JPH0951510A - Device and method for signal conversion - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain an image signal of high resolution by classifying input image signal in block units in stages according to an activity code. SOLUTION: As for an SD(standard resolution) image signal S1 inputted to an up converter, the activity classification part of a classification part 12 determines a threshold according to a dynamic range DR, and sets a tap pattern corresponding to the characteristics of the dynamic range DR of the SD image signal S1 among three kinds of spatial class tap pattern (wide-area tap pattern, standard-area tap pattern, and narrow-area tap pattern) according to a class code c0 obtained as the result. Thus, the class tap pattern on which the dynamic range DR characteristics of the input SD image signal are reflected is set. According to a class code C1, the number (k) of requantized bits of each tap for spatial class classification is varied according to the class code c1 to switch the level resolution of the tap.

Description

Detailed Description of the Invention

[0001]

[Table of Contents] The present invention will be described in the following order. TECHNICAL FIELD OF THE INVENTION Conventional Technology (FIGS. 28 to 30) Problems to be Solved by the Invention (FIGS. 28 to 30) Means for Solving the Problems (FIGS. 1 to 27) Embodiments of the Invention (1 ) First Example (FIGS. 1 to 8) (2) Second Example (FIGS. 1 to 9) (3) Third Example (FIG. 4) (4) Fourth Example (FIG. 10) (5) ) Fifth Example (6) Sixth Example (FIGS. 11 to 17) (7) Seventh Example (FIG. 18) (8) Eighth Example (FIGS. 19 to 21) (9) Ninth Example Example (FIGS. 22 to 25) (10) Other Examples (FIGS. 26 to 27)

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal conversion device and a signal conversion method, for example, an NTSC system (a committee for examining American television systems-National Television).
It is suitable to be applied to an up-converter for converting a standard resolution signal (SD: Standard Definition) such as a system (determined by the System Committee) into a high resolution signal (HD: High Definition) such as a high vision. .

[0003]

2. Description of the Related Art Conventionally, in this type of up-converter, an HD image signal is formed by increasing the number of pixels by performing frequency interpolation processing on the SD image signal. For example, as shown in FIG. 28, such an up-converter is arranged on the scanning line 1 of the HD image in the horizontal and vertical directions with respect to the SD image signal having a large “◯” mark and a large “Δ” mark. By performing double frequency interpolation on each, an HD image signal having a small “◯” mark and a small “Δ” mark is generated.

In this case, as an example of interpolation by the up converter, there is a method of generating HD pixels at four kinds of positions from the field data of the SD image signal. For example, paying attention to the SD pixels marked with "◎" in the figure, there are four types in the vicinity.
The HD pixels at the positions of mode1, mode2, mode3 and mode4 are generated by interpolating the surrounding SD pixels. As the interpolation filter used at this time, there are a two-dimensional non-separable filter 2 in space shown in FIG. 29 and a horizontal / vertical separable filter 3 shown in FIG.

Practically two-dimensional non-separable filter 2
Is HD pixel mode1, mode2, mode3 and mo at four positions
Interpolation processing is independently performed for each de4 by the two-dimensional filters 4A to 4D, and each interpolation result is serialized in the selection unit 5 to generate an HD image signal. On the other hand, the horizontal / vertical separable filter 3 uses the vertical interpolation filter 6A for mode1 and mode3.
Is executed, and the vertical interpolation filter 6B is used to
And the processing for mode 4 is executed to form two scanning line data of the HD image signal. Next, the horizontal filters 7A and 7B are used for each scanning line to interpolate the HD pixels at four types of positions, and the interpolated results are serialized in the selection unit 8 to generate an HD image signal. Has been done.

[0006]

In the conventional up converter as described above, even when the ideal filter is used as the interpolation filter, the spatial resolution is the same as that of the SD image signal although the number of pixels increases. In addition, in reality, since an ideal filter cannot be used, S
There is a problem that only an HD image signal whose resolution is lower than that of the D image signal can be generated.

As one method for solving such a problem,
For example, as disclosed in Japanese Patent Laid-Open No. 5-328185, the SD image signal is classified into several classes based on the characteristics of the input SD image signal, and the prediction data for each class is generated by learning in advance. A class classification adaptive processing method for generating a high-resolution HD image signal using a prediction coefficient has been proposed.

However, according to this method, when an HD image signal is generated using the class classification adaptive processing method, when a prediction coefficient is generated by learning, an appropriate class classification is performed according to the characteristics of the input SD image signal. If not performed, there is a problem that the prediction accuracy of the HD image signal is reduced. That is, if the class classification capability is not sufficient, HD image signals that should originally be classified into different classes are classified into the same class. Therefore, the prediction coefficient obtained by learning predicts the average value of HD image signals having different properties, and as a result, there is a problem that the resolution restoration capability is reduced.

The present invention has been made in consideration of the above points, and a low resolution image signal having a higher resolution and a higher resolution can be obtained by appropriate class classification corresponding to various signal characteristics of an input image signal. An object of the present invention is to propose a signal conversion device and a signal conversion method that can convert an image signal.

[0010]

In order to solve such a problem, in the first invention, a signal converter for converting an input first image signal into a second image signal different from the first image signal. In the first method, the activity of the first image signal in the space is evaluated, and the activity code is output.
Of the second class classification means for executing stepwise class classification based on the activity code and outputting the class code based on the result of the class classification, and the second class classification means using the second image signal. A prediction coefficient storage unit that stores a prediction coefficient for predicting and generating an image signal;
By performing a prediction calculation on the first input image signal using the prediction coefficient read from the prediction coefficient memory according to the activity code and / or the class code,
A prediction calculation means for generating the second image signal is provided.

In a second aspect of the present invention, in the signal conversion device, the first class classification means for evaluating the in-spatial activity of the first image signal and outputting the activity code, and stepwise on the basis of the activity code. A second class classification unit that performs class classification and outputs a class code based on the classification result, and a predicted value generated as an interpolated pixel signal of the first image signal are stored, and the activity code and And / or a predicted value storing / reading unit that reads and outputs a corresponding predicted value according to the class code.

Further, in the third invention, in the signal conversion method for converting the input first image signal into the second image signal different from the first image signal, the spatial activity of the first image signal is obtained. To output an activity code, perform stepwise class classification based on the activity code, output a class code based on the classification result, and output the first code according to the activity code and / or the class code. The prediction coefficient stored in the prediction coefficient storage means for predicting and generating the second image signal using the image signal is read, and the read prediction coefficient is used for the first input image signal. Then, the second image signal is generated by performing the prediction calculation.

Further, in the fourth invention, in the signal conversion method for converting the input first image signal into a second image signal different from the first image signal, the spatial activity of the first image signal is obtained. To output an activity code, perform stepwise class classification based on the activity code, output a class code based on the class classification result, and predict the predicted value according to the activity code and / or the class code. The predicted value generated as the interpolated pixel signal of the first image signal stored in the storage means is read and output.

As a result, in the first to fourth aspects of the invention, the in-spatial activity of the input image signal for each block unit is evaluated and classified, and the input image signal in block units is obtained according to the activity code obtained as a result. By performing the general class classification, the activity code of the first step and the result of the class classification of the previous step are reflected in the class classification of the subsequent steps, and thus highly accurate class classification can be performed.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings.

(1) First Embodiment FIG. 1 shows SD by applying a class classification adaptive process as a whole.
2 shows an up-converter using a two-dimensional non-separable filter for generating an HD image signal from an image signal. The SD image signal S1 input to the up converter 10 through the input terminal IN is sent to the class classification unit 12 for each block unit consisting of a predetermined number of pixels centering on the SD pixel of interest, and the SD image signal S1 is further predicted. It is sent to the calculation unit 13. The class classification unit 12 uses the input SD image signal S1.
Based on the SD pixel NO feature of the peripheral SD image signal S1 for the target pixel of
The class code d0 is generated as a prediction coefficient ROM (Read Only Memory) which is a storage means as address data.
Send to 14.

