JP3533697B2 - Adaptive filter device and adaptive filter processing method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive filter device for classifying a video signal including a luminance signal and a color signal and performing an adaptive filter process according to the class, and an adaptive filter processing method thereof.

[0002]

2. Description of the Related Art A color television signal used in a general color television or the like includes an NTSC system or P
AL method is adopted.

Both the NTSC and PAL color television signals are luminance signals (Y signals).
And a color signal (C signal) composed of an I (wideband) signal and a Q (narrowband) signal (in the PAL system, an RY signal and a BY signal) that are color difference signals are decoded colors that are multiplexed by frequency interleaving. It is a television signal. This decoded color television signal is called a composite signal.

On the other hand, a color television separated into three signals of a luminance signal Y in the NTSC system and color difference signals I and Q (luminance signal Y, color difference signals RY and BY signals in the PAL system). The John signal is called the component signal. To generate a component signal from the composite signal, use a bandpass filter,
This can be obtained by separating the composite signal into a luminance signal (Y signal) and a color signal (C signal) (Y / C separation), and further converting the color signal C into an I signal and a Q signal.

As a method of separating the composite signal from Y / C, some band-limiting filters are used instead of the band-pass filter, and outputs from these band-limiting filters are adaptively selected. Alternatively, a method of mixing these outputs has also been proposed.

It is known that the band of the luminance signal and the band of the chrominance signal of these composite signals are not steadily multiplexed with respect to position and time. Therefore, when the Y / C separation is performed by fixing the filter coefficient of the band limiting filter, the luminance signal Y and the chrominance signal C cannot be completely separated.
Also, in the Y / C separation method by adaptive selection and mixing,
The state of the signal to be taken out is too large for the number of band limiting filters, again in this case the complete separation of the composite signal is not possible.

In order to solve such a problem, the applicant of the present invention has a class of digital signal data which can easily and completely perform Y / C separation according to the specification and drawings of Japanese Patent Application No. 5-241186. A classifying device, an adaptive Y / C separating device, a classifying method of digital signal data, and an adaptive Y / C separating method are proposed.

According to the present invention, the digital signal data is divided into blocks, divided into classes, and the color signals are separated from the frequency-multiplexed signal data by using the filter coefficient obtained by converting the digital signal data into the numerical values corresponding to the class divisions. Is improving.

[0009]

By the way, in class classification by correlation detection, correlation in a desired direction is detected by subtraction of in-phase pixels. In the determination regarding the pixels in the field of the class classification by this correlation detection method, for example, taps as shown in FIG. 9A are used. That is, it can be seen that the in-phase pixels (∘) are arranged every other line in the vertical direction because the field signal is handled, and are arranged only every three pixels in the horizontal direction.

Further, as shown in FIG. 9B, the tap used in the predictor for Y / C separation is a pixel (H0, V0) on the vertical line with respect to the pixel (H0, V0) in the vertical direction. , V-1),
(H0, V + 1), and pixels (H-2, V0) and (H + 2, V0) that are one pixel apart from the pixel (H0, V0) in the horizontal direction.

By the way, FIG. 9 performed for class classification
The tap for correlation detection in (a) and the tap in FIG. 9 (b) used for actual Y / C separation are different taps. Since the Y / C separation process is performed with different taps in this way, there is a high possibility that the correlation determination will be erroneous.

Further, when learning the filter coefficient for each class, if the correlation judgment is performed by collecting only the data having the characteristic such as the correlation in the vertical direction, the judgment is likely to be erroneous. There is a fear that the filter coefficient of will not be learned.

Therefore, the present invention has been made in view of the above-mentioned circumstances, and improves the filter performance so that a filter characteristic adapted to an input signal can be obtained, and, for example, an error due to correlation determination is suppressed. An object of the present invention is to provide an adaptive filter device and an adaptive filter processing method capable of performing the same.

