JP2003179887A - Digital image signal generating apparatus and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a digital image signal generating apparatus that selects a pixel in use depending on a result of still picture discrimination so as to determine a class of a target pixel with high accuracy in the case of generating a pixel value of a digital image signal with higher resolution from a received digital image signal. <P>SOLUTION: The digital image signal generating apparatus performs the same processing as that of an encoder employing the MUSE system and generates in advance a coefficient to accurately interpolate a value of interleaved pixels through learning by using a resulting signal and an original High Vision signal. Least square method arithmetic circuits 65a, 65b determine a coefficient for minimizing the square of an error between an estimate value and a true value by each class and store the results to memories 67a, 67b. The estimated value is generated by linear combination between the coefficient and values of surrounding pixels. The arithmetic circuit 65a determines the coefficient as a plurality of pixels in use when using pixels of the same field and the arithmetic circuit 65b determines the coefficient as a plurality of pixels in use when using pixels of the same frame. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital image signal generation apparatus applied to a decoder for a high resolution video signal which compresses the amount of transmission information by sub-sampling, for example, a MUSE decoder which is a compression method for a high definition signal, and Regarding the method.

[0002]

2. Description of the Related Art As one method for band compression or information amount reduction when recording or transmitting a digital image signal, there is a method of reducing the amount of transmission data by thinning out pixels by subsampling. One example is the multiple sub-Nyquist sampling encoding method in the MUSE method. This system can compress a high-definition signal into a band of about 8 MHz.

[0003]

In the conventional MUSE method, a two-dimensional spatial filter is used for interpolation when decoding data which has been subsampled once or twice at the time of encoding. However, MUS
In the E method, since the transmission information amount is compressed by utilizing the visual characteristic that the resolution in the diagonal direction is low, there is a problem that the resolution in the diagonal direction lost at the time of encoding cannot be recovered.

Therefore, an object of the present invention is to provide a digital image signal generating apparatus and method which are applied to a MUSE type decoder and solve the above-mentioned problems.

[0005]

According to a first aspect of the present invention, a digital image signal generator for generating a pixel value of a digital image signal having a higher resolution than a digital image signal from an input digital image signal is used as a generation target. Stationary determination means for determining stationary of the first pixel of interest,
Classifying means for determining the class of the pixel of interest on the basis of a plurality of pixels having a local correlation with the first pixel of interest, and a coefficient for each class acquired in advance by learning to generate the pixel. And a pixel value generation unit that is coupled to the memory unit and that generates a pixel value of the first pixel of interest by calculating a plurality of pixels and a coefficient. The plurality of pixels having a local correlation with the pixel of interest are selected in accordance with the determination result by the stillness determination means.

According to the present invention, when the target pixel is classified into a class, if the target pixel is determined to be a moving portion in accordance with the determination result of the stillness determination means, it is temporally and spatially close to the target pixel, for example, the same. If the classification is performed based on a plurality of pixels in the field and it is determined that the pixel of interest is a still portion, the classification is performed based on a plurality of pixels spatially near the pixel of interest, for example, a plurality of pixels in the same frame. In this way, the pixels used for class classification are made different according to the stillness determination result, so it is possible to perform class classification using a plurality of pixels that have a local correlation with the pixel of interest, improving the accuracy of class classification. it can.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. First, the main part of the MUSE encoder will be described with reference to FIG. A high-definition signal is converted into a digital signal by an A / D converter, and the Y (luminance) signal and Pr (R) are calculated by matrix calculation.
-Y component) signal and Pb (BY component) signal are formed,
It is supplied to the input terminals indicated by 1, 2, and 3 in FIG.

The Y signal is supplied to the inter-field prefilter 4. A field offset sub-sampling circuit 5, a low-pass filter 6 and a sampling frequency conversion circuit 7 are connected to the filter 4. The field offset subsampling circuit 5 shifts the subsampling phase by 1 pixel between fields, and the output thereof is supplied to the low-pass filter 8. Hara Y
The sampling frequency of the signal is 48.6 MHz, and the sampling frequency of the sub-sampling circuit 5 is 24.3 M.
12.15 MHz by the low pass filter 8 at
The frequency components above z are removed, and the data is interpolated to restore the sampling frequency to 48.6 MHz.

A sampling frequency conversion circuit 9 is connected to the low-pass filter 8, and the sampling frequency is 32.4 MH by the sampling frequency conversion circuit 9.
converted to z. The output signal of this circuit 9 is TCI (Time
Compressed Integration) The switch 10 is supplied.
The signal path from the sub-sampling circuit 5 to the conversion circuit 9 is provided for processing the static region.