In this case, the prediction coefficient ROM 14 stores, in the class code, the prediction coefficient obtained by pre-learning, which is used when predictively calculating HD interpolation pixels for generating a high resolution image signal from a low resolution image signal. It is stored as predicted data d1 associated with d0. Thus, the prediction coefficient ROM 14 reads the prediction data d1 using the class code d0 as address data, and uses this as the prediction calculation unit 1
Send to 3.

On the other hand, the prediction calculation unit 13 is arranged to detect the SD image signal S1.
By executing a predetermined prediction calculation using the prediction data d1 with respect to, an HD interpolation pixel is generated from the SD image signal S1. The SD image signal S1 supplied to the prediction calculation unit 13 is supplied via a delay unit (not shown). The delay time of this delay unit corresponds to the time until the prediction data d1 is supplied to the prediction calculation unit 13.

The prediction calculation unit 13 is composed of four prediction calculation units 13A to 13D, and each prediction calculation unit 13A.
13D to 13D execute the sum of products operation using the prediction data d1 for the SD image signal S1. As a result, each of the prediction calculation units 13A to 13D has four types of positions on the scanning line 1, respectively.
The predicted values d2, d3, d4 and d5 of the HD interpolation pixels corresponding to mode1, mode2, mode3 and mode4 are generated. The HD interpolation pixels d2, d3, d4, and d5 generated by the prediction calculation units 13A to 13D are sent to the selection unit 15. The selection unit 15 rearranges the predicted values d2, d3, d4 and d5 in time series using a buffer memory (not shown) and outputs the rearranged output values OUT as the HD image signal S2.

FIG. 2 shows the configuration of the class classification unit 12 shown in FIG. As is clear from FIG. 2, the input terminal IN
The SD image signal S1 input through the above is first supplied to the activity classification unit 21. The activity classification unit 21 responds to the supplied SP image signal S1 by, for example,
The spatial activity is classified for each block unit consisting of 3 × 3 9 pixels centered on the pixel of interest, and the characteristic thereof is evaluated and judged. In addition, the activity classification unit 21 determines the class code based on the classification and evaluation of the spatial activity.
c0 is generated, and the selection unit 25 and ADRC (Adaptive Dyn
amic Range Coding) to the class classification unit 26.

On the other hand, the SD image signal S1 is sent in parallel to the wide area tap selecting section 22, the standard tap selecting section 23 and the narrow area tap selecting section 24 which set the tap patterns of three different kinds of pixels. Then, each of the wide area tap selection section 22, the standard tap selection section 23 and the narrow area tap selection section 24 selects the tap patterns p0, p1 and p2 of the spatial class for the input SD image signal S1.

FIG. 3 shows the activity classification unit 21 of FIG.
Is shown. As shown in FIG. 3, the input terminal IN
The SD image signal S1 input from the first is sent to the processing unit 30. The processing unit 30 receives the input SD image signal S
The dynamic range DR of a plurality of SD pixels centering on the target SD pixel 1 is detected. Dynamic range DR
Is, for example, the maximum value MAX in the neighboring area obtained from 9 pixels (indicated by ⊚ in the figure) centered on the pixel of interest shown in FIG.
And the minimum value MIN in the neighborhood area,

[Equation 1] Is defined by The dynamic range DR of a plurality of SD pixels centering on the SD pixel of interest is sent to the threshold value determination unit 31.

The threshold value judging section 31 uses a predetermined threshold value,
A comparison is made with the dynamic range DR. As a result, the threshold determination unit 31 outputs the class code c0 obtained by the threshold processing. That is, the spatial activity is determined by the threshold processing of the threshold determination unit 31 by determining the magnitude of the dynamic range in three stages (that is, the spatial activity has three stages of high, medium, and small). ,
The determination result is output as a 2-bit class code c0.

In general, it is considered that the spatial activity is high when the dynamic range DR is large, and the spatial activity is low when the dynamic range DR is small. As a result, the first-stage class classification by the dynamic range is performed in the activity classifying unit 21.

Next, the class classification of the next stage in the wide area tap selection section 22, the standard tap selection section 23, the narrow area tap selection section 24 and the ADRC class classification section 26 will be specifically described.

First, of the three types of cluster tap pattern selection units described above, the standard tap selection unit 23 inputs the input S
Paying attention to the standard spatial variation of the D image signal S1, FIG.
A general cluster tap pattern as shown in (B) is selected. On the other hand, the wide area tap selection unit 22 pays attention to the stationary space of the input SD image signal S1.

That is, the wide area tap selection unit 22 is shown in FIG.
A cluster tap pattern for a wide area as shown in (A) is selected. Further, the narrow area tap selection unit 24 pays attention to the non-stationary spatial fluctuation of the input SD image signal S1, and responds to the non-stationary signal change by a cluster table for the narrow area as shown in FIG. 5C. Select a pattern.

The wide area tap selecting section 22, the standard tap selecting section 23 and the narrow area tap selecting section 24 send the respectively selected cluster tap patterns p0, p1 and p2 to the selecting section 25. The selecting section 25 selects one of the cluster tap patterns p0, p1 and p2 using the class code c0 sent from the activity classification section 21 as a selection control signal, and selects the selected cluster tap pattern p3ADRC. It is sent to the class classification unit 26.

That is, the selection unit 25 uses the class code c0.
, If the spatial activity is low, the cluster tap pattern p0 from the wide area tap selection unit 22 is selected, and if the class code c0 indicates that the spatial activity is high, the cluster area from the narrow area tap selection unit 24 is selected. Select the up pattern p2.

The ADRC class classification unit 26 sets the number k of requantization bits of ADRC using the class code c0 obtained according to the dynamic range DR of the input SD image signal S1 as a control signal. As a result, the level resolution of each tap of the tap pattern p3 of the space class can be changed and set according to the dynamic range DR of the SD image signal S1.

ADRC requantizes pixels with a quantization step width defined as requantization. The ADRC code c1 (in the formula, ci is used according to the number i of SD pixels of the cluster pattern) is the dynamic range DR, the number k of requantization bits, the SD pixel xi, and the minimum pixel level in the vicinity thereof. With MIN,
Next formula

[Equation 2] Is represented by

The level resolution of the spatial cluster tap pattern is specifically switched by switching the requantization bit number k in the ADRC operation of the equation (2) according to the class code c0. Thus, the level resolution can be adaptively switched and set according to the dynamic range DR of the input signal. That is, the ADRC class classification unit 26 is configured to set the level resolution more finely as the dynamic range DR is larger. In this way, the class classification unit 12 generates the class code d0 including the class code c0 and the ADRC code c1. This class code d0 is sent to the prediction coefficient ROM 14 in the subsequent stage as address data.

The prediction coefficient ROM 14 stores prediction data d1 and class codes c0 and c1 used when generating HD interpolation pixels.
The class code d0, which is a combination of and, is read as address data and sent to the prediction calculation unit 13. Prediction calculation unit 1
Each of 3A to 13D executes a prediction calculation using the SD pixel xi which is the SD image signal S1 and the prediction coefficient wi which is the prediction data d1 for each class, and the positions mode1 to mo on the scanning line 1 are calculated.
An estimated pixel y ′ of the HD interpolation pixel corresponding to de4 is generated.

The SD pixel xi used at this time is, for example,
Pixels of interest arranged as shown in FIG. 6 (indicated by ◎ in the figure)
And 13 pieces of prediction tap data composed of peripheral pixels (indicated by circles in the figure). Therefore, in this case, the prediction coefficient wi is composed of 13 prediction coefficients for each of the prediction calculation units 13A to 13D. The SD pixels used in each of the prediction calculation units 13A to 13D are the same, but the prediction coefficient ROM
The prediction coefficient wi from 14 is the prediction calculation units 13A to 13D.
Since the prediction coefficient is different for each of the
The OM 14 stores four prediction coefficient groups each including 13 prediction coefficients Wi corresponding to one class.

