JPH1056622A - Image signal converter and method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate a high definition(HD) signal with higher resolution by taking property of a composite signal characteristic such as a phase of a subcarrier into account. SOLUTION: A motion detection section 62 receiving a composite signal generates a luminance signal in a pseudo way to detect a motion d41. A weight generating section 63 generates a weight d45 in response to the motion d41. A pre-processing section 65 receiving the composite signal generates a Y signal by 3-dimension Y/C separation. A processing section 64 uses a classification section 69, a prediction coefficient ROM 70 and a predict arithmetic section 71 to conduct classification adaptive processing by using the Y signal and the result of arithmetic operation is outputted as d48. Similarly, classification adaptive processing is conducted in the processing section 66 and the result of arithmetic operation is outputted as a signal d49. A weight sum arithmetic section 67 conducts optimum sum arithmetic operation by changing the weight of the signals d48, d49 depending on the signal d45.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal conversion apparatus and method using class classification adaptive processing capable of obtaining an image signal having a higher resolution than an input image signal.

[0002]

2. Description of the Related Art Conventionally, as a device for converting an image signal into a different format, for example, a standard TV signal (SD (Stan)) is used.
dard Definition) signal) to HD (High Definition)
There is an up-converter for converting to a format signal.
The technology used in this upconverter will be described below. First, the standard TV signal (SD signal) and HD
FIG. 10 shows a spatial arrangement example of each pixel of a signal. here,
To simplify the description, the number of pixels of the HD signal is doubled in each of the horizontal and vertical directions. Paying attention to the SD pixels indicated by ◎ in the figure, there are HD pixels at four types of neighboring positions. The modes for predicting HD pixels existing at these four types of positions are referred to as mode1, mode2, mode3, and mode4, respectively.

In a conventional upconverter, an interpolation pixel is generated by applying an interpolation filter to an input SD signal, and an HD format signal is output. As a simple configuration example of this upconverter, it is conceivable to generate HD pixels at four types of positions from field data of an SD signal. The configuration of the interpolation filter used there is classified into a two-dimensional non-separable filter in space that does not separate vertical processing and horizontal processing, and a vertical / horizontal separable filter that performs these processing separately. . FIGS. 11 and 1 show configuration examples of these interpolation filters.
It is shown in FIG.

The non-separable interpolation filter shown in FIG. 11 uses a two-dimensional spatial filter. An SD signal is supplied from the input terminal 81, and the input SD signal
Two-dimensional filter 82 for de1, two-dimensional filter 8 for mode2
3, and are supplied to the two-dimensional filter 84 for mode 3 and the two-dimensional filter 85 for mode 4, respectively. That is, the interpolation processing is executed using an independent two-dimensional filter for each of the HD pixels at the four types of positions. As a result, each filter 8
The outputs 2 to 85 are serialized as HD signals in the selector 86, and the output HD signal is extracted from the output terminal 87.

The interpolation filter shown in FIG. 1 uses a vertical / horizontal separable filter. An SD signal is supplied from an input terminal 91, and the input SD signal is supplied to the vertical interpolation filters 92 and 93 by the HD signal.
The scan line data of the book is generated. For example, the vertical interpolation filter 92 performs processing for mode 1 and mode 2, and the vertical interpolation filter 93 performs processing for mode 3 and mode 4.
Is performed.

When these processes are performed, the output signals from the vertical interpolation filters 92 and 93 are supplied to horizontal interpolation filters 94 and 95. This horizontal interpolation filter 9
In 4 and 95, a horizontal filter is used for each scanning line.
The HD pixels at the different positions are interpolated and supplied to the selection unit 96. The selector 96 serializes the supplied HD signal, and extracts the output HD signal from the output terminal 97.

However, in the conventional up converter, even if an ideal filter is used as an interpolation filter, the number of pixels is increased, but the spatial resolution is not different from that of the SD signal. Actually, since an ideal filter cannot be used, there is a problem that an HD signal whose resolution is lower than that of an SD signal can only be generated.

Therefore, in order to solve these problems,
It has been proposed to apply a classification adaptive process for interpolation. This class classification adaptive processing performs a class classification based on, for example, a feature of the luminance level of the input SD signal, and performs linear interpolation between a prediction coefficient corresponding to the classified class and a plurality of pixel values of the input SD signal forming the prediction tap. This is a process of generating an HD signal by primary combination. At this time, the used prediction coefficients are obtained by learning in advance for each class.

[0009]

As described above, it is possible to generate an HD signal having a higher resolution than an SD signal by the class classification adaptive processing. The up-converter based on the class classification adaptive processing proposed earlier involves an input SD signal related to processing on a component signal such as a luminance signal.

However, it is also possible to directly convert a composite SD signal into an HD signal. When a composite SD signal is input, this composite SD
The method of converting a signal to a component SD signal and then converting it to an HD signal and the method of directly converting an input composite SD signal to an HD signal differ in the performance of conversion depending on the degree of image motion. is there. In the previously proposed classification adaptive processing, only one method is used, and the two are not adaptively selected and switched, so that the conversion performance is insufficient.

Accordingly, the present invention provides an HD signal having a high conversion performance by, for example, switching an HD signal obtained by a plurality of different processes in accordance with the characteristics of an input composite SD signal or performing weighted addition. It is an object of the present invention to provide an image signal conversion device and method capable of generating the image signal.

[0012]

According to a first aspect of the present invention, there is provided an image signal converting apparatus for converting a first digital image signal into a second digital image signal having a larger number of pixels than the first digital image signal. A plurality of preprocessing means for generating a plurality of signals by performing different preprocessing on the digital image signal; and a plurality of preprocessing means for performing a classification adaptive process on each of the plurality of signals to generate a plurality of converted signals. An image signal conversion apparatus comprising: a classification adaptive processing means; a control signal generation means for generating a control signal from a first digital image signal; and a switching means for switching a plurality of conversion signals by the control signal. is there.

Further, according to a second aspect of the present invention, there is provided an image signal converting apparatus for converting a first digital image signal into a second digital image signal having a larger number of pixels than the first digital image signal. A plurality of preprocessing means for generating a plurality of signals by performing different preprocessing on the plurality of signals; and a plurality of class classification adaptive processes for performing a plurality of conversion processing on the plurality of signals to generate a plurality of converted signals. Means, control signal generating means for generating a control signal from the first digital image signal, and addition calculating means for performing weighted addition of a plurality of converted signals in accordance with the control signal. An image signal conversion device.

According to a tenth aspect of the present invention, there is provided an image signal conversion method for converting a first digital image signal into a second digital image signal having a larger number of pixels than the first digital image signal. Generating a plurality of signals by performing different pre-processing on the plurality of signals; performing a class classification adaptive process on each of the plurality of signals to generate a plurality of converted signals; An image signal conversion method characterized by comprising a step of generating a signal and a step of switching a plurality of conversion signals by a control signal.

Further, the invention according to claim 11 is the first invention.
Generating a plurality of signals by subjecting the first digital image signal to different pre-processing in an image signal conversion method for converting the first digital image signal into a second digital image signal having a larger number of pixels than that of the first digital image signal And apply class classification adaptive processing to each of the plurality of signals,
Generating a plurality of converted signals; generating a control signal from the first digital image signal; and performing weighted addition to the plurality of converted signals in accordance with the control signal. Image signal conversion method.

Different pre-processing is performed on the input SD signal, and up-conversion is performed on the pre-processed signal in the class classification adaptive processing. And input S
A high-performance output HD signal can be obtained by switching the converted signal obtained by appropriate conversion processing or performing weighted addition according to the control signal based on the D signal.
More specifically, three-dimensional Y / C separation is performed on the input composite signal, up-conversion is performed in the classification adaptive processing using the separated Y signal or C signal, and the classification is performed on the input composite signal. Up-conversion is performed in adaptive processing. Then, the conversion performance can be improved by switching the up-converted converted signal or performing weighted addition according to the control signal based on the input composite signal.

[0017]

Embodiments of the present invention will be described below in detail with reference to the drawings. First, in order to facilitate understanding of the present invention, an up-converter using the previously proposed classification adaptive processing will be described. In an upconverter using the classification adaptive processing, the input SD
The input signal is classified into several classes based on the characteristics of the signal, and an output HD signal is generated according to an adaptive prediction method for each class generated in advance by learning.

As an example, the input SD as shown in FIG.
A class generation tap is set for a signal (8-bit PCM (Pulse Code Modulation) data), and a class is generated based on the waveform characteristics of the input SD signal. In the example of FIG. 1A, a class is generated with seven taps (seven SD pixels) around the target SD pixel (画素). For example, 1 bit ADRC (Adaptive Dynamic Range) for 7 tap data
When coding is applied, the minimum value of seven pixels is removed based on the dynamic range defined from the data of seven pixels, and then the pixel value of each tap is adaptively quantized by 1 bit, so that 128 classes are generated. Is done.

ADRC is developed as a VTR signal compression system, but is suitable for expressing the waveform characteristics of an input signal with a small number of classes. In addition to the ADRC, the following classification methods can be adopted.

1) Use PCM data directly. 2) Apply DPCM (Differential PCM) to reduce the number of classes. 3) Apply VQ (Vector Quantization) to reduce the number of classes. 4) Frequency transformation (DCT (Discrete Cosine Transform)
Coding), Hadamard transform, Fourier transform, etc.).

The adaptive processing is performed for each of the classified classes. An example of the adaptive processing is a prediction processing using a prediction coefficient for each class generated by learning in advance. FIG. 1B shows an example of a prediction tap used in the prediction processing. In this example, a prediction tap is formed from 13 taps in a frame centered on a target SD pixel. An example of the prediction equation is shown in equation (1).

[0022]

(Equation 1) y ': estimated HD pixel value x _i : SD signal prediction tap pixel value w _i : prediction coefficient

As described above, the HD pixel value is estimated by the product-sum operation of the prediction coefficients generated for each class and the input data, for example, by linear linear combination. FIG. 2 shows a circuit configuration of the classification adaptive processing. An input SD signal is supplied from an input terminal denoted by reference numeral 1, and the supplied input SD signal is supplied to the classifying unit 2 and the prediction tap selecting unit 3. The class classification unit 2 generates a class for the input SD signal based on the class tap as shown in FIG. 1A described above.
The generated class is output from the classification unit 2 with the prediction coefficient RO
M4.

The prediction coefficient ROM 4 outputs a prediction coefficient responding using the generated class as an address. The prediction coefficient is supplied from the prediction coefficient ROM 4 to the prediction calculation unit 5. The prediction tap selection unit 3 selects a prediction tap including 13 taps from the input SD signal as shown in FIG. 1B described above. The prediction tap consisting of the selected 13 taps is
It is supplied from the prediction tap selection unit 3 to the prediction calculation unit 5. The prediction operation unit 5 executes the prediction operation shown in Expression (1) from the supplied prediction coefficient and prediction tap, and outputs the operation result from the output terminal 6.

In this example, in order to predict the HD pixels of mode1 to mode4, the class tap composed of seven pixels shown in FIG. 1A is shared, but the class tap may be changed for each of mode1 to mode4. Similarly, although the prediction tap including 13 pixels shown in FIG. 1B is shared, the prediction tap can be changed based on the class supplied as described above, and the prediction tap is changed for each of mode1 to mode4. It is possible to change it.

The above-described prediction coefficients are generated in advance by learning, and the learning method will be described. An example in which a prediction coefficient based on the linear combination model of Expression (1) is generated by the least square method will be described. The least squares method is applied as follows. As a generalized example, consider the following equation, where X is input data, W is a prediction coefficient, and Y is an estimated value.

Observation equation: XW = Y (2)

(Equation 2)

The least squares method is applied to the data collected by the above-mentioned observation equation (2). In the example of equation (1), n = 13 and m is the number of learning data. Consider the residual equation of equation (4) based on the observation equation of equation (2).

Residual equation: XW = Y + E (4)

(Equation 3)

From the residual equation of equation (4), the most probable value of each w _i is:

(Equation 4) It is considered that the condition for minimizing is satisfied. That is, it is only necessary to consider the condition of Expression (5).

[0031]

(Equation 5)

Considering n conditions based on i in equation (5),
W _1, w ₂ to meet this,..., It may be calculated w _n. Thus, equation (6) is obtained from the residual equation (4).

[0033]

(Equation 6)

Equation (7) is obtained from equations (5) and (6).

(Equation 7)

Then, a normal equation (8) is obtained from the equations (4) and (7).

(Equation 8)

In the normal equation of the equation (8), since the same number of equations as the number n of unknowns can be established, each w _i
Can be obtained. Then, the simultaneous equations are solved using the sweeping-out method (Gauss-Jordan elimination method).

In the learning in this case, the above-mentioned linear first-order model is set between the target signal and the teacher signal, and the prediction coefficients are generated in advance by the least square method. FIG. 3 is a flowchart illustrating an example of the learning method. In this flowchart, the control of the learning process starts from step S1, and in the learning data formation in step S1, for example, learning data is formed from the above-described 13 taps shown in FIG. 1B. Here, a control in which a dynamic range in a block near a target SD pixel is smaller than a predetermined threshold value, that is, a low activity range is not treated as learning data. This is because a dynamic range having a small dynamic range is likely to be affected by noise and an accurate learning result may not be obtained.

At the end of the data in step S2, if the processing of all the input data, for example, the data of one frame or one field has been completed, the control shifts to the prediction coefficient determination in step S5. Step S3
Transfers control to class determination. In the class determination in step S3, as shown in FIG. 1A described above, a class is determined based on the motion evaluation value of the pixel position near the SD pixel of interest. In the normal equation of step S4, the above equation (8)
Is created. After processing all data,
From the end of the data in step S2, the control moves to step S5. In the prediction coefficient determination in step S5, the normal equation is solved by using a matrix solution, and the prediction coefficient is determined. The prediction coefficient is stored in the memory by the registration of the prediction coefficient in step S6, and this flowchart ends. The above is the outline of the classification adaptive processing by the prediction method.

FIGS. 4 and 5 show an example of an upconverter using the above-described classification adaptive processing. This figure 4
FIG. 5 and FIG. 5 show a configuration corresponding to the above-described up-converter in FIG. 11 and FIG. 12, and generate HD pixel data based on the relationship between SD pixels and HD pixels shown in FIG. The input SD signal supplied from the input terminal 10 is supplied to the prediction tap selection unit 11 and the class classification unit 12. The class classification unit 12 generates a class d0 like the above-described class tap shown in FIG. 1A, and generates the class d0 as a prediction coefficient ROM 13 as an address.
Supplied to The prediction tap selection unit 11 selects a prediction tap as shown in FIG. 1B described above. The selected prediction tap is supplied to the mode1 prediction operation unit 14, the mode2 prediction operation unit 15, the mode3 prediction operation unit 16, and the mode4 prediction operation unit 17.

In the prediction coefficient ROM 13, as described above, a prediction coefficient obtained by learning in advance reads a coefficient corresponding to the class d0. The read prediction coefficient is
As d5, the mode1 prediction operation unit 14, the mode2 prediction operation unit 15, the mode3 prediction operation unit 16, and the mode4 prediction operation unit 1
7. In the mode1 prediction operation unit 14, a product-sum operation is executed from the selected prediction tap and the prediction coefficient d5, and an HD pixel d1 of mode1 is generated.
The pixel d1 is supplied to the selection unit 18. Similarly, in the prediction operation units 15, 16, and 17, a product-sum operation is performed from the selected prediction tap and the prediction coefficient d5, and the HD pixel d
2, d3 and d4 are generated and supplied to the selection unit 18.

At this time, each of the prediction operation units 14, 15, 1
In 6 and 17, the product-sum operation of the prediction tap from the prediction tap selection unit 11 and the prediction coefficient d5 is performed. At the time of this product-sum operation, the sign of the prediction coefficient d5 is changed according to the mode.
When the product-sum operation is performed between the mode1 prediction operation unit 14 and the mode2 prediction operation unit 15, for example, the sign of the prediction coefficient d5 is reversed between positive and negative, and similarly, the mode3 prediction operation unit 16 and the mo
Also between the de4 prediction calculation units 17, the sign of the prediction coefficient d5 is reversed between positive and negative, and the product-sum operation is executed. Further, between the mode1 prediction operation unit 14 and the mode3 prediction operation unit 16, for example, the prediction taps are exchanged line-symmetrically with respect to the vertical line on the target SD pixel, and the prediction coefficient d5 and the product-sum operation are executed. To
In the mode2 prediction operation unit 15 and the mode4 prediction operation unit 17,
The prediction taps are exchanged symmetrically with respect to the vertical line on the target SD pixel, and the product-sum operation is executed.

The selecting unit 18 rearranges the HD pixels generated from the target SD pixels in a desired time series, and
The HD pixel is output via. This selection unit 18 includes
Also includes the memory required for sorting. The above is an example of the configuration of the class classification adaptive processing when the prediction calculation of the two-dimensional non-separable structure is used.

On the other hand, FIG. 5 shows an example of the configuration of the class classification adaptive processing in the case of performing the prediction calculation of the vertical / horizontal separable configuration. An input SD signal is supplied from an input terminal 20, and the input SD signal is supplied to a prediction tap selection unit 21 and a class classification unit 22. The class classification unit 22 generates a class d11 like the class tap shown in FIG. 1A described above, and supplies the generated class d11 to the vertical coefficient ROM 23 and the horizontal coefficient ROM 24 as addresses.
The prediction tap selected by the prediction tap selection unit 21 is supplied to the vertical prediction calculation units 25 and 26.

In the vertical coefficient ROM 23, the one corresponding to the class d11 is read out from the prediction coefficients previously obtained by learning as described above, and the read out prediction coefficients are:
The vertical prediction coefficient d16 is supplied to the vertical prediction calculation units 25 and 26. In the horizontal coefficient ROM 24, as described above, the class d is calculated based on the prediction coefficient obtained by learning in advance.
11 is read out, and the read prediction coefficient is used as the horizontal prediction coefficient d17 as the horizontal prediction calculation unit 27.
And 28.

The vertical prediction calculation unit 25 generates a vertical estimated value by a product-sum operation from the selected prediction tap and the vertical prediction coefficient d16. This vertical estimation value is generated at the position of the HD pixel. Next, horizontal class classification adaptive prediction is performed using the vertical estimation value, thereby generating an HD pixel value. The generated vertical estimation value is supplied to the horizontal prediction calculation unit 27 as d12. Similarly, in the vertical prediction operation unit 26, a vertical estimation value d13 is generated by a product-sum operation from the selected prediction tap and the vertical prediction coefficient d16, and supplied to the horizontal prediction operation unit 28.

The horizontal prediction operation section 27 has a memory for storing the supplied vertical estimation value d12, selects a prediction tap from the stored vertical estimation value, and calculates a difference between the selected prediction tap and the horizontal prediction coefficient d17. HD pixels are generated by the product-sum operation. The generated HD pixel is supplied to the selection unit 29 as d14. Similarly, in the horizontal prediction calculation unit 28, the prediction tap selected from the memory and the horizontal prediction coefficient d
The HD pixel is generated by the product-sum operation with 17. The generated HD pixel is supplied to the selection unit 29 as d15. In the above-described prediction calculation units 25, 26, 27, and 28, the signs and / or taps of the prediction coefficients are switched according to the mode.

The vertical and horizontal prediction calculation units 25 to 28
Are divided into two types, one for mode 1 and mode 2 scanning lines and the other for mode 3 and mode 4 scanning lines.
Indicates the former, and the prediction calculation units 26 and 28 indicate the latter. In the selection unit 29, the supplied HD pixels d14 and d15 are finally rearranged through an output terminal 30 by optimal rearrangement. With the above processing, the classification adaptive processing in the case where the prediction calculation is made into the vertical / horizontal separable configuration is realized.

The above-described class classification adaptive processing is a technique called a prediction method using a prediction operation as shown in the example of equation (1). In addition to this, there is a class classification adaptive process using the center of gravity method. The centroid method is not a prediction calculation method that outputs a prediction calculation value, but a method that uses an average value of a target signal distribution corresponding to each class as an output value. In the prediction calculation method, a prediction coefficient corresponding to each class is generated in advance by learning. In the centroid method, an output value corresponding to each class is generated in advance by learning. Therefore, the basic configuration of the centroid method is a simple one in which the class classification unit and the predicted value ROM are connected in series. The centroid method has a lower degree of freedom in prediction than the prediction calculation method, and thus is often inferior to the prediction method in overall performance, but has the advantage of a very small amount of hardware. Therefore, these are used properly depending on the application.

By using such a classification adaptive processing, an up-converted image with improved resolution can be obtained from the up-converter using the above-mentioned conventional interpolation filter.

The above-described classification adaptive processing can be applied to a component SD signal, for example, a luminance signal.
When directly up-converting a composite signal, it is necessary to consider characteristics specific to the composite signal, such as the phase of the subcarrier, in addition to the classification adaptive processing.
6 shows an example of a subcarrier phase when a composite signal is sampled at a sampling frequency of 4 fsc, FIG. 6A shows a subcarrier phase of a pixel in an even frame, and FIG. 6B shows a subcarrier phase in an odd frame. 4 shows a subcarrier phase of a pixel.

As shown in the figure, the pixel position, line number,
The phase changes based on the field number and the frame number. In the NTSC composite signal, the phase of the subcarrier is inverted for each line and each frame, and the phase returns to the original state in units of four fields. In an up-conversion from a component signal, it is general to sequentially form class taps and prediction taps in the vicinity of a pixel to be processed. However, as is clear from FIG. 6, each pixel value in the vicinity of the composite signal is
Reflects subcarrier fluctuations. Even in the case of a flat signal having the same color and the same luminance value, in a composite signal, each pixel value changes depending on the subcarrier phase.

Therefore, in the up-conversion from the composite signal, a tap configuration in consideration of the subcarrier phase is essential instead of sequentially taking a class tap or a prediction tap from the vicinity of the target pixel. The processing configuration for performing up-conversion from a composite signal using the class classification adaptive processing is realized by the configuration shown in FIG. 2 as in the case of the component signal, but it is necessary to change the structure of the class tap and the prediction tap.

According to the present invention, when the input SD signal is a composite signal or the like, high-performance up-conversion is realized by applying different pre-processing and then applying the class classification adaptive processing to each of them. A conversion signal is generated by using a classification adaptive process for each signal obtained by performing different pre-processing on the input signal, and the conversion signal is switched based on a separately detected control signal.
Get the signal. One embodiment of the present invention is shown in FIG.

The input SD signal d20 supplied from the input terminal 41 is supplied to the control signal generator 42, the preprocessors 43 and 45,
47. The control signal generator 42 generates a control signal d21 from the supplied input SD signal. The generated control signal d21 is supplied from the control signal generator 42 to the selector 49 in units of pixels or blocks. In the pre-processing units 43, 45 and 47, three different types of pre-processing are performed on the input SD signal, and the pre-processed signals are classified as adaptive classification processing units 44, 46 and d24 as d22, d23 and d24. 48. When the classification adaptive processing units 44, 46, and 48 use the prediction method, the configuration is as shown in FIG. The output signals d25, d26 and d27 of the classification adaptive processing units 44, 46 and 48 are supplied to the selection unit 49. The selector 49 selects the output signals d25, d26 and d27 based on the control signal d21 from the control signal generator 42.
The selected main signal is output from the output terminal 50 to the HD signal d2.
8 is output.

FIG. 8 is a block diagram showing another embodiment for performing weighted addition on a plurality of processing signals. 8, the same blocks as those in FIG. 7 described above are denoted by the same reference numerals, and description thereof will be omitted. The control signal d21 from the control signal generation unit 42 is supplied to the weight generation unit 51 in pixel units or block units. The weight generating unit 51, the control signals d21 supplied, weighting factor a _i of the threshold determination three kinds of signals, such as by d25, d26 and d27 are generated. The generated addition weight _ai is supplied to the weighted addition operation unit 52 as d31. In the weighted addition operation unit 52, the operation of Expression (9) is performed on the output signals d25, d26, and d27 from the classification adaptive processing units 44, 46, and 48. The HD signal obtained by the calculation result of the equation (9) is output as an output terminal 53 as d32.
Output from

Y ″ = a ₀ × y ′ ₀ + a ₁ × y ′ ₁ + a ₂ × y ′ ₂ (9) y ″: final HD output value y ′ _i : estimated HD pixel value by each pre-processing a _i : addition Weight (however, sum is assumed to be 1.0)

The addition weights a _i can be generated in advance by learning as targets of the classification adaptive processing. In that case, the weight generating section 51 has a built-in weight coefficient ROM.

As a first example of the above-mentioned pre-processing units 43, 45 and 47, processing of one-dimensional Y / C separation, two-dimensional Y / C separation and three-dimensional Y / C separation can be mentioned. The input composite SD signal is subjected to Y / C separation, Y and C signals of the components are generated, and the component signals are subjected to class classification adaptive processing, and up-conversion is performed.

In the second example of the pre-processing units 43, 45 and 47, one of the pre-processing units 43, 45 and 47 has a Y / C
Separation is performed, and the generated component Y signal is subjected to class classification adaptive processing to generate an up-converted component Y signal, and the other is directly subjected to class classification adaptive processing without performing Y / C separation on the composite SD signal. To generate an up-converted component C signal.

Further, in the third example of the pre-processing units 43, 45 and 47, the averaging in the time direction is performed on a stationary part, that is, a frame cyclic noise reducer is applied, and a one-dimensional or two-dimensional A dimensional median filter is applied, and a quasi-stationary part between the two is subjected to a three-dimensional median filter, subjected to a class classification adaptive process, and capable of up-conversion.

Next, FIG. 9 shows a block diagram of a more specific configuration example of the other embodiment described above. The input SD signal d40 supplied from the input terminal 61 is a composite signal in which a carrier chrominance signal is superimposed on a luminance signal. The subcarriers of the carrier chrominance signal have a phase relationship as shown in FIG. The input SD signal d40 is supplied to a motion amount detection unit 62, processing units 64 and 66, and a preprocessing unit 65. The movement amount detection section 62 detects the movement amount of the input SD signal d40. Since an ideal motion amount cannot be detected in a composite signal, a luminance signal is pseudo-generated by a simple low-pass filter or the like, and the motion amount is detected using the luminance signal. The detected motion amount is supplied from the motion amount detection unit 62 to the weight generation unit 63 as d41.

The weight generator 63 generates weights for weighted addition operation according to the amount of motion d41. Briefly,
A threshold value is determined for the amount of motion, and a weight is generated. This threshold is empirically determined while considering the final image quality. As a method of generating weights, a method of preparing an HD signal serving as a teacher signal, generating weights by learning in advance, and storing the weights in a ROM can be considered. In this method, the process becomes the classification adaptive processing itself. Simply, the amount of motion becomes a class, and a weight corresponding to the amount of motion is read out. The weight d45 is supplied from the weight generation unit 63 to the weighted addition operation unit 67.

In the configuration shown in FIG. 9, the processing unit 64 includes a class classification unit 69, a prediction coefficient ROM 70, and a prediction calculation unit 7.
1 In the processing section 64, the supplied input composite SD signal d40 is supplied to the classification section 69. In the classifying section 69, the input composite S
A class is generated according to the D signal d40. The generated class is supplied from the class classification unit 69 to the prediction coefficient ROM 70 as d42. In the prediction coefficient ROM 70, a prediction coefficient corresponding to the class is read. This prediction coefficient is obtained in advance by learning, and is supplied to the prediction calculation unit 71 as d46.

In the pre-processing unit 65, for example, in order to suppress the image quality deterioration peculiar to the composite signal, the Y / C separation by the three-dimensional Y / C separation is executed.
43 is supplied to the prediction calculation unit 71. Prediction calculation unit 7
1, the Y signal d43 from the preprocessing unit 65 and the prediction coefficient d
The product-sum operation with 46 is performed, and the operation result is supplied to the weighted addition operation unit 67 as d48. As described above, the processing unit 64 applies the classification adaptive processing to the separated Y signal d43, and outputs the calculation result as d48. This processing unit 64 corresponds to the classification adaptive processing unit shown in FIG. Further, although not shown, it is also possible to perform signal conversion on the C signal by class classification adaptive processing.

On the other hand, the processing unit 66 comprises a prediction calculation unit 72, a class classification unit 73, and a prediction coefficient ROM 74, and performs signal conversion directly by class classification adaptive processing without performing preprocessing. In the processing section 66, the supplied input composite SD signal d40 is
And supplied to the class classification unit 73. Classifier 7
In 3, the class is generated according to the input composite SD signal d40. The generated class is supplied from the class classification unit 73 to the prediction coefficient ROM 74 as d44.
In the prediction coefficient ROM 74, a prediction coefficient corresponding to the class is read. This prediction coefficient is obtained in advance by learning, and is supplied to the prediction calculation unit 72 as d47. In the prediction operation unit 72, a product-sum operation of the input SD signal d40 and the prediction coefficient d47 is performed.
Is supplied to the weighted addition operation unit 67.

In the weighted addition operation unit 67, the d4 from the processing unit 64 is calculated according to the weight d45 from the weight generation unit 63.
8 and d49 from the processing unit 66 are calculated. In the weighted addition operation unit 67, at places where the amount of motion is large, the input composite SD signal is not subjected to Y / C separation or the like, but is directly subjected to the class classification adaptive processing, and the weight of the up-converted d49 is obtained. Becomes larger. Further, in a portion where the amount of motion is small, the input composite SD signal is subjected to three-dimensional Y / C separation, the resulting component Y signal is subjected to class classification adaptive processing, and the weight of the up-converted d48 is changed. growing. In this case, the above equation (9) is a calculation in the case of two inputs, and the calculation result is taken out from the output terminal 68 as d50.

In the configuration shown in FIG. 9, a component Y signal is generated at a portion with a small amount of motion where a relatively stable Y signal can be obtained by three-dimensional Y / C separation, and the generated component Y is generated. Apply a class classification adaptive process to the signal. Otherwise, no processing is performed on the composite signal, and the classification adaptive processing is performed directly. Therefore, in order to avoid image quality deterioration such as dot interference peculiar to the composite signal caused by superposition of subcarriers, it is possible to improve the final image quality.

In the above-described embodiment, a method using a prediction operation in the class classification adaptive processing has been described. However, the present invention is not limited to this method, and it is also possible to use a centroid method. Is also possible.

[0069]

According to the present invention, a plurality of converted signals obtained by subjecting an input SD signal to different pre-processing and then performing a classification adaptive process are converted to a control signal obtained from the input SD signal. By performing weighted addition or switching according to the signal, high-performance signal conversion can be performed.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing an arrangement of pixels for explaining a class classification according to the present invention.

FIG. 2 is a block diagram showing a general configuration example for explaining a class classification according to the present invention.

FIG. 3 is a flowchart for learning a prediction coefficient according to the present invention.

FIG. 4 is a block diagram showing an example of a two-dimensional non-separable upconverter using a class classification according to the present invention;

FIG. 5 is a block diagram showing an example of an upconverter having a vertical / horizontal separable configuration using a class classification according to the present invention.

FIG. 6 is a schematic diagram illustrating an example of phases of subcarriers of a composite signal of an even frame and an odd frame according to the present invention.

FIG. 7 is an embodiment of an up-converter to which the present invention is applied.

FIG. 8 is another embodiment of the upconverter to which the present invention is applied.

FIG. 9 is a specific example of an upconverter to which the present invention is applied.

FIG. 10 is an arrangement diagram showing an arrangement of SD pixels and HD pixels.

FIG. 11 shows a conventional two-dimensional non-separable upconverter.

FIG. 12 shows a conventional vertical / horizontal separable upconverter.

[Explanation of symbols]

62: motion amount detection unit, 63: weight generation unit, 6
4, 66 ... processing unit, 65 ... pre-processing unit, 67 ...
.. Weighted addition operation units, 69, 73 ... class classification units, 70, 74 ... prediction coefficient ROMs, 71, 72 ...
・ Prediction calculation unit

Claims (11)

[Claims] 1. An image signal conversion apparatus for converting a first digital image signal into a second digital image signal having a larger number of pixels than the first digital image signal, wherein different preprocessing is performed on the first digital image signal. A plurality of pre-processing means for generating a plurality of signals according to: a plurality of class classification adaptive processing means for performing a class classification adaptive process on each of the plurality of signals to generate a plurality of converted signals; An image signal conversion device comprising: control signal generation means for generating a control signal from a digital image signal; and switching means for switching the plurality of conversion signals by the control signal. 2. An image signal conversion device which converts a first digital image signal into a second digital image signal having a larger number of pixels than the first digital image signal, wherein different preprocessing is performed on the first digital image signal. A plurality of pre-processing means for generating a plurality of signals according to: a plurality of class classification adaptive processing means for performing a class classification adaptive process on each of the plurality of signals to generate a plurality of converted signals; An image signal conversion device comprising: control signal generation means for generating a control signal from a digital image signal; and addition operation means for performing weighted addition on the plurality of conversion signals according to the control signal. . 3. The image signal conversion device according to claim 2, wherein the weighting factor is generated by using a class classification adaptive process. 4. The image signal conversion device according to claim 1, wherein the class classification adaptive processing uses a prediction operation. 5. The image signal conversion device according to claim 1, wherein the class classification adaptation process uses a centroid method. 6. The image signal conversion device according to claim 1, wherein the class classification adaptation process uses a prediction operation and a centroid method. 7. The image signal conversion device according to claim 1, wherein the plurality of pre-processing units are one-dimensional Y / C separation, two-dimensional Y / C separation, or three-dimensional Y / C separation.
An image signal conversion device comprising at least two of the separations. 8. The image signal conversion device according to claim 1, wherein the plurality of pre-processing units perform Y / C separation on the first digital image signal, and output a component signal. An Y / C separation unit for performing a Y / C separation on the first digital image signal and outputting a component signal without performing the Y / C separation. 9. The image signal conversion device according to claim 1, wherein the plurality of preprocessing units are a noise reducer applied to a stationary part, a noise reducer applied to a quasi-stationary part, and a moving part. An image signal conversion device comprising at least two of the noise reducers applied to the image signal. 10. An image signal conversion method for converting a first digital image signal into a second digital image signal having a larger number of pixels than the first digital image signal, wherein different preprocessing is performed on the first digital image signal. Generating a plurality of converted signals by performing class classification adaptive processing on each of the plurality of signals to generate a plurality of converted signals; and generating a control signal from the first digital image signal. Selecting the plurality of conversion signals according to the control signal. 11. An image signal conversion method for converting a first digital image signal into a second digital image signal having a large number of pixels, wherein the first digital image signal is subjected to different preprocessing. Generating a plurality of signals; performing a classification adaptive process on each of the plurality of signals to generate a plurality of converted signals; and generating a control signal from the first digital image signal. Performing a weighted addition in accordance with the control signal to the plurality of converted signals.