The sampling frequency conversion circuit 11 is connected to the low pass filter 6 for band limitation, and 48.6.
The sampling frequency is converted from MHz to 32.4 MHz. The output of this circuit 11 is supplied to the TCI switch 12. The signal from the TCI switch 12 is supplied to the mixing circuit 17 via the two-dimensional sub-sampling filter 16. A signal path from the low-pass filter 6 to the sub-sampling filter 16 is provided for processing the motion area. In the mixing circuit 17, the output signal of the filter 16 and the output signal of the TCI switch 10 are mixed.

A motion vector detection circuit 13 is connected to the sampling frequency conversion circuit 7. A motion filter 14 and a motion detection circuit 15 are connected to the motion vector detection circuit 13. The motion filter 14 includes
The output signal of the sampling frequency conversion circuit 11 is also supplied. The output of the motion filter 14 is supplied to the motion detection circuit 15. A control signal for controlling the mixing ratio of the mixing circuit 17 is generated based on the detection result (motion amount) of the motion detection circuit 15.

The color signals Pr and Pb from the input terminals 2 and 3 are supplied to the line sequential circuit 23 through the vertical low pass filters 21 and 22, respectively. The line-sequential color signal from the line-sequencing circuit 23 is supplied to the low-pass filter 24,
Components of z or more are removed and then supplied to the field offset subsampling circuit 26. The line-sequential color signal is supplied to the field offset sub-sampling circuit 27 via the low-pass filter 25 for band limitation. Time compression circuit 2 for sub-sampling circuit 27
8 are connected.

The low-pass filter 24 and the sub-sampling circuit 26 are processing circuits for the stationary region, and the low-pass filter 25, the sub-sampling circuit 27 and the time compression circuit 28 are processing circuits for the moving region. The output signals of the sub-sampling circuit 26 and the time compression circuit 28 are supplied to the TCI switches 10 and 12, respectively,
It is time-axis multiplexed with the luminance signal component processed as described above.

The output signal of the mixing circuit 17 is supplied to the frame / line offset sub-sampling circuit 31.
The sub-sampling pattern here is inverted between frames and between lines, and the sampling frequency is set to 16.2 MHz. The output signal of the sub-sampling circuit 31 is transmitted through the transmission gamma correction circuit 32 to the MUSE
Is supplied to the formatting circuit 33. Although not shown in the figure, a time-axis-compressed audio signal, sync signal, VIT signal, etc. are added to the formatting circuit 33, and a MUSE signal of about 8 MHz is taken out from the output terminal 34.

Subsampling of the MUSE encoder described above will be schematically described with reference to FIG. The processing of the static region is shown on the upper side, and the processing of motion quantization is shown on the lower side. FIG. 2 shows the sampling state of the signal at each point in FIG. Further, the processing of the C signal is the same as that of the Y signal, and thus the description thereof will be omitted. A digital Y signal is supplied from the input (point A) of the field offset subsampling circuit 5, and an output signal subsampled in a pattern in which the sampling phase is shifted by one pixel for each field is generated at point B.

At the output (point C) of the low-pass filter 12, the interpolated signal (sampling frequency is 48.
6 MHz) is generated. At the output (point D) of the sampling frequency conversion circuit 9, a signal whose sampling frequency has been converted to 32.4 MHz appears.

On the other hand, the input of the low-pass filter 6 (point a)
Is supplied with a digital Y signal similar to that at the point A. In the motion area, the field offset sub-sampling is not performed, and the output of the sampling frequency conversion circuit 11 (b
At the point), the Y signal similar to that at the point D is generated.

The Y signals that have been subjected to the respective processing of the static region and the moving region are mixed in the mixing circuit 17, and the mixing circuit 1
The output of 7 is supplied to the frame / line offset sub-sampling circuit 31. Output of this circuit 31 (point E)
In, an output signal sampled so as to have an offset of one pixel in the horizontal direction between frames and between lines is generated.

FIG. 3 shows a part of a MUSE decoder to which the present invention can be applied. The MUSE signal received, converted into a baseband signal, and converted into a digital signal is supplied to the interframe interpolation circuit 41, the interfield interpolation circuit 42, and the motion part detection circuit 43, respectively. The moving portion detection circuit 43 detects a moving area and controls the mixing ratio of the signals processed by the moving area and the still area.