The estimated pixel y'of the HD interpolation pixel has been described above.
Formula using 13 SD pixels x i and prediction coefficient w i

(Equation 3) Is converted and generated. That is, the prediction calculation unit 1
3A to 13D are the SD pixels X respectively supplied.
Using i and the prediction coefficient Wi, the prediction calculation is performed by the equation (3) to generate the HD interpolation pixel. The prediction coefficient wi used here is obtained by learning in advance, and the prediction coefficient RO
It is stored in M14.

Here, the learning procedure for generating the prediction coefficient Wi for each class stored in the prediction coefficient ROM 14 will be described with reference to the flowchart of FIG. The prediction coefficient Wi can be obtained in accordance with the prediction coefficient learning procedure RT1 shown in FIG. 7. First, when the prediction coefficient learning procedure RT1 is started in step SP1, the following step SP2
In order to learn the prediction coefficient w i, learning data corresponding to an already known image is generated.

Specifically, in the HD image shown in FIG. 28, the HD interpolation pixel is set as the HD attention pixel, and this HD attention pixel is predicted by a set of learning data composed of HD interpolation pixels and SD pixels in the periphery. It is represented by a linear linear combination model using. The prediction coefficient used at this time is obtained for each class by using the least squares method. In addition, when generating a large number of learning data by using a plurality of images instead of using only one image when generating the learning data, a more accurate prediction coefficient can be obtained.

At the subsequent step SP3, step SP2
It is determined whether or not the number of learning data generated in (3) has been generated enough to obtain the prediction coefficient. If it is determined that the number of learning data is less than the required number, step SP
Go to 4. In this step SP4, the class learning data is classified into classes. In the classification, first, the local flatness of the learning sampling data is detected, and the pixel used for the classification is selected according to the detection result. As a result, it is possible to eliminate the influence of noise by excluding the input signal having a small change from the learning target. After that, the classification of the class learning data is performed by the input SD image signal S
This is done by performing the same processing as in classifying 1.

That is, in class classification of the learning data, first, the dynamic range DR of the learning data is classified and evaluated, and the class code c0 is set. Then, wide area,
One of the tap patterns p3 is selected as a spatial class from the three types of standard and narrow area tap patterns based on the class code c0. As shown in FIG. 8, the class code c0 and the ADRC code c1 thus obtained are obtained.
A class code d0 is set by combining and, and the class code d0 is stored in the ROM 14 in association with the prediction data d1.

At step SP5, a normalization equation is formed for each class based on the classified learning data. The processing in step SP5 will be specifically described. Here, for the sake of generalization, a case will be described in which there are n sampling pixels as learning data. First, the pixel level x1, ..., Xn of each sampling pixel
And the pixel level y before the sub-sampling of the interpolation pixel of interest is represented by a prediction formula by an n-tap linear linear combination model with prediction coefficients w1, ..., Wn for each class. This prediction formula is

(Equation 4) Shown in Prediction coefficient w1, ..., W in equation (4)
The pixel level y is estimated by finding n.

Next, an example of generating the prediction coefficients w1, ..., Wn by the method of least squares will be shown. The least squares method is applied as follows. As a generalized example, consider the following observation equation, where X is input data, W is a prediction coefficient, and Y is an estimated value.

(Equation 5) The least squares method is applied to the data collected by the observation equation (5). In the example of equation (5),
n = 13, m is the number of learning data.

First, the following residual equation will be considered based on the observation equation (5).

(Equation 6) From the residual equation of equation (6), the most probable value of each wi is

(Equation 7) It is considered that the condition for minimizing is satisfied. That is, the partial derivative by wi of equation (7) is

(Equation 8) In this case, n conditions based on i in the equation (8) are considered, and w1, w2, ..., Wn satisfying these conditions may be calculated. Therefore, the following equation is obtained from the residual equation (6).

[Equation 9] From the expressions (9) and (8), the following expression

(Equation 10) Is obtained. Then, the following normal equation is obtained from the equations (6) and (10).

[Equation 11] With the normal equation of the equation (11), since the same number of equations as the number n of unknowns can be established, the most probable value of each wi can be obtained. This normal equation can be solved using the sweep method (Gauss-Jordan elimination method).

In the prediction coefficient learning procedure RT1, in order to obtain the undetermined coefficients w1, ..., Wn for each class, the step S is repeated until the normalization equations of the same number as the unknown number n are formed.
The loop of P2-SP3-SP4-SP5-SP2 is repeated.

When the required number of normalization equations are obtained in this way, an affirmative result is obtained in step SP3 for the determination as to whether or not the learning data has ended, and the processing is performed in step SP6 for determining the prediction coefficient. Move.

In this step SP6, the normalization equation of the equation (11) is solved to predict the prediction coefficient w1 for each class.
Determine wn. The prediction coefficient thus obtained is subjected to address division RO for each class in the next step SP7.
It is registered in the storage means such as M. At this time, for one class code, four prediction coefficient groups including prediction coefficients w1, ..., Wn respectively corresponding to the prediction calculation units 13A to 13D are registered. By the above learning processing, the prediction coefficient of the class classification adaptive processing is generated, and the prediction coefficient learning procedure RT1 is ended at the next step SP8.

The operation of the above-described up-converter 10 and each part of the up-converter having the above-mentioned structure will be described. First, the SD image signal S1 input to the up converter 10 through the input terminal IN is sent to the class classification unit 12 and the prediction coefficient calculation unit 13 in parallel. The class classification unit 12 generates a class code d0 based on the SD image signal S1 and sends it to the prediction coefficient ROM 14.

The prediction coefficient ROM 14 reads the prediction data d1 previously obtained by learning according to the class code d0, and sends it to the prediction coefficient calculation unit 13. In each of the prediction calculation units 13A to 13D, the prediction coefficient calculation unit 13 uses the SD image signal S1 sent from the input terminal IN and the prediction data d1 sent from the prediction coefficient ROM 14 to determine the 4 on the scanning line 1. H corresponding to one position (mode1 to mode4)
Generate D interpolation pixels.

The activity classifying unit 21 first detects the dynamic range DR of a plurality of SD pixels centering on the SD pixel of interest of the input SD image signal S1, and sets this dynamic range DR as a predetermined threshold value. The class code c0 is output by comparing and determining between the two. In general,
When the dynamic range DR is large, the spatial activity is high. On the contrary, when the dynamic range DR is small, the spatial activity is low.

On the other hand, the input SD image signal S1 in block units is arranged in parallel to the wide area tap selecting section 22, the standard tap selecting section 23 and the narrow area tap selecting section 24 which set the tap patterns of three different types of pixels. Sent out. Then, the wide area tap selection unit 22, the standard tap selection unit 23, and the narrow area tap selection unit 24 set tap patterns p0, p1, and p2 of the space class, respectively.

Based on the class code c0, the selecting section 25 selects, for example, an SD image signal S1 having a small dynamic range DR and a low activity as shown in FIG.
A cluster tap pattern p0 of a signal change in a relatively wide area as shown in (3) is selected, and a gradual signal change is reflected in the class. On the other hand, for the SD image signal S1 having a large dynamic range DR and a high activity,
By selecting a cluster tap pattern p2 in a narrow area as shown in FIG. 5C, the signal change in the narrow area is expressed by the maximum number of classes. As a result, the selection unit 25 selects a space class consisting of the tap pattern p3 reflecting the signal change in accordance with the signal characteristic viewed from the dynamic range DR, and sends it to the ADRC class classification unit 26 in the next stage.