[0014]

In order to solve the above-mentioned problems, an adaptive filter device according to the present invention is an adaptive filter device which changes a filter characteristic in accordance with an input signal. In an adaptive filter device that changes, a class classification unit that classifies an input signal into blocks and classifies it into any one of a plurality of classes, and the block output signal from this class classification unit and the classified class are input. And a filter processing unit that has a filter characteristic based on the filter coefficient from the filter coefficient output unit and outputs the filter coefficient, , A prediction filter means for linearly predicting a true predicted value for class classification, and an output signal from this prediction filter means According to the class classified by the class classification means of the constituent part of the preceding stage among the constituent parts of these plural stages, the constituent part having a class classification means for classifying into any of the classes The filter coefficient is output from the filter coefficient output means of the component of the next stage, and the filter processing unit outputs the filter coefficient according to the class classified by the class classification unit of the final stage of the class classification unit. And a predictive filter unit that has a filter characteristic based on the filter coefficient from the filter coefficient output unit and that linearly predicts a true predicted value for an input signal. There is.

Here, the input signal is a signal in which the luminance signal Y and the color signal C are multiplexed, and this input signal is separated into the luminance signal Y and the color signal C by the filtering process. The first linear prediction unit uses a fixed filter coefficient. The first linear prediction unit may use a variable characteristic filter whose filter characteristic changes based on the filter coefficient obtained according to the classification result of the input signal.

Further, in order to solve the above-mentioned problems, the adaptive filter processing method according to the present invention is an adaptive filter processing method in which the filter characteristic is changed according to the input signal, and the input signal is divided into a plurality of classes. A class classification processing step of classifying into any of the above, and a filtering processing step in which the blocked output signal from this class classification processing step and the classified class are input,
In the class classification processing step, a filter coefficient output step of outputting a filter coefficient, and a filter characteristic based on the filter coefficient from the filter coefficient output step, the input signal is a true value for class classification. Of the prediction process of linearly predicting the prediction value of, and a classifying process of classifying the output signal from the prediction filter process into any one of a plurality of classes are repeated a plurality of times, According to the class classified by the class classification process of the previous partial processing step among the partial processing steps of, the filter coefficient in the filter coefficient output step of the next partial processing step is output, and the filter processing step is performed. Then, the filter that outputs the filter coefficient according to the class classified by the last class classification process of the above class classification process. A coefficient output step has a filter characteristic based on the filter coefficients from the filter coefficient output step is characterized by having a prediction filter step of linear prediction a prediction value of the true value for the signal to be inputted.

[0017]

In the adaptive filter device according to the present invention, the class classification unit predicts the predicted value of the true value for class classification for the input signal by the filter characteristic based on the filter coefficient from the filter coefficient output means. The filter means performs linear prediction, and the output signal from the prediction filter means is classified into any one of a plurality of classes by a plurality of stages, and the preceding stage of these plurality of stages is classified. The filter coefficient of the constituent part of the next stage is output according to the specified class, and the filter coefficient of the filter processing unit is output with the class code classified into the corresponding class by the class classification circuit of the constituent part of the final stage as an address. Means (ROM). In the ROM, the filter coefficient of this address is supplied to the predictive filter means and subjected to a filtering process, whereby the filter coefficient adapted to each tap for the input signal is set to obtain a desired filter characteristic.

By filtering the input signal in which the luminance signal Y and the color signal C are multiplexed, a sufficiently appropriate luminance signal Y and color signal C are separated.

By using a filter having a fixed filter coefficient as the prediction filter means of the first stage constituent part of the class classification unit, the predicted value of the true value for class classification can be obtained.

The variable characteristic that the filter coefficient changes based on the filter coefficient obtained according to the class classified by the class classification means of the preceding constituent part in the predictive filter means of the constituent parts of the second and subsequent constituent parts of the class classification part By using a filter, it becomes possible to perform more appropriate filter processing.