That is, the stationary area is interpolated by the interframe interpolation circuit 41 using the image data of the previous frame. However, when the entire image moves, like the panning of the camera, a process of moving and superimposing the image one frame before according to the motion vector transmitted as the control signal is performed. The output signal of the inter-frame interpolation circuit 41 is a low-pass filter 44, a sampling frequency conversion circuit (32.4 MHz to 48.6 MH).
z) 45, a field offset sub-sampling circuit 46, and an inter-field interpolation circuit 47 to a mixing circuit 48. A signal having a sampling frequency of 24.3 MHz is obtained from the sub-sampling circuit 46.

The motion area is spatially interpolated by the field interpolation circuit 42. A sampling frequency conversion circuit 49 from 32.4 MHz to 48.6 MHz is connected to the interpolation circuit 42, and the output signal thereof is supplied to the mixing circuit 48. The mixing ratio of this mixing circuit 48 is
It is controlled by the output signal of the moving part detection circuit 43. Although not shown, the output signal of the mixing circuit 48 is supplied to the TCI decoder and separated into Y, Pr, and Pb signals. Further, R / G / B signals are obtained after D / A conversion, inverse matrix calculation, and gamma correction.

The processing of the above decoder will be schematically described with reference to the sampling pattern of FIG. Input signal (E
The sampling state of (point) is the same as the above-mentioned encoder output (point E). The still area is passed through the inter-frame interpolation circuit 4, and at its output (point F), a video signal in which thinned pixels are interpolated is generated. Sampling frequency conversion circuit 4
At 5 (point G), the sampling frequency is 48.6 MHz
The video signal converted into appears.

At the output (point H) of the field offset sub-sampling circuit 46, a signal subjected to offset sampling with a pixel shift for each field is generated.
A signal in which pixels are interpolated is generated at the next output (point I) of the inter-field interpolation circuit 47. This is supplied to the mixing circuit 48.

At the output (point f) of the field interpolation circuit 42 for processing the motion area, a video signal interpolated by the pixels in the field is generated. The sampling frequency conversion circuit 49 outputs 48.6 to its output (point h).
A video signal with a sampling frequency of MHz is generated.
This is supplied to the mixing circuit 48.

In the MUSE method described above, sub-sampling is performed twice on the stationary area and interpolation is performed twice, and sub-sampling and interpolation are performed once on the moving area. Conventionally, a filter is used for these interpolations, but as a result, there is a problem that the resolution in the diagonal direction is lost as described above. The present invention solves this problem. Therefore, the present invention can be applied to any of the interframe interpolation circuit 41, the field interpolation circuit 42, and the interfield interpolation circuit 47 in the above MUSE decoder. Applicable.

As an example, FIG. 5 shows an embodiment in which the present invention is applied to the field interpolation circuit 42 for the motion area. In FIG. 5, reference numeral 51 is an input terminal for an offset sub-sampled digital image signal. Reference numeral 52 is a time series conversion circuit for converting an input signal into a block structure signal. That is, the time-series conversion circuit 52 synchronizes a plurality of pixels required for classification and interpolation calculation.

The output signal of the time series conversion circuit 52 is supplied to the interpolation calculation circuit 53 and the class classification circuit 55. The interpolation calculation circuit 53 is connected to a coefficient memory 54 that stores coefficients acquired by learning in advance as described later. In the coefficient memory 54, a table 54a in which the first coefficient is stored and a table 54 in which the second coefficient is stored.
b and are included.

A class code c is generated from the class classification circuit 55. The pattern of the two-dimensional (in-field or in-frame) level distribution of the block of the block including the pixel of interest, which is the target of interpolation, that is, the class is determined. The class code c indicates this class, and the class code c is supplied to the coefficient memory 54 as its address.

In FIG. 5, a signal indicating the amount of movement of the pixel of interest is supplied from the input terminal 57 to the comparison circuit 58. As the motion amount signal, for example, the output signal of the motion part detection circuit 43 of the MUSE decoder (FIG. 3) can be used. Specifically, the signal indicating the amount of movement has a value in the range of, for example, 0 to 16 proportional to the amount of movement. In the comparison circuit 58, the threshold value TH is compared, and when the motion amount signal is larger than the threshold value TH, the pixel of interest is determined to be a moving pixel, and when it is below the threshold value TH, the pixel of interest is stopped. Judge as a pixel. Although TH is set appropriately, TH = 3 is an example.