The ADRC classifying unit 26 sets the requantization bit number k of each tap for spatial class classification small for the SD image signal S1 having a small dynamic range DR by using the class code c0 as a control signal. . As a result, the level resolution of each tap is lowered, and the ADRC code c1 can be output on the assumption of stationarity. On the other hand, the SD image signal S having a large dynamic range DR
For 1, the number of requantization bits k of each tap for spatial class classification is set to a large value and the ADRC code c1 with high level resolution is output. As a result, the unsteady signal change of the SD image signal S1 having a large dynamic range DR and a high activity can be reflected in the class.

In this way, the class classification unit 12 switches the tap patterns of the pixels used for class classification according to the dynamic range DR of the input SD image signal S1 and also requantizes the taps for each class classification. The level resolution is adaptively set by switching several k. As a result, it is possible to perform an appropriate class classification according to the characteristics of the dynamic range DR of the input SD image signal S1.

The class classification unit 12 combines the class code c0 and the ADRC code c1 with the class code d0.
To the prediction coefficient ROM 14 of the next stage. Prediction coefficient RO
M14 is the prediction data d1 based on this class code d0.
Is read out and sent to the prediction calculation unit 13. The prediction calculation unit 13 converts the SD pixel using this prediction data d1 to generate an HD interpolation pixel. Then, the generated HD interpolation pixels are supplied to the selection unit, rearranged in time series by the selection unit 15, and output as an HD image signal. Thus, the selected prediction data d1 reflects the characteristics of the input SD image signal S1 due to the dynamic range DR, and the precision of the HD interpolation pixel generated by conversion from the SD pixel can be improved. Image signal S
A spatial resolution improvement of 2 can be achieved.

According to the above configuration, the up converter 1
The SD image signal S1 input to 0 is threshold-value-determined by the activity classification unit 21 according to the dynamic range DR, and three types of spatial cluster tap patterns (wide area tap patterns, It is possible to set a tap pattern according to the characteristic of the dynamic range DR of the SD image signal S1 from the standard area tap pattern or the narrow area tap pattern), and thus, a cluster pattern reflecting the dynamic range DR characteristic of the input SD image signal S1. Pattern can be set.

Further, according to the above configuration, the class code c1
By changing the requantization bit number k of each tap for spatial class classification according to the above, and switching the level resolution of each tap, the stationarity or non-stationarity of the signal change is classified by the level resolution of the tap. Can be reflected in the classification. Thus, an HD image signal having a high spatial resolution reflecting the signal characteristics of the input SD image signal S1 is obtained by appropriately classifying the tap pattern and the level resolution of the tap according to the dynamic range DR characteristics of the input SD image signal S1. S2 can be generated.

(2) Second Embodiment FIG. 9 shows an activity classification unit 35 of the up converter according to the second embodiment. The activity classification unit 35 inputs the SD image signal S1 for each block unit composed of a plurality of SD pixels centering on the SD pixel of interest.
The activity in the space is evaluated, and the class of the pixel of interest is classified according to the characteristic. That is, the SD signal S1 input from the input terminal IN is supplied to the ADRC class classification unit 36, and the ADRC class classification unit 36 performs ADRC on the SD signal S1 composed of a plurality of SD pixels centering on the SD pixel of interest. Execute the class classification process by.

The ADRC code c output from the ADRC class classification unit 36 (S
In the same manner as in the first embodiment, the dynamic range DR, the number of requantization bits k, the SD pixel xi and the minimum pixel level MIN in the vicinity thereof are used to calculate formula

(Equation 12) Is generated by.

The ADRC code c generated by the ADRC classifying unit 36 is sent to the post-processing unit 37 in the next stage. The post-processing unit 37 calculates the standard deviation σ for the ADRC code c, for example, which represents the variation degree of the level distribution pattern according to the ADRC code c. And the post-processing unit 3
7 sends the calculated standard deviation σ with respect to the ADRC code c to the threshold determination unit 38 in the next stage. The threshold determination unit 38
The class code c0 is generated and output by the threshold judgment of the standard deviation σ with respect to the ADRC code c. The standard deviation σ for the ADRC code c is the ADRC code ci, A
Average value ca of DRC codes ci and number of ADRC codes n
Using

(Equation 13) Is represented by

The selecting section 25 shown in FIG. 2 uses the class code c0 thus obtained to select the tap pattern p3 of the spatial class in the same manner as in the first embodiment. Further, the ADRC class classification unit 26 adaptively sets the level resolution of each tap based on the class code c0. As a result, the spatial cluster pattern and the level resolution are set based on the classification result of the spatial activity of the input SD image signal S1, and the same effect as the first embodiment can be obtained.

(3) Third Embodiment As a third embodiment, when classifying, for example, FIG.
The standard deviation σ is calculated from the input data distribution of 9 pixels in the vicinity of the SD pixel of interest, and the calculated standard deviation σ
The class code c0 data may be generated by determining the threshold value of. That is, in the processing unit 35 of the activity classification unit 21 shown in FIG. 3, the standard deviation σ is calculated from the data distribution of 9 pixels near the input target pixel.

That is, the processing unit 35 is supplied with the SD signal S1 input from the input terminal IN, and the processing unit 35 calculates the standard deviation σ from the data distribution of 9 pixels near the input target pixel. Then, the processing unit 30 sends the calculated standard deviation σ to the threshold value determination unit 31 in the next stage. The threshold judgment unit 38 judges the threshold of the standard deviation σ to obtain the class code.
Generate and output c0. The standard deviation σ is SD pixel xi,
Using the average value xi in the neighborhood area and the number n of pixels in the neighborhood area,

[Equation 14] Is defined by

As described above, in the third embodiment, the class classification is executed by the threshold judgment using the standard deviation σ.
Generally, when the standard deviation is large, the spatial activity is high, and when the standard deviation is small, the spatial activity is low. Therefore, by switching the spatial cluster tap pattern and switching the level resolution of the tap for spatial class classification based on the threshold determination of the standard deviation σ, the same effect as that of the first embodiment described above can be obtained. Obtainable.

(4) Fourth Embodiment Further, as a fourth embodiment, a frequency distribution table in which the ADRC code c representing the variation degree of the level distribution pattern by the ADRC code c is registered is generated, and the generated frequency distribution table is generated. The class code c0 data may be generated by using the threshold value determination. That is, in the frequency distribution table as shown in FIG. 10, the ADRC code c existing between the threshold value 0 and the threshold value 1 is measured, and the classification is performed according to the comparison. In this case, for example, the spatial activity is low when the ADRC code is bound in a predetermined section, while the spatial activity is low when the ADRC code c is widely distributed.
It can be said that space activity is high.

That is, similarly to the second embodiment, the SD signal S input from the input terminal IN is input to the ADRC class classification unit 36.
1 is supplied to the ADRC classifying unit 36,
A class classification process by ADRC is executed on a plurality of SD pixels centering on a pixel.

The ADRC code c generated in the ADRC class classification unit 36 is sent to the post-processing unit 37 in the next stage. The post-processing unit 37 generates a frequency distribution table in which the ADRC code c is registered as shown in FIG. 10, which represents the variation degree of the level distribution pattern according to the ADRC code c. Then, the post-processing unit 37 sends the data representing the frequency distribution table for the generated ADRC code c to the threshold determination unit 38 in the next stage. The threshold determination unit 38 uses the ADRC code c
In the frequency distribution table for, the threshold determination is executed by measuring the number of pixels existing between the threshold 0 and the threshold 1, and the class code c0 is generated and output.

Therefore, based on the threshold value judgment using the frequency distribution table for the ADRC code c, the spatial cluster tap pattern is switched and the level resolution of the tap for spatial class classification is switched, as described above. The same effect as the first embodiment can be obtained.

(5) Fifth Embodiment Further, as a fifth embodiment, the absolute value of the difference between adjacent pixels of SD pixel values is registered in the frequency distribution table, and the spatial activity is evaluated using the frequency distribution table. Thus, the class code c0 may be generated. In this case, the spatial activity is high when there are many pixels having large difference absolute values, and conversely, the spatial activity is low when there are many pixels having small difference absolute values.

