncation process.

DR = MAX-MIN Q = [(x-MIN) * ( ²ⁿ / DR)] x ^ = [Q * (DR / ²ⁿ ) + MIN] A plurality of ADRC-encoded codes The normalized patterns are assembled and expressed as a class by pattern classification.

FIG. 7 schematically shows the table obtained by the table calculation circuit 13. For example, an 8-bit representative pixel value is calculated corresponding to each of the 9-bit class codes. This table is used as the ROM table 4 or 15.

If the pattern or texture that the user wants to add is different from the surface of the lawn in the above-mentioned example, for example, in the case of wood grain, the input signal supplied to the input terminal 11 of FIG. 2 is changed. Then, by learning, the representative pixel value in the case of the grain of wood is acquired and stored in the ROM table 15.

Returning to the configuration of FIG. 1, when the CG signal is supplied, the class classification circuit 3 generates a class code, and the representative pixel value of the texture designated by the class code is read from the ROM table 4. It The selector 2 receives the area signal from the area signal generation circuit 5.
Is selected, and the pixel value included in the designated area is changed to the representative pixel value. Therefore, the CG signal from the output terminal 6 is a signal in which the designated area has been changed to have a desired texture.

Another embodiment of the present invention will be described with reference to FIG. Similar to the configuration of the embodiment shown in FIG. 1, a desired pattern and texture are added to the area designated by the area signal. In another embodiment, a ROM
Table 7 is a class-to-coefficient table. This table is acquired in advance by learning.

The coefficients read from the ROM table 7 are supplied to the arithmetic circuit 8. The arithmetic circuit 8 is supplied with the CG signal through the blocking circuit 9. Assuming that the block structure is as shown in FIG. 5, the coefficients w ₁ to w ₉ are read from the ROM table 7. Then, the coefficient and the surrounding pixel a
By linear combination of the through i (each pixel value is x ₁ ~x _9), the value of the target pixel X is generated.

X = w ₁ × x ₁ + w ₂ × x ₂ + ... + w ₉ × x ₉ (1) Here, a plurality of pixels used for this class classification are a plurality of pixels used for interpolation calculation. The pixels x _{1 to} x ₉ need not be the same, and may be different.

For learning to determine the coefficients, the arrangement of FIG. 9 is used. This configuration is the same as the configuration of FIG. 2 described above. In the coefficient determination circuit indicated by 13, the coefficient is determined for each class by the least square method, and the determined coefficient is stored in the ROM table 16. Coefficient determination circuit 13
Is supplied with an input signal corresponding to a pattern, texture, etc., a class code, and an output signal of the two-dimensional low-pass filter 12. In the coefficient determination circuit 13, the coefficient that minimizes the sum of squares of the error between the predicted pixel value predicted by the equation (1) and the true value of the pixel of interest is determined by the least square method.

The learning method is shown in the flowchart of FIG. In FIG. 10, step 31 is a step of an imaging process for collecting learning data (images including patterns and textures that the user wants to use). The captured image is digitized and stored in the memory (step 32). A class code is generated using the image signal stored in this memory (step 3).
3). This is because the pixels around the pixel of interest in the block are
This is a process of digitizing.

The process surrounded by the broken line after step 33 is the process of determining the coefficient by the method of least squares. As will be described later, the observation equation is created in step 35, the normal equation is created in step 36, and the normal equation is obtained by the sweep method in step 37. Then, the obtained coefficient is output in step 38.

The calculation for obtaining the coefficient by the method of least squares will be described below. The observation equation is generally expressed by the following equations (2) and (3). x _ij is the pixel data in the input image, and y _i is the true value corresponding to the interpolation value. Basically,
It suffices to solve the simultaneous equations and obtain the coefficients w ₁ to w _n .

[Observation equation] XW = Y (2)

[0032]

[Equation 1]

That is, as shown in equation (4), the coefficient matrix is determined so that the residual matrix E is added and the square error is minimized.

[Residual Equation]

[Equation 2]

From the equation (4), in order to find the most probable value of the coefficient w _i , w satisfying the condition of minimizing Σe _i ² (Σ means accumulation of i = 1 to i = m) ₁ , w ₂ , ...
・, W _n should be found. That is, n ₁ , which satisfy the following equation (5), satisfy w ₁ , w ₂ , ...
·, Seek w _n.

[0036]

(Equation 3)

From the equation (4), the following equation (6) is established.

[0038]

[Equation 4]

The condition of the equation (5) is i = 1, 2, ..., N
The following equations (7) are obtained respectively.

[0040]

(Equation 5)

Here, the following normal equation is obtained from the equations (4) and (7).

[0042]

(Equation 6)

Since this is a simultaneous equation with a number corresponding to the number n of unknowns, each w _i which is the most probable value can be obtained from this normal equation. To be precise, if the matrix having (Σx _jn x _jn ) elements for each w _i in Expression (8) is regular, this normal equation can be solved. Actually, the Gauss-Jordan elimination method (sweeping method) is used to solve this normal equation.

The coefficient w _i is calculated for each class indicated by the index code. The coefficient for each class thus obtained is stored in the ROM table (ROM table 7 in FIG. 8 and ROM table 16 in FIG. 9) and used for pixel value generation.

[0045]

According to the present invention, when a user wants to add a desired pattern or texture to an image of computer graphics, learning is performed by using an actual image of the pattern or texture in advance and the representative pixel is used. Values or coefficients for prediction are determined. Therefore, desired patterns and textures can be added without increasing the load on the computer graphics creation tool.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration for learning according to an embodiment of the present invention.

FIG. 3 is a block diagram of an example of a texture calculation circuit.

FIG. 4 is a block diagram of an example of a class classification circuit.

FIG. 5 is a schematic diagram used for explaining a block structure.

FIG. 6 is a schematic diagram illustrating an example of class classification processing.

FIG. 7 is a schematic diagram showing an example of a table of class vs. representative pixel value.

FIG. 8 is a block diagram of another embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration for learning according to an embodiment of the present invention.

FIG. 10 is a flowchart showing a process at the time of learning.

[Explanation of symbols]

 2 selector 3 class classification circuit 4 ROM table 13 coefficient determination circuit by least square method

Claims (3)

[Claims] 1. An image generation apparatus for adding a pattern or texture to an image created by computer graphics, wherein a plurality of pixels in the computer graphics signal around a pixel of interest are used to classify the pixel of interest. A table for storing representative pixel values for a desired pattern or texture obtained by learning in advance for each class, and for the pixel of interest in the area to which the pattern or texture is to be added, And an image generating device which selectively outputs the representative pixel value read from the table in correspondence with the class from the class determining means, instead of the pixel of the computer graphics signal. 2. The image generating apparatus according to claim 1, wherein a normalized representative pixel value is stored in the table. 3. An image generation apparatus for adding a pattern or texture to an image created by computer graphics, wherein a plurality of pixels in the computer graphics signal around a pixel of interest are used to classify the pixel of interest. For determining the value of the pixel of interest included in the signal of the pattern or texture, and the linearity of a plurality of pixel data and a plurality of coefficients located in the vicinity of the pixel of interest and included in the computer graphics signal. Prediction is performed by linear combination, the plurality of coefficients are determined so as to minimize the sum of squares of the error between the predicted pixel data and the true value of the pixel of interest, and the determined coefficient is determined for each class. In the stored table and in the area where you want to add the pattern or texture, click from the class determining means. Image generating apparatus, which comprises means for selectively outputting a predicted pixel value formed by using the coefficient read from the table corresponding to the raster, instead of the pixel of the computer graphics signal. .