The output signal (determination signal) of the comparison circuit 58 is supplied to the time series conversion circuit 52 and the coefficient memory 54.
The peripheral pixels output by the time series conversion circuit 52 are switched according to the determination signal. That is, when the determination signal indicates that the pixel of interest is a moving pixel, the time-series conversion circuit 52 outputs the peripheral pixels in the field, and when the determination signal indicates that it is a still pixel, this is the frame. The peripheral pixels inside are output. More specifically, the time series conversion circuit 52 is provided with a selector or an address generation circuit controlled by a determination signal.

Further, according to the determination signal, the coefficient memory 54
Tables 54a and 54b are selectively used. That is, for a moving pixel, the first coefficient of the table 54a is output to the interpolation calculation circuit 53, and for a still pixel, the second coefficient of the table 54b is output to the interpolation calculation circuit 53. At the time of learning which will be described later, the first coefficient of the table 54a is determined by referring to the peripheral pixels in the field, and the second coefficient of the table 54b is determined by referring to the peripheral pixels in the frame.

Class code c from class classification circuit 55
Is supplied to the coefficient memory 54, the coefficient corresponding to the class is stored in the table 54a or 54b of the coefficient memory 54.
Read from. By linear linear combination of the coefficient from the memory 54 and the value of the peripheral pixel from the time series conversion circuit 52,
An interpolated value for the pixel of interest is formed. The interpolation calculation circuit 53 outputs the interpolation value of the thinned pixel to the output terminal 56. In the interpolation calculation circuit 53, the interpolation value y ′ is generated by the linear linear combination of the following equation.

Y ′ = w1 x1 + w2 x2 + ... + wn xn (1) x1 to xn are values of pixels around the pixel of interest, and w1
~ Wn are coefficients determined in advance for each class.

The above-mentioned coefficient memory 54 stores the first and second coefficients created in advance by learning.
FIG. 6 shows an example of a configuration for learning. A high resolution digital image signal for learning is supplied from an input terminal 61. As this input signal, still image signals having different patterns can be used.

The input digital image signal is supplied to the frame / line offset sub-sampling circuit 63 through the two-dimensional sub-sample filter 62, as in the MUSE encoder. The output of the circuit 63 is supplied to the time series conversion circuits 64a and 64b, and the data of a plurality of reference pixels are synchronized. Time series conversion circuit 64a, 6
The output signal of 4b is a least squares arithmetic circuit 65a, 65b
And the classification circuits 66a and 66b, respectively.

The time series conversion circuit 64a simultaneously synchronizes a plurality of pixels which are pixels in the same field as the pixel of interest and around the pixel of interest. The other time series conversion circuit 64b synchronizes a plurality of pixels in the same frame as the pixel of interest and around the pixel of interest. Then, the class classification circuit 6
As shown in FIG. 7, 6a is performed based on the level distribution of four reference pixels (whose levels are a, b, c, and d) in the same field around the pixel of interest (interpolation pixel). . That is, as shown in FIG. 8, the class classification circuit 66a calculates the average value Av of the reference pixels a to d, then compares each value of the reference pixels with the average value Av, and responds to the comparison result. Generate class code c. In the example of FIG. 8, (a <A
The class code c of (0101) is formed based on the comparison result of v, b ≧ Av, c <Av, d ≧ Av.

The class classification circuit 66b similarly generates a class code c. However, the class classification circuit 66b
Classification is performed using three reference pixels b, d, and e (FIG. 7) in the same frame. As a reference pixel,
The choice of what is optional is merely an example. The class code c generated by the class classification circuits 66a and 66b is the least squares arithmetic circuit 65a.
And 65b. For these arithmetic circuits 65a and 65b, the time series conversion circuits 64a and 64b are provided.
And the true value of the pixel of interest from the input terminal 61 are respectively supplied.

The class classification circuit 5 of the interpolator of FIG.
5 classifies the target pixel into classes, as in the above-described class classification circuits 66a and 66b. In FIG. 5, since the time series conversion circuit 52 outputs a plurality of pixels in the field or a plurality of pixels in the frame according to the determination signal, one class classification circuit 55 classifies using the pixels in the field and Classification using pixels is selectively performed. If necessary, class classification circuit 55
A determination signal may be supplied to

Another example of the class classification circuits 55, 66a, 66b is ADRC (Adaptive Dynamic Range Coding). ADRC is a technique for adaptively removing redundancy in the level direction by utilizing local correlation of images. More specifically, 1-bit ADRC can be used. That is,
The maximum value and the minimum value of the block including the reference pixel described above are detected, the dynamic range that is the difference between the maximum value and the minimum value is detected, the value of the reference pixel is divided by the dynamic range, and the quotient is 0.5. Are compared to 0, and those smaller than 0.5 are encoded as '1', and those smaller than 0.5 are encoded as '0'.