That is, instead of the ADRC class classification unit 36 of the activity classification unit 35 shown in FIG. 9, an adjacent pixel difference calculation unit for calculating the absolute value of the difference between adjacent pixels is provided. That is, the adjacent pixel difference calculation unit is supplied with the SD signal S1 input from the input terminal IN, and the adjacent pixel difference calculation unit calculates the difference between adjacent pixels with respect to a plurality of peripheral SD pixels centering on the SD pixel of interest. Is calculated, and the absolute value of the calculated difference is generated. The absolute difference value generated by the adjacent pixel difference calculation unit is sent to the post-processing unit 37 in the next stage.

The post-processing section 37 generates a frequency distribution table in which the absolute difference values are registered, which represents the variation degree of the level distribution pattern according to the ADRC code c. And the post-processing unit 37
Sends the data representing the frequency distribution table for the generated absolute difference values to the threshold determination unit 38 in the next stage. Threshold determination unit 3
8 is the threshold value 0 in the frequency distribution table for the absolute difference value.
The threshold value determination is executed by measuring the number of pixels existing between the threshold value 1 and the threshold value 1, and the class code c0 is generated and output. Therefore, by switching the spatial cluster tap pattern and switching the level resolution of the tap for spatial class classification based on the threshold determination of the absolute difference value of the adjacent pixels, the same as in the above-described first embodiment. The effect of can be obtained.

(6) Sixth Embodiment FIG. 11 shows an up converter 50 according to the sixth embodiment. The up converter 50 performs one-dimensional Laplacian processing in a plurality of directions in the Laplacian filters 51A to 51E of the first-stage class classification unit 50A,
By comprehensively judging those values, the first-stage class classification is executed. Then, according to the classification result of the first-class classification unit 50A, the cluster pattern of the second-class classification unit 50B is set. Then, the class classification unit 50B at the next stage executes class classification using this cluster tap pattern.

The sixth embodiment evaluates the spatial and temporal activities in the first-stage class classification. The configuration of the sixth embodiment will be described with reference to FIG. The SD image signal S1 input from the input terminal IN is supplied to the class classification unit 50A. And the class classification unit 5
The SD image signal S1 supplied to 0A is supplied to the Laplacian filters 51A to 51E and the delay unit 56B.
The five types of Laplacian filters 51A to 51E perform Laplacian processing on the input SD image signal S1 in block units for each frame or each field of the supplied SD image signal S1 in Laplacian processing in different directions for each filter. Outputs values LO to L4.

That is, the Laplacian processing is performed by the one-dimensional Laplacian filters 51A to 51A as shown in FIG.
According to E, the horizontal direction (FIG. 12 (A)), the vertical direction (FIG. 12 (B)), the lower right diagonal direction (the diagonal direction from the upper left end to the lower right end) (FIG. 12 (C)), Diagonal lower left direction (diagonal direction from the upper right end to the lower left end)
(D)) and the Laplacian process in each of the time direction (FIG. 12E).

First, the Laplacian filters 51A to 51
The Laplacian values L0 to L3 obtained as a result of the Laplacian processing through D are the absolute value circuits 52A of the next stage.
To 52D respectively. Absolute value circuit 52A-5
2D applies absolute values to the Laplacian values L0 to L3, respectively. And the absolute value a0-a3 obtained as a result
Is sent to the maximum value detector 53.

The maximum value detector 53 detects the maximum value from the absolute values a0 to a3, and thresholds the detected maximum value using the threshold value TH0. As a result, the input SD image signal S
A flatness of 1 is detected, and a value (flatness value) L10 representing the flatness expressed in 1 bit is output. Further, the maximum value detector 53 outputs a value (maximum value detection direction value) a10 indicating the direction in which the maximum value expressed by 2 bits is detected. At the same time, the Laplacian filter 51E sends the Laplacian value L4 in the time direction to the absolute value circuit 52E.

The absolute value circuit 52E converts the Laplacian value L4 into an absolute value, and obtains the absolute value obtained as a result.
a4 is sent to the comparator 54. The comparator 54 threshold-processes the absolute value a4 using the threshold TH1 to output a value (time-direction change value) a11 that represents a change in the time direction expressed by one bit. Then, 4-bit data obtained by adding the above-mentioned maximum value detection direction value a10 and flatness value L10 to this time direction change value a11 is sent to the class classification unit 50B of the next stage as a control signal CT.

Since the Laplacian value is basically the sum of the spatial differences between the pixel of interest and the pixels on both sides, the Laplacian value increases as the change in adjacent pixels increases. The first-stage class classification unit 50A applies a one-dimensional Laplacian filter to the activity in a predetermined direction in the space in order to detect the direction in which an edge in the space is located.
The characteristics are roughly classified.

The control signal CT output from the class classification unit 50A is the selectors 55A to 55I of the class classification unit 50B.
And to the delay circuit 56A. Selectors 55A-5
5I is connected to the register array 57 through a line to which SD image data selected by the control signal CT is supplied. The register array 57 includes a delay circuit 56B.
The SD image signal S1 for several lines delayed through is transmitted.

The selectors 55A to 55I are selectively switched according to the control signal CT, select the SD image data supplied from the register array 57 according to the index, and the pixel data for 9 pixels is set to 1 in the next stage. It is sent to the bit ADRC class classification unit 58. That is, the cluster tap pattern of the next class classification unit 50B is selected according to the control signal CT.

The ADRC class classification unit 58 uses the selector 5
Using a cluster tap pattern formed by 9 taps selected by 5A to 55I, 1 bit A
Performs DRC class classification and expresses in 9 bits
RC code c is output. Therefore, the classification unit 5
The 0B class is classified into 512 (29) classes.

As a result, in the up converter 50,
The first class classification unit 50A (16 ways) and the next class classification unit 50B (512 ways) can be classified into 8192 classes. Thus, depending on the class selected by the spatial activity of the first stage, the ADR of the next stage
By selecting the optimum tap pattern in the C class classification, it is possible to perform a more accurate class classification that reflects the characteristics of the spatial activity in the class classification in the subsequent stages.

The prediction value calculation unit 50C at the next stage has a prediction coefficient RAM 59 for storing prediction coefficients of HD interpolation pixels, a prediction tap pattern setting unit 60 for setting a prediction tap pattern for generating HD interpolation pixels, a prediction tap pattern and prediction. The prediction calculation unit 61 generates HD interpolation pixels by calculation using coefficients. The prediction coefficient RAM 59 is 1
ADR supplied from bit ADRC classifier 58
The prediction coefficient is read using the two signals of the C code c and the control signal CT delayed through the delay circuit 56A as address data, and the read prediction coefficient is used as the prediction calculation unit 61.
Send to

On the other hand, the SD image signal S1 sent from the register array 57 is sent to the predictive tap pattern setting section 60. The predictive tap pattern setting unit sets a predictive tap pattern used in the predictive calculation unit 61 in the subsequent stage. That is, from the input SD image signal S1, the prediction tap pattern that is predictively calculated with the coefficient read from the prediction coefficient RAM by the prediction calculation unit 61 in the subsequent stage is output. The prediction calculation unit 61 generates and outputs an HD interpolation pixel by linear linear combination using the pixel data of the prediction tap pattern and the prediction coefficient.

Here, the 4-bit control signal CT generated by the class classification unit 50A and the cluster tap pattern used by the class classification unit 50B selected by the control signal CT will be described.