Further, in the adaptive filter processing method according to the present invention, in the class classification processing step, the input signal is classified under the filter characteristic based on the filter coefficient output from the filter coefficient output means. The partial processing step of linearly predicting the predicted value of the true value and classifying the output signal into one of a plurality of classes is repeated a plurality of times, and the previous partial processing step among the plurality of partial processing steps is performed. The filter coefficient of the next partial processing step is output according to the class classified by, the filter coefficient is output according to the class classified by the final partial processing step, and the output filter coefficient is output. It has a filter characteristic based on, and linearly predicts the true predicted value for the input signal, so that the adaptive filter at each tap for the input signal Set the coefficient so desired filter characteristics can be obtained.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of an adaptive filter device and an adaptive filter processing method according to the present invention will be described below with reference to the drawings. Here, a preferred example of the adaptive Y / C separation device will be given in which the adaptive filter device of the present invention is applied to classify a video signal including a luminance signal and a chrominance signal and adaptive filter processing is performed according to the class classification. Luminance signal Y
The adaptive Y / C separation of frequency-multiplexed digital signal data including the color signal C and the color signal C will be described.

The adaptive Y / C demultiplexer classifies the output signal of the prediction filter into one of the classes as schematically shown in FIG. 1, and the adaptive Y / C corresponding to the classified class.
A / C separation device 1 for separating a color signal C from the supplied digital composite signal data, an I / Q signal separation processing unit 2 for separating the color signal C into an I signal and a Q signal, And the adder 3 for extracting the luminance signal Y.

The Y / C separation device 1 includes a class classification unit 1a for classifying digital composite signal data (hereinafter referred to as signal data) supplied via an input terminal 4, and a classification output by the class classification unit 1a. The data P for each class and the adaptive Y / C separation processing unit 1b for performing Y / C separation processing adapted to the signal data B divided into blocks and divided into classes.

The class classification unit 1a in the Y / C separation device 1 is, for example, as shown in FIG. 2, a block processing unit 11, a ROM 12 for outputting a fixed filter coefficient, and a linear prediction as a first prediction filter means. The unit 13, the class classification circuit 14 as the class classification means, the ROM 15 that outputs the filter coefficient according to the output signal from the class classification circuit 14, the linear prediction unit 16 based on the filter coefficient from the ROM 15, and the class classification circuit 17. It is configured.

Here, the Y / C separation device includes a block processing section 11, a ROM 12, and a ROM 12 which are surrounded by at least a broken line.
The linear predicting unit 13 and the class classifying circuit 14 include the class classifying unit 1.
a may be configured to improve the filter performance by supplying the adaptive filter coefficient at each tap with respect to the input signal to the adaptive Y / C separation processing unit 1b which is the second prediction filter means described later. You can

The block processing section 11 is a frequency-multiplexed and scan-converted NTS supplied via the input terminal 4.
The C system one-dimensional signal data is divided into two-dimensional signal data, which are supplied to the linear prediction unit 13 and the class classification circuit 14 in consideration of the timing. The block-shaped signal data is represented by a symbol B.

The ROM 12 stores the first filter coefficient for the tap prepared in advance as an initial value. In the ROM 12, when the blocked signal data from the blocking processing unit 11 is supplied to the linear prediction unit 13, the filter coefficient stored according to the supply of the signal data is supplied to the linear prediction unit 13.

The linear predictor 13 uses the linear predictive equation for the blocked signal data X (i) and the filter coefficient a (0, i) corresponding to, for example, 18 taps.

[0030]

[Equation 1]

Then, the Y / C separation predicted value S (0) is calculated by substituting into The Y / C separation prediction value output by the linear prediction unit 13 is S (0) at n = 0 indicating the first stage. The linear prediction unit 13 outputs the Y / C separated prediction value S (0) to the class classification circuit 14.