An ADRC that generates an output of a bit number other than 1 bit may be adopted. Not limited to ADRC, DPC
M (Differential pulse code modulation), BTC (Blo
A compression coding encoder such as ck Trancation Coding) can be used as the class classification circuits 55, 66a, and 66b. Further, it is possible to use the value of the reference pixel as it is for classifying. VQ (Vector Quantization) can also be used for information compression.

The least squares arithmetic circuits 65a and 65b are
For each class, the coefficient is determined so that the square of the error between the estimated value y ′ of the pixel of interest represented by a linear linear combination of the value of the peripheral pixel and the coefficient and its true value y is minimized. Then, the determined coefficients are stored in the memories 67a and 67b of the coefficient memory 67, respectively. The one stored in the memory 67a is used as the table 54a in the interpolation device of FIG. 5, and the one stored in the memory 67b is the table 54b.
Used as.

The determination of the coefficient by the method of least squares will be described with reference to the flowchart of FIG. Step 71
The control of the learning process is started from, and in the learning data formation of step 72, learning data corresponding to a known image is formed. The value of the peripheral pixel in the field (in the case of the arithmetic circuit 65a) or in the frame (in the case of the arithmetic circuit 65b) is adopted as the learning data. The true value y of the pixel of interest and the values x1 to xn of the peripheral pixels are a set of learning data.

Here, control is performed so that blocks having a dynamic range smaller than the threshold value are not treated as learning data. This is because those having a small dynamic range are easily affected by noise, and an accurate learning result may not be obtained. At the end of the data in step 73, if the processing of all the input data, for example, one frame of data has been completed, the control proceeds to the prediction coefficient determination in step 76, and if not completed, the control proceeds to the class determination in step 74.

The class determination in step 74 is based on the value of a predetermined pixel in the field or frame, as described above. In the normal equation addition in step 75, the normal equation of Expression (9) described later is created. After the processing of all the data is completed, the control is transferred to the step 76 from the end of the data in step 73. This step 76
In determining the prediction coefficient, the normal coefficient is solved using a matrix solution method to determine the prediction coefficient. The prediction coefficient store in step 77 stores the prediction coefficient in the memory, and step 78
Then, the control of the learning process ends.

Step 75 in FIG. 9 (normal equation generation)
The process of step 76 (determination of prediction coefficient) will be described in more detail. Let y be the true value of the pixel of interest and y be its estimated value.
′ And the values of the surrounding pixels are x1 to xn,
An n-tap linear primary combination y '= w1 x1 + w2 x2 + ... + wn xn (2) is set for each class by coefficients w1 to wn. Before learning, wi is an undetermined coefficient.

As described above, learning is performed for each class,
When the number of data is m, the equation (2) is represented by the equation (3). yj '= w1 xj1 + w2 xj2 + ... + wn xjn (3) (where j = 1, 2, ... m)

In the case of m> n, w1 to wn are not uniquely determined, so that the elements of the error vector E are predicted errors in the respective learning data xj1, xj2, ... Xjn, yj.
The j is defined as the following expression (4). ej = yj- (w1xj1 + w2xj2 + ... + wnxjn) (4) (where j = 1, 2, ... m) Next, the coefficient that minimizes the following equation (5) is obtained, and the least squares method is used. The optimum prediction coefficients w1, w2, ..., Wn are determined.

[0048]

[Equation 1]

That is, when the partial differential coefficient by wi of the equation (5) is obtained, the following equation (6) is obtained. Formula (6)
(I = 1, 2, ..., N).

[0050]

[Equation 2]

Since each wi may be determined so that the equation (6) becomes 0,

[0052]

[Equation 3]

Using a matrix as

[0054]

[Equation 4]

It becomes This equation is generally called a normal equation. A normal equation is a simultaneous equation with exactly n unknowns. As a result, the undetermined coefficients w1, w2, ..., Wn that are the most probable values can be obtained. Specifically, since the matrix on the left side of the equation (9) is generally positive definite symmetric, the simultaneous equations of the equation (9) can be solved by a method called Cholesky method, the undetermined coefficient wi can be obtained, and the class code The address wi is stored in the memory.