FIG. 13 shows indexes (I0 to I3) representing 16 (24) combinations of 4-bit control signals CT generated by the classifying unit 50A. 14 to 17 show an example of the tap pattern structure of the class classification unit 50B in the next stage provided corresponding to the indexes (I0 to I3). Regarding the relationship between the control signal CT and the tap pattern, there are cluster taps in the direction of large change (direction of large Laplacian value) in the spatial direction,
The relationship is such that the larger the maximum Laplacian value, the smaller the spread of the tap space. Furthermore, when the Laplacian value in the time direction is large, the tap pattern is composed of taps in the same field, and when the Laplacian value in the time direction is small, the tap pattern is composed of taps on different fields (for example, frames).

The tap patterns shown in FIGS. 14A to 14D are provided corresponding to the indexes (0000) to (0011), respectively, and are selected for those having a large change in the horizontal direction. It

When comparing the cluster patterns shown in FIGS. 14A to 14D,
Indexes (0010) and (001) shown in (C) and (D)
The horizontal arrangement intervals of the cluster tap pattern of 1) are shown in FIGS. 14A and 14B as index (0000) and
The maximum Laplacian value in the horizontal direction is represented by making it narrower than the arrangement interval of the cluster pattern of (0001). That is, the larger the maximum horizontal Laplacian value, the smaller the spatial spread of the cluster tap pattern, and the smaller the maximum horizontal Laplacian value, the larger the spatial spread of the cluster tap pattern. To do.

Further, as shown in FIGS. 14B and 14D, when the bits in the time direction (fourth bit) of the indexes (0001) and (0011) are on, the cluster tip pattern is set to the same field. By arranging it inside, the change in the time direction is greatly expressed. On the other hand, FIG. 14 (A)
And (C), the indexes (0000) and (0
When the bit in the time direction (010) (the fourth bit) is off, the taps are arranged on different fields to represent the change in the time direction small.

Further, the vertical arrangement intervals of the cluster pattern of indexes (0110) and (0111) shown in FIGS. 15 (C) and (D) are shown in FIG. 15 (A) and (B). The maximum Laplacian value in the vertical direction is represented by making it narrower than the arrangement interval of the cluster tap patterns of (0100) and (0101). In other words, the larger the maximum value of the vertical Laplacian value, the smaller the spatial spread of the cluster tap pattern,
The smaller the maximum Laplacian value in the vertical direction, the larger the spatial spread of the cluster tap pattern.

Further, as shown in FIGS. 15B and 15D, when there is a change in the time direction of the indexes (0101) and (0111), that is, the bit in the time direction (4th bit).
When is on, the cluster tap patterns are arranged in the same field to greatly represent the change in the time direction. On the other hand, as shown in FIGS. 15A and 15C, when the bits in the time direction of the indexes (0100) and (0110) are off, the taps are arranged on different fields, so that the time direction is changed. Express the change in

16A and 16B, the clustering patterns of the indexes (1010) and (1011) shown in FIGS. 16C and 16D are arranged in the lower right direction. The Laplacian value in the downward-rightward direction, which is the maximum value, is represented large by making it narrower than the arrangement interval of the cluster tap patterns of (1000) and (1001). That is, the larger the maximum Laplacian value in the downward-rightward direction, the smaller the spatial spread of the cluster-up pattern, and the smaller the maximum Laplacian value in the downward-rightward direction, the wider the spatial spread of the cluster-up pattern. To increase.

Further, as shown in FIGS. 16B and 16D, when the bits in the time directions of the indexes (1001) and (1011) are ON, the cluster tap patterns are arranged in the same field. Greatly represent changes in the time direction. On the other hand, as shown in FIGS. 16 (A) and 16 (C), when the bits in the time directions of the indexes (1000) and (1010) are off, the taps are arranged on different fields to change the time direction. Express the change in

Further, the arrangement intervals of the cluster tape patterns of the indexes (1110) and (1111) shown in FIGS. 17C and 17D in the downward left direction are shown in FIGS. 17A and 17B.
The maximum Laplacian value in the lower left direction, which is the maximum value, is expressed by making it narrower than the arrangement interval of the cluster pattern of indexes (1100) and (1101). That is, the larger the maximum downward Laplacian value in the left downward direction, the smaller the spatial spread of the cluster tap pattern, and the smaller the maximum downward Laplacian value in the spatial spread of the cluster tap pattern. To increase.

As shown in FIGS. 17B and 17D, when the bits in the time direction of the indexes (1101) and (1111) are on, the cluster tap patterns are arranged in the same field. Greatly express the change in direction. On the other hand, as shown in FIGS. 17A and 17C, when the bits in the time direction of the indexes (1100) and (1110) are off, the taps are arranged on different fields to change the time direction. Express the change in

As described above, in the cluster tap pattern, there are cluster taps in the direction of large change (direction of large Laplacian value) in the spatial direction, and the larger the maximum value of the Laplacian value, the smaller the spread of the tap space. Is set. Further, when the Laplacian value in the time direction is large, the tap pattern is composed of taps in the same field, and when the Laplacian value in the time direction is small, the tap pattern is composed of taps on different fields (for example, frames).

In the above structure, the up converter 50 and the up converter 5 of the sixth embodiment described above are used.
The operation of each unit of 0 will be described. First, when the SD image signal S1 is input to the first class classification unit 50A, the Laplacian filters 51A to 51D apply Laplacian filters to a plurality of different level directions in the space. The resulting Laplacian value L0 ~
L3 is converted into absolute values by the absolute value conversion circuits 52A to 52D and output as absolute values a0 to a3.

Then, the absolute values a0 to a3 are detected by the maximum value detector 5
3, the maximum value is obtained from the absolute values of the supplied Laplacian values. At this time, the direction of the edge in the space of the input image signal is detected depending on which direction of the Laplacian (L10) is the maximum absolute value of the Laplacian values. Then, the value indicating the direction of the detected maximum value (maximum detection direction value) a10 is represented by 2 bits.
Further, by thresholding the maximum value, the flatness of the input SD image signal S1 can be expressed, and the value (flatness value) L10 representing the flatness is expressed by one bit.

Further, the Laplacian filter 51E can obtain the Laplacian value L4 in the time direction. The Laplacian value L4 is converted into an absolute value by the absolute value conversion circuit 52E and output as the absolute value a4. Then, the absolute value a4 is thresholded by the comparator circuit using the threshold value TH1, and is output as a value (time-direction change value) a11 representing a change in the time direction. In this way, it is possible to capture the rough characteristics of the input SD image signal S1.

The class classifying unit 50A uses the 4-bit control signal CT consisting of the 2-bit maximum value detection direction value a10, the 1-bit flatness value L10, and the 1-bit time direction change value a11 to the next-stage class classifying unit. Send to 50B.

The class classification unit 50B at the next stage sets the spatial cluster cut pattern based on the indexes (I0 to I3) corresponding to the 4-bit control signal CT. That is, the class classification unit 50B switches the selectors 55A to 55I according to the control signal CT, selects the cluster tap pattern for the input SD image signal S1 obtained through the register array 57, and sends it to the ADRC class classification unit 58. . As a result, highly accurate class classification can be performed based on the cluster pattern reflecting the activity of the input SD image signal S1.

Further, the SD image signal S1 is sent from the register array 57 to the predicted value calculation unit 50C at the next stage. In the prediction tap pattern unit 60, the prediction calculation unit 61 calculates the prediction coefficient RAM according to the control signal CT and the ADRC code c.
A prediction coefficient read from 59 and a prediction tap pattern for executing the prediction calculation are set, and the set prediction tap pattern is sent to the prediction calculation unit 61. Then, in the prediction calculation unit 61, the control signals CT and ADRC
Prediction calculation processing is executed by linear linear combination using the prediction coefficient read from the prediction coefficient RAM 59 according to the code c and the prediction tap pattern from the prediction tap pattern unit 60, and is output as an HD interpolation pixel.

According to the above configuration, the Laplacian filters 51A to 51E of the first-stage class classifying unit 50A perform the one-dimensional Laplacian processing in a plurality of directions and comprehensively judge the values to classify the first-stage class. Is executed and the cluster cut pattern of the next-stage class classification unit 50B is set according to the classification result of the first-stage class classification unit 50A, and then the next-stage class classification unit 50B uses this cluster cut pattern. By executing the class classification, it is possible to perform a highly accurate class classification reflecting the activity characteristic of the input SD image signal S1, and thus it is possible to obtain the same effect as that of the first embodiment.

(7) Seventh Embodiment FIG. 18 in which parts corresponding to those in FIG.
4 shows an up converter 65 according to an embodiment. Seventh
In the embodiment, similarly to the sixth embodiment, the next-class classification unit 5 is used according to the first-class classification result using the Laplacian filters 51A to 51E of the first-class classification unit 50A.
A cluster tap pattern for 0B class classification is set, and the class classification of the next stage is performed based on this cluster tap pattern, thereby executing the class classification in two steps.

The configuration of the up converter 65 of the seventh embodiment will be described with reference to FIG. The SD image signal S1 input from the input terminal IN is supplied to the class classification unit 65A. Then, the SD image signal S1 supplied to the class classification unit 65A is supplied to each of the Laplacian filters 51A to 51E and the delay unit 56A.

Five types of Laplacian filters 51A-5
1E performs Laplacian processing on the input SD image signal S1 in block units for each frame or field of the supplied SD image signal S1 in different directions for each filter, and outputs Laplacian values LO to L4. . These Laplacian filters 51A to 51E are the sixth ones shown in FIG.
The same Laplacian filter used in the up converter 50 of the embodiment is used. The Laplacian values L0 to L4 are sent to the quantizers 66A to 66E at the next stage, respectively.

The quantizers 66A to 66E execute the non-linear quantization after converting the Laplacians L0 to L4 into absolute values, and output the quantized values q0 to q4. For example, the quantizer 66
A to 66E are input SD image signal S1 by quantization.
Is converted into two kinds of quantized values q0 to q4 of 0 and +1, it can be classified into 32 (2 5) kinds of classes in total.
That is, quantization is performed so that 0 is assigned when the absolute value is small and 1 is assigned when the absolute value is large. Then, the quantized values q0 to q4 are sent to the synthesizing unit 67, and the synthesizing unit 67
Generates a 5-bit class code by synthesizing the quantized values q0 to q4, and sends this as a control signal CT representing the first-stage class classification to the next-class classification unit 65B.

Similar to the sixth embodiment, the class classification unit 65B switches the selectors 55A to 55I based on the control signal CT and selects the tap pattern of 9 pixels based on the SD pixels sent through the register 57. Selector 5
The tap pattern of 9 pixels selected by 5A to 55I is sent to the 1-bit ADRC class classification unit 58,
The 1-bit ADRC classifying unit 58 executes the ADRC on the inputted cluster pattern, thereby executing the second class classification to generate 512 (29) kinds of class codes. As a result, 16384 different classes can be formed by multiplying the classes of the first-class classification unit 65A and the classes of the second-class classification unit 65B.

The ADRC class classification unit 65 generates a 9-bit class code, and together with the first-stage class code supplied via the delay circuit 56B, the predicted value calculation unit 65C.
Is supplied to. The calculation unit 65C is the same as that of the sixth embodiment described above, and thus the description thereof is omitted. Further, regarding the cluster tap pattern used in the class classification unit 65B corresponding to the control signal CT from the first-stage class classification unit 65A,
Illustration is omitted. However, as the cluster tap pattern, the tap pattern is set such that the larger the Laplacian value, the smaller the space spread of the tap pattern, and the tap spreads in the direction of the larger Laplacian value. If the Laplacian value in the time direction is large, the tap pattern is composed of taps within the same field, and if the Laplacian value in the time direction is small, the tap pattern is composed of taps on different fields (for example, frames). It

With the above structure, the same effect as that of the above-described first embodiment can be obtained.

(8) Eighth Embodiment FIG. 19 in which parts corresponding to those in FIGS. 1 and 2 are assigned the same reference numerals shows a class classification unit 70 of an up converter according to the eighth embodiment. The SD image signal S1 input from the input terminal IN is sent in parallel to the ADRC class sorting unit 71 and the plurality of cluster selection units 72 (72A to 72G). The ADRC class classification unit 71 uses the four pixels around the SD pixel of interest from the input SD image signal S1.
The class classification is executed according to the pattern of the pixel level distribution by the DRC, and the ADRC code c obtained as a result is sent to the selection unit 73. On the other hand, in the selecting section 73, cluster tap selecting sections 72A to 72G for setting a spatial cluster tap pattern according to the signal characteristic of the input SD image signal S1.
Each tap pattern classified as a spatial class is transmitted.

That is, each cluster tap selection section 72A-
In 72G, as shown in FIGS. 20 and 21, the level distribution (73A, 73B, 74A, 74B, 75A, 75) by each ADRC code c obtained as a result of applying 1-bit ADRC to the surrounding 4 pixels with respect to the pixel of interest.
B, 76A, 76B, 77A, 77B, 78A, 7
8B, 79A, 79B, 80A, 80B), cluster tap patterns (81, 82, 83, 84, 85, 86, 87) for representing respective signal characteristics are set in association with each other.

For example, with respect to each of the level distributions 73A and 73B based on the ADRC code c, the characteristic of the level distribution is seen in the lower right diagonal direction (once the diagonal direction from the upper left end to the lower right end). Therefore, since the edge of the image is considered to be in that direction, the cluster tap pattern 81 in the diagonally downward right direction is made to correspond. Similarly, AD
With respect to each of the level distributions 74A and 74B based on the RC code c, since the characteristic of the level distribution is seen in the lower left diagonal direction (the diagonal direction from the upper right end to the lower left end), the cluster tap pattern 82 in the lower left diagonal direction. Correspond to.

Regarding the level distributions 75A and 75B, 76A and 76B based on the ADRC code c, the characteristic of the level distributions that are biased to the left and right sides can be seen, so the prediction taps that are biased to the left and right sides are collected. The cluster tap patterns 83 and 84 are made to correspond to each other.

Further, with respect to each of the level distributions 77A and 77B, 78A and 78B by the ADRC code c, since the characteristic of the level distribution of the taps biased to the upper side and the lower side can be seen, the cluster taps biased to the upper side and the lower side, respectively. The cluster tap patterns 85 and 86 that have been collected are associated with each other. Furthermore, the level distributions 79A and 79B and the level distributions 80A and 80B according to the ADRC code c
Since a characteristic of a steady level distribution can be seen for each of the above, the cluster tap pattern 87 using all the cluster taps is made to correspond.

As a result, the cluster-up selectors 72A ...
The cluster tap pattern set in 72G is selected by the selection unit 73 according to the ADRC code c,
The selected cluster tap pattern P10 is A in the next stage.
It is sent to the DRC class classification unit 26. The ADRC class categorizing unit 26 performs 1-bit ADRC on the SD image signal S1 according to the selected cluster pattern P10, and the resulting ADRC code d0 is used as address data for reading the prediction coefficient. Prediction coefficient R
Send to OM14.

The operation of the class classification unit 70 of the up converter according to the eighth embodiment having the above configuration will be described. First, the input SD image signal S1 is supplied to the ADRC class classifying unit 71 in the first stage, and the ADRC class classifying unit 71 in the first stage executes 1-bit ADRC by four pixels of the peripheral pixels for the target pixel, so that the class code c is obtained. Is generated. Further, the input SD image signal S1 is classified into a plurality of different cluster tap patterns 81 to 87 reflecting the signal characteristics in the cluster tap selection sections 72A to 72G.

Then, according to the class code c from the ADRC class classification unit 71 in the first stage, one cluster tap pattern P1 is selected from the plurality of cluster tap patterns 81 to 87.
0 is selected and sent to the ADRC class classification unit 26 in the next stage. The ADRC class classification unit 26 receives the input SD image signal S1 forming the selected cluster pattern P10.
1-bit ADRC is executed to generate the class code d0, and the prediction coefficient R in the subsequent stage is used as address data.
Send to OM14.

According to the above configuration, when the address data for selecting the prediction coefficient is generated, the ADRC of the first stage is used.
The cluster tap pattern is selected according to the spatial activity of the input SD image signal S1 detected by the class classification unit 71. Then, the class code d0 reflecting the spatial activity can be generated by executing the ADRC class classification 26 of the second stage using the selected cluster pattern P10, and thus the first code The same effect as the embodiment can be obtained.

Further, according to the above configuration, by changing the prediction tap pattern used for the prediction calculation according to the spatial activity, it is possible to shorten the calculation processing when the prediction tap is not small.

As in the first embodiment, the AD of the first stage
Based on the output signal from the RC class classification unit 71, the second
The number of quantization bits of the ADRC class classification unit 26 of the stage may be switched.

(9) Ninth Embodiment FIG. 22 shows an up converter 90 according to the ninth embodiment. The up-converter 90 performs rough class classification of spatial activity by one-bit ADRC in the first-stage class classification unit, and executes more detailed multi-bit ADRC in the next-stage class classification unit. In the up converter 90, the SD image signal S1 input through the input terminal IN is transferred to the first stage block 91 and the delay circuit 9
7 and 101 respectively. First stage block 91
23, n is centered on the pixel data of interest in the current frame or field (indicated by ⊚ in the figure)
Block into × m pixels (for example, 5 × 5 pixels in the figure),
The block data b1 obtained as a result is 1 bit A
It is sent to the DRC class classification unit 92.

The 1-bit ADRC class classification unit 92 outputs 5
1-bit ADRC for block data b1 of × 5 pixels
ADRC code resulting from executing the process
Adjust the timing of c10 through delay circuit 93A
Send to OM94. Furthermore, the 1-bit ADRC class classification unit 92 calculates the dynamic range D calculated at the time of calculation.
R is sent to the comparison circuit 95. At this time, the 1-bit ADRC class classifying unit 92 simultaneously sends the dynamic range DR and the pixel minimum level MIN to the multi-bit ADRC class classifying unit 96.

Here, the comparison circuit 95 sends the comparison result CR obtained by comparing the dynamic range DR with the threshold value TH to the ROM 94 through the delay circuit 93B. Then, based on this comparison result CR, the dynamic range DR is a threshold value.
If it is smaller than TH (CR = 0), the multi-bit ADRC class classification unit 96 does not execute the classification. On the other hand, when the dynamic range DR is larger than the threshold TH (CR = 1), the multi-bit ADRC class classification unit 9
Perform classification according to 6. However, in the case of this embodiment, the multi-bit ADRC class classification unit 96 is always executed.

Therefore, in the ROM 94 in the latter stage,
By ignoring the ADRC code c11 from the multi-bit ADRC class classifier 96, it is considered that the multi-bit ADRC class classifier is not executed. Also, if
In order to prevent the multi-bit ADRC class classification unit 96 from being executed, as shown by the dotted line in FIG.
CR may be supplied to the multi-bit ADRC class classification unit 96, and control may be performed so that the multi-bit ADRC class classification unit 96 is not executed.

On the other hand, the SD image signal S1 whose timing has been adjusted by delaying it from the input terminal IN through the delay circuit 97 is supplied to the next stage block 98 and the next stage block 9
23, for example, as shown in FIG. 23, the block data b2 of 9 pixels consisting of 3 × 3 pixels including the target pixel is set and sent to the multi-bit class classification unit 96. The multi-bit ADRC class classification unit 96 classifies the block data b2 using the dynamic range DR and the pixel minimum level MIN calculated by the 1-bit ADRC class classification unit 92, and the resulting ADRC code c1
1 is sent to the ROM 94.

As shown in FIG. 24, the ADR obtained as the classification result of the 1-bit ADRC class classification unit 92.
The table of the class code d0 having a hierarchical structure corresponding to the class classification by the C code c10 and the ADRC code c11 obtained as the classification result of the multi-bit ADRC class classification unit 96 is obtained by learning in advance, and the ROM9
4 is stored. The classification unit 90 determines the comparison result CR.
Based on the ADRC codes c10 and c11 obtained by the class classification of the 1-bit ADRC class classification unit 92 and the multi-bit ADRC class classification unit 96,
The d0 is read and sent to the prediction coefficient RAM 99 at the subsequent stage. Also in this prediction coefficient RAM 99, the prediction coefficient is registered according to the class code d0 which is obtained in advance by learning and is hierarchical.

From the prediction coefficient RAM 99, the class code
A prediction coefficient group is sequentially read out using d0 as address data, and is sent to the prediction calculation unit 100 that calculates an HD interpolation pixel by performing a product-sum calculation with the SD image signal S1. On the other hand, the predictive calculation unit 100 is supplied with the SD image signal S1 delayed by the delay circuit 101 to the predictive tap setting unit 102, and the predictive tap pattern setting unit 102 is shown in FIG. The prediction tap pattern as shown is sent to the prediction calculation unit 100. The predictive tap setting unit 102 generates an interpolated pixel using a predictive tap pattern as shown in FIG. As a result, in the prediction calculation unit 100, a linear prediction calculation is executed between the prediction coefficient group corresponding to the SD image signal S1 and interpolation pixels are generated.

The operation of the up-converter 90 of the ninth embodiment and each part of the up-converter 90 having the above structure will be described.

First, the SD image signal S1 input to the first stage block 91 of the up converter 90 is extracted in block units of 5 × 5 pixels centering on the SD pixel of interest, and classified by the 1-bit ADRC classifying unit 92. It Then, the dynamic range DR output from the 1-bit ADRC class classification unit 92 is subjected to threshold processing in the comparison circuit 95 using a predetermined threshold TH.

At this time, when the dynamic range DR is larger than the threshold value TH, the multi-bit ADRC class classification unit 96 executes the class classification to execute the 1-bit ADRC class classification unit 92 and the multi-bit ADRC class classification unit. The class code d0 is read from the ROM 94 based on the classification result of 96. On the other hand, the dynamic range D
If R is smaller than the threshold value TH, 1-bit A is set without executing the class classification of the multi-bit ADRC class classification unit 96.
The class code is read from the ROM 94 only based on the result of the class classification by the DRC class classification unit 92.

In this way, the 1-bit ADRC class classification unit 92 and / or the multi-bit ADRC class classification unit 9
According to the result of the classification of 6, the class code d0 is RO
It is read from M94 and sent to the prediction coefficient RAM 99 as address data. In this way, by roughly evaluating the spatial activity in the first stage and performing fine classification according to the spatial activity in the second stage,
Appropriate class classification that reflects the characteristics of the spatial activity of the input SD image signal S1 can be performed.

The prediction coefficient RAM 99 reads the prediction coefficient according to the class code d0 and sends it to the prediction calculation unit 100. The prediction calculation unit 100 includes a prediction tap pattern setting unit 1
The prediction tap pattern selected in 02 is subjected to the prediction calculation using the above-described prediction coefficient to obtain H
The D interpolation pixel is generated and output.

According to the above construction, by classifying coarsely in the first stage and finely classifying in the second stage, according to the signal characteristics of the input SD image signal S1 captured in the first stage, the sub-stages are more appropriately subdivided. The classified classification can be performed. This allows
The same effect as that of the above-described embodiment can be obtained, and further, when the classification is completed only in the first stage, the classification processing can be shortened.

(10) Other Embodiments In the above embodiments, the ADRC class classification method is used as the class classification method by data compression of the input image signal, but the present invention is not limited to this. ,
For example, DPCM (Differential Pulse Code Modulatin)
), VQ (Vector Quantization) and MSB (Most Sig