The class classification circuit 14 uses the Y / C based on the blocked signal data and the coefficient at the desired tap.
The difference between the separation value and the Y / C separation prediction value S (0) obtained by the estimation supplied from the linear prediction unit 13 is estimated as the error amount. Whether or not the estimated error amount is data suitable for Y / C separation at this tap is determined by threshold processing. The class classification circuit 14 uses the predetermined numerical value thus obtained according to the determination, so-called class code (class information) P (0), as an address in the ROM 1
Output to 5. In the threshold processing, for example, the average value of the signal data in the block may be used as the threshold.

As a circuit configuration for performing such class classification, there is a configuration shown in FIG. 3, for example. The class classification circuit of FIG. 3 is provided with, for example, five coefficient circuits 141a to 141e having desired Y / C separation tap coefficients. The product-sum units 142a to 142e are provided corresponding to the five coefficient circuits 141a to 141e. Accumulator 142a
To 142e, the block-wise signal data B respectively supplied and the coefficients from the coefficient circuits 141a to 141e are subjected to sum-of-products operation to obtain desired Y / C separation values R1 to R5, and are supplied to the comparators 143a to 143e. Supply. Comparator 143a-
The 143e inputs the Y / C separated values R1 to R5 and the Y / C separated predicted value S (0) from the linear prediction unit 13 and compares them. In the comparators 143a to 143e, the error amount | R1− obtained from this comparison in order to perform the threshold processing.
S (0) |, | R2-S (0) |, | R3-S (0)
|, | R4-S (0) |, | R5-S (0) | are estimated and compared with the threshold value TH (0) as described above.

In the comparison of the comparators 143a to 143e, the level of each bit is set to "1" or "0" according to the following conditions, for example. That is, the comparator 143a
When the error amount | R1-S (0) | is smaller than the threshold value TH (0), the bit 0 is set to the level "1", and the bit 0 is set to the level "0" otherwise. Comparator 143b
Sets the bit 1 to the level "1" when the error amount | R2-S (0) | is smaller than the threshold value TH (0), and otherwise sets the bit 1 to the level "0". Comparator 143c
Sets the bit 2 to the level "1" when the error amount | R3-S (0) | is smaller than the threshold value TH (0), and otherwise sets the bit 2 to the level "0". Comparator 143d
When the error amount | R4-S (0) | is smaller than the threshold value TH (0), the bit 3 is set to level "1", and the bit 3 is set to level "0" otherwise. Comparator 143e
When the error amount | R5-S (0) | is smaller than the threshold value TH (0), the bit 4 is set to level "1", and the bit 4 is set to level "0" otherwise. In this way, the so-called class code P (0) is output as binary data (bit 4, bit 3, bit 2, bit 1, bit 0) and 5 bits.

The ROM 15 stores filter coefficients obtained by learning the blocked signal data B using, for example, the least square method. Also, RO
M15 has various Y / C separation taps. The class code P (0) output from the class classification circuit 14 is supplied to the ROM 15 as address data, and the data of the address corresponding to this is supplied to the linear prediction unit 16
Output to.

The linear predictor 16 performs the same linear prediction as the linear predictor 13 described above. The second prediction of the color signal C is performed using the equation (1) (n = 1). The linear prediction unit 16 uses the Y / C separation predicted value S of the second color signal C.
(1) is output to the class classification circuit 17.

The class classification circuit 17 uses the first color signal C
Y / C separation predicted value S (0) of the class classification circuit 17 and the Y / C separation predicted value S obtained by the estimation supplied from the linear prediction unit 16 according to the coefficient at the desired tap of the class classification circuit 17. The difference from (1) is estimated as the amount of error. The class classification circuit 17 determines whether the estimated error amount is data suitable for Y / C separation at this tap by threshold processing. In this way, a predetermined numerical value obtained according to the judgment, so-called class code (class information)
ROM 18 shown in FIG. 4 with P (1) as address data
Output to. The threshold processing is the same as described above.
The configuration of the class classification circuit 17 is the same as that shown in FIG.

As shown in FIG. 3, the class classification circuit has, for example, 5 coefficient circuits each having a coefficient of a desired Y / C separation tap.
One is provided. A product-sum device is provided corresponding to these five coefficient circuits, and the product-sum device is represented by performing a product-sum operation on the Y / C separation predicted value S (0) supplied and the coefficient from the coefficient circuit. The Y / C separation value is calculated and supplied to each comparator. Each of the comparators represents the data R1 to R5 corresponding to each Y / C separation predicted pixel position represented using the Y / C separation prediction value S (0) and the Y / C separation from the linear prediction unit 16. The predicted value S (1) is input and compared. In each comparator, the amount of error | R1-S (1) |, | R2-S (1) |, | R3- obtained from this comparison in order to perform threshold processing.
S (1) |, | R4-S (1) |, | R5-S (1) |
Is compared with the threshold value TH (1) as described above.

In the comparison of each comparator, the level of each bit is set to "1" or "0" according to the following conditions, for example. That is, the comparator corresponding to the comparator 143a sets the bit 0 to the level "1" when the error amount | R1-S (1) | is smaller than the threshold value TH (1), and otherwise sets the bit 0 to the level. Set to "0". Comparator 143b
When the error amount | R2-S (1) | is smaller than the threshold value TH (1), the comparator corresponding to
Otherwise, bit 1 is set to level "0". The comparator corresponding to the comparator 143c determines the error amount | R.
When 3-S (1) | is smaller than the threshold value TH (1), bit 2 is set to level "1", and bit 2 is set to level "0" otherwise. The comparator corresponding to the comparator 143d sets the bit 3 to the level "1" when the error amount | R4-S (1) | is smaller than the threshold value TH (1), and otherwise sets the bit 3 to the level "0". " Comparator 143e
When the error amount | R5-S (1) | is smaller than the threshold value TH (1), the comparator corresponding to
In other cases, bit 4 is set to level "0". In this way, the so-called class code P (1) is
Binarized data (bit 4, bit 3, bit 2, bit 1, bit 0) and 5 bits are output.

The class classification unit 1a shown in FIG. 1 performs Y / C separation processing in which the classified data P (1) for each class and the Y / C separation prediction value S (1) are applied as blocked signal data B. It is supplied to the adaptive Y / C separation processing unit 1b. The adaptive Y / C separation processing unit 1b includes a ROM 18 that outputs a filter coefficient according to a class and a linear prediction unit 19 as a second prediction filter means.

The ROM 18 stores the prediction filter coefficients of various taps immediately before the end, which are obtained in the process of learning the prediction filter coefficients which is performed in parallel with the conventional linear prediction processing. The ROM 18 handles the supplied data P (1) for each class as address data, and outputs the filter coefficient stored corresponding to the address to the linear prediction unit 19. The linear prediction unit 19 uses the Y / C separated prediction value S (1).
Then, the final color signal C is predicted by performing the sum-of-products calculation of

The linear predictor 19 in the adaptive Y / C separation processor 1b supplies the final color signal C separated by this prediction to the I / Q signal separation processor 2 and the adder 3, respectively. The I / Q signal separation processing unit 2 outputs I and Q signals based on the separated final color signal C from output terminals 5 and 6.
Further, the adder 3 is supplied with signal data at one end side, and subtracts and inputs the final color signal C at the other end side. The adder 3 adds these signals and outputs a luminance signal Y at an output terminal 7
Output from.

With this configuration, the luminance signal Y and the chrominance signal C or the I and Q signals can obtain a Y / C separation value having a small correlation determination error amount close to the true value. , It is possible to improve the filter performance of Y / C separation adapted to the input signal, and to suppress the error due to the correlation judgment, for example.

In this embodiment, the blocking processing unit 11 outputs R to the first prediction filter means and the fixed coefficient.
ROM 15 and R for outputting a filter coefficient that is variable according to the class code from the class classification circuit 14 recursively with a part of the linear prediction unit 13 that performs linear prediction with the filter coefficient from the OM 12 and the class classification circuit 14 as a unit
The linear prediction unit 16 that performs linear prediction with the filter coefficient from the OM 15 and the portion that forms a group of the class classification circuit 17 are connected in a cascade manner. You may comprise by the part.

Next, the operation of the adaptive Y / C separation device will be described with reference to the schematic diagrams of the tap relations of FIGS. 5, 6 and 8 and the flowchart of FIG. As shown in FIG. 5, the adaptive Y / C separation device basically has, for example, vertical tap positions (H0, V-1), (H0, V + 1), and horizontal tap positions (H-
2, VO), (H + 2, V0), (H0, V0) of field number 0 (# 0) and field number 2 (# 2) of tap positions in the time direction
(H0, V0) of (3) is compared with the Y / C separation value at the desired Y / C separation tap and the true value, and the tap with the smallest error is used. Here, the coefficient at each tap position described above is 0.5. Field number 0 (# 0) (H0, V-1)
Value X (2) and (H0, V0) value X (5), the desired separation tap value R
1 is represented by 0.5 * X (2) + 0.5 * X (5).

However, it is impossible to know the true value on the Y / C separation side. Therefore, it has to be estimated by some method instead of the true value. The estimation method for this is
For example, a method of linearly predicting the true value of the luminance signal Y from the weighted average obtained by multiplying the coefficient at each tap position and the tap value is used (see FIG. 6).

Here, the symbols ◯ and ● of each tap in FIGS. 5 and 6 indicate that the phase difference of the subcarriers is 180.
In-phase and 90-degree opposite phase are shown. Moreover, the square shown by the broken line indicates a tap unrelated to the separation tap.

As shown in FIG. 6, the estimated value (Y / C separation predicted value) for the true value of the luminance signal Y is compared with the value separated by the desired tap. In this comparison, the difference between the estimated value for the true value of the luminance signal Y and the value separated by the desired tap is estimated as the error amount. Whether the estimated error amount is suitable for tap separation is determined by threshold processing. Among the various separation taps prepared, a class is configured and learned so as to select the separation taps corresponding to the estimated error quantity. When these series of processes are recursively repeated until the error quantity converges, , The difference between the separation value by the desired tap and the estimated value with respect to the true value, that is, the error amount can be minimized.

A procedure applying this concept will be described with reference to the flowchart of FIG. Further, the above-mentioned drawings are used as necessary. Start Y / C separation by recursive classification. In step S10, the initial coefficient in the separation prediction for each tap is set. For example, the coefficient of the true value estimation tap in FIG. 6 is the field number 0 (# 0),
Field number 2 (# 2) (H-2, V-1), (H + 2, V-1), (H
-2, V + 1), (H + 2, V + 1) 0.03125, (H0, V-1), (H-2, V0),
0.0625 is used for (H + 2, V0) and (H0, V + 1), and 0.125 is used for (H0, V0) as initial values.

In step S11, signal data (input video signal) is input to each tap of the initial coefficient. This signal data is divided into one dimension into two dimensions and the step S
Proceed to 12.

In step S12, the product sum of the initial coefficient and the blocked signal data as the first prediction filter processing step is obtained. The sum of products is the Y / C separation predicted value S (0).

Here, the symbols X (1) to X (18) of the signal data are assigned to each tap as shown in FIG.

In step S13, as shown in FIG. 5, when the Y / C separation value at a desired tap, for example, 5 taps,
Data R1 corresponding to the pixel position predicted by Y / C separation
R5 R1 = 0.5 * X (2) + 0.5 * X (5) R2 = 0.5 * X (5) + 0.5 * X (8) R3 = 0.5 * X (4) + 0.5 * X (5) R4 = 0.5 * X (5) + 0.5 * X (6) R5 = 0.5 * X (5) + 0.5 * X (14) is obtained. Next, the difference between the R1 to R5 and the Y / C separation predicted value S (0) is defined as the error amount, and the error amount and the threshold value T
Compare with H (0). When it is smaller than the threshold value TH (0), the level "1" is output, and otherwise, the level "0" is output.
Is output (class classification process). The result of these five comparisons is the class code P. This class code P is output as an address.

When learning, a normal equation is generated from the signal data, the blocked signal data B and the class code P. By solving the simultaneous linear equations obtained from this normal equation, the coefficient of a new tap is obtained.

In step S14, the newly obtained coefficient is output as a filter coefficient (filter coefficient output step). In step S15, the total error amount is calculated as described above. It is determined whether or not this total error amount has decreased and converged. If it has not converged yet (No),
The learning of the filter coefficient of each tap is repeated by the procedure from step S12. For this learning, the least squares method using a normal equation is used. If the total error amount has converged, the process proceeds to step S16.

In step S16, the adaptive Y / C separation process corresponding to the second predictive filtering process is performed using the filter coefficient of this tap. This process outputs the optimally separated color signal C in the same manner as in the case where it is performed by the adaptive Y / C separation unit 1b as shown in FIG.

In step S17, the color signals C to I
The signal and the Q signal are separated. In step S18,
The optimally separated luminance signal Y is output by subtracting and inputting the optimally separated color signal C into the signal data.

Thus, the color signal C and the luminance signal Y can be optimally separated by recursively repeating until the difference between the desired tap and the true value is converged as an error amount.

The color signal C and the luminance signal Y can be optimally separated from each other by recursively repeating the class classification and the linear prediction using the above configuration, and the error due to the correlation judgment can be suppressed, for example. .

Even if the filter coefficient in the first linear predictor is set to a fixed coefficient, a variable characteristic filter whose filter characteristic changes based on the filter coefficient obtained according to the classification result of the input signal is used. Optimal filter processing can be performed. Furthermore, the first linear prediction unit causes the class classification based on the linear prediction with the fixed coefficient, and the linear prediction is performed with the filter coefficient from the variable filter corresponding to the output signal of this class classification to perform the class classification.
It is possible to accelerate the convergence of the error amount, set more optimal filter coefficients, and perform optimal Y / C separation.

Further, the color signal C and the luminance signal Y can be optimally separated by using the method described above. It is monitored according to the data obtained for each class until the total error amount obtained from the estimation by the filter coefficient at the time of learning converges to a predetermined value, and the filter coefficient is learned according to the monitoring result to classify the class. By repeating, the optimum filter coefficient can be set and the signal data can be Y / C separated.

In the present invention, the class is classified by comparing the linearly predicted data with the separated value of the desired tap, but the invention is not limited to this class classification, and the linearly predicted data and the blocked signal data may be used. Even if compared and classified, the optimum filter coefficient can be set and the signal data can be set to Y / C.
Can be separated.

[0063]

In the adaptive filter device according to the present invention, the color signal C and the luminance signal Y can be optimally separated by recursively repeating the class classification and the linear prediction, and for example, the error due to the correlation determination can be suppressed. be able to.

Even if the filter coefficient in the first linear predictor is set to a fixed coefficient, a variable characteristic filter whose filter characteristic changes based on the filter coefficient obtained according to the classification result of the input signal is used. Optimal filter processing can be performed. Furthermore, the first linear prediction unit causes the class classification based on the linear prediction with the fixed coefficient, and the linear prediction is performed with the filter coefficient from the variable filter corresponding to the output signal of this class classification to perform the class classification.
It is possible to accelerate the convergence of the error amount, set more optimal filter coefficients, and perform optimal Y / C separation.

Further, the adaptive filter processing method of the present invention can also optimally separate the color signal C and the luminance signal Y. It is monitored according to the data obtained for each class until the total error amount obtained from the estimation by the filter coefficient at the time of learning converges to a predetermined value, and the filter coefficient is learned according to the monitoring result to classify the class. By repeating, you can set the optimum filter coefficient and set the signal data to Y /
C can be separated.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of an adaptive Y / C separation device to which an adaptive filter device according to the present invention is applied.

FIG. 2 is a block diagram showing a configuration of a class classification unit of the adaptive Y / C separation device.

FIG. 3 is a block circuit diagram showing a configuration of a class classification circuit in the class classification unit.

FIG. 4 is a block diagram showing a configuration of adaptive Y / C separation processing of the adaptive Y / C separation device.

FIG. 5 is a schematic diagram illustrating a relationship between five types of separation taps in the adaptive Y / C separation device.

FIG. 6 is a schematic diagram illustrating a relationship between true value estimation taps by linear prediction of the adaptive Y / C separation device.

FIG. 7 is a flowchart illustrating an adaptive filter processing method according to the present invention.

FIG. 8 is a schematic diagram illustrating a relationship between blocked data of pixel positions used for Y / C separation.

FIG. 9 is a schematic diagram illustrating the relationship between taps for class classification and taps used for prediction.

[Explanation of symbols] 1 Y / C separation device 2 I, Q signal separation processing unit 3 adder 4 input terminals 5, 6, 7 output terminals 1a Class classification section 1b Adaptive Y / C separation processing unit 11 Block processing unit 12, 15 ROM 13, 16 Linear predictor 14, 17 class classification circuit 141a to 141e Coefficient circuit 142a-142e product sum 143a to 143e comparator

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H04N 9/78 H03H 17/02 621 H03H 21/00

Claims (5)

(57) [Claims] 1. An adaptive filter device for changing a filter characteristic according to an input signal, wherein a class classification unit for classifying an input signal into blocks and classifying it into one of a plurality of classes, and the block from the class classification unit. A filter processing unit to which the converted output signal and the classified class are input, wherein the class classification unit is based on a filter coefficient output unit that outputs a filter coefficient, and a filter coefficient based on the filter coefficient from the filter coefficient output unit. A predictive filter means having a filter characteristic and linearly predicting a predicted value of a true value for classifying an input signal, and an output signal from the predictive filter means is selected from a plurality of classes. And a plurality of stages having a class classification means for classifying into a plurality of stages. According to the class classified by, the filter coefficient of the filter coefficient output unit of the next stage is output, and the filter processing unit is classified by the class classification unit at the final stage of the class classification unit. A filter coefficient output unit that outputs a filter coefficient according to a class, and a prediction filter that has a filter characteristic based on the filter coefficient from the filter coefficient output unit and that linearly predicts a true predicted value for an input signal. An adaptive filter device comprising: 2. The input signals are a luminance signal Y and a color signal C.
Is a multiplexed signal, and the input signal is filtered to separate the luminance signal Y and the color signal C from each other. 3. The adaptive filter device according to claim 1, wherein the predictive filter means of the first-stage component of the class classification unit is a filter having a fixed filter coefficient. 4. The predictive filter means of the constituent parts of the second and subsequent stages of the class classification unit changes the filter coefficient based on the filter coefficient obtained according to the class classified by the class classification means of the former constituent part. 2. The adaptive filter device according to claim 1, wherein the adaptive filter device is a variable characteristic filter. 5. An adaptive filter processing method for changing a filter characteristic according to an input signal, wherein a class classification processing step of classifying an input signal into blocks and classifying it into any one of a plurality of classes; Of the block output signal and the classified class are input. In the class classification processing step, a filter coefficient output step of outputting a filter coefficient and a filter coefficient output step from the filter coefficient output step Under the filter characteristic based on the filter coefficient, a prediction filter step for linearly predicting a true predicted value for classifying an input signal, and an output signal from this prediction filter step for a plurality of classes. The partial treatment step having a class classification step of classifying into any of the Depending on the classified class by class classification step of the previous partial process steps of the extent,
The filter coefficient is output in the filter coefficient output step of the next partial processing step. In the filter processing step, the filter coefficient is set according to the class classified by the final class classification step of the class classification processing step. A filter coefficient output step of outputting and a prediction filter step of linearly predicting a true predicted value for an input signal, having a filter characteristic based on the filter coefficient from the filter coefficient output step Adaptive filtering method.