[0056]

According to the present invention, when a pixel of interest is classified, when the pixel of interest is determined to be a moving portion, a plurality of pixels temporally and spatially adjacent to the pixel of interest are used for class classification. If the target pixel is determined to be a still part, the classification is performed using a plurality of pixels spatially close to the target pixel, and thus the class is determined using a plurality of pixels having a strong local correlation with the target pixel. Classification can be performed, and the accuracy of class classification can be improved.

[Brief description of drawings]

FIG. 1 is a partial block diagram of a MUSE encoder.

FIG. 2 is a schematic diagram for explaining subsampling of a MUSE encoder.

FIG. 3 is a partial block diagram of a MUSE decoder to which the present invention can be applied.

FIG. 4 is a schematic diagram for explaining an interpolation process of a MUSE type decoder.

FIG. 5 is a block diagram of an embodiment in which the present invention is applied to an interpolating device for sub-sampling signals.

FIG. 6 is a block diagram of an example of a configuration at the time of learning for determining a coefficient according to the present invention.

FIG. 7 is a schematic diagram of an example of an array of pixels used for class classification.

FIG. 8 is a schematic diagram showing an example of class classification.

FIG. 9 is a flowchart for explaining learning for obtaining a coefficient.

[Explanation of symbols]

41 Frame interpolator 42 field interpolation circuit 47 field interpolator 53 Interpolation calculation circuit 54 coefficient memory 58 Comparison Circuit for Stationary Judgment

Claims (6)

[Claims] 1. A digital image signal generation device for generating a pixel value of a digital image signal having a higher resolution than the digital image signal from an input digital image signal, wherein stillness of a first pixel of interest to be generated is determined. Stationary determination means for performing, class classification means for determining the class of the first pixel of interest based on a plurality of pixels having a local correlation with the first pixel of interest, and generating a pixel, Pixel value for generating a pixel value of the first pixel of interest by combining the memory means in which the coefficient for each class acquired by learning in advance is stored and the memory means, and calculating a plurality of pixels and the coefficient A plurality of pixels having a local correlation with the first pixel of interest in the class classification means, Flip with digital image signal generating apparatus characterized by being selected. 2. The class classification means, when the stillness determination means determines that the first pixel of interest is a moving part, temporally and spatially close to the first pixel of interest. When a class is determined using a plurality of pixels of the first target pixel and the stillness determining unit determines that the first target pixel is a stationary portion, a plurality of pixels spatially close to the first target pixel are determined. 2. The digital image signal generation device according to claim 1, wherein the pixel is used to determine the class. 3. The memory means stores the coefficient as
The first coefficient and the second coefficient acquired by learning in advance are stored, and when the stillness determination means determines that the first pixel of interest is a moving portion, the first coefficient
2. The digital image according to claim 1, wherein when the first pixel of interest is determined to be a still portion, the second coefficient is used as the coefficient when the coefficient of 1 is used as the coefficient. Signal generator. 4. The first and second coefficients are generated during the learning based on a first digital image signal and a second digital image signal having a resolution lower than that of the first digital image signal. In doing so, the true value of the second pixel of interest in the second digital image signal and a plurality of pixels in the second digital image signal that are temporally and spatially adjacent to the second pixel of interest. The first coefficient is generated for each class based on pixels, and the first coefficient is generated for each class based on the true value of the second pixel of interest and a plurality of pixels spatially adjacent to the second pixel of interest. The digital image signal generation device according to claim 3, wherein the second coefficient is generated. 5. The first and second coefficients are the second coefficient.
5. The digital image signal generating device according to claim 4, wherein the square of the error between the true value of the target pixel and the pixel value generated by the calculation is determined by the least square method. 6. A digital image signal generation method for generating a pixel value of a digital image signal having a higher resolution than that of the digital image signal from an input digital image signal, wherein the stillness of a first pixel of interest to be generated is determined. A stillness determination step to perform, a class classification step for determining the class of the first pixel of interest based on a plurality of pixels having a local correlation with the first pixel of interest, and a memory step, A pixel value generation step of generating a pixel value of the first pixel of interest by calculating a plurality of pixels and a coefficient for each class acquired by learning in advance, wherein in the class classification step, the first A plurality of pixels having a local correlation with the pixel of interest are selected according to the determination result in the stillness determination step. Ijitaru image signal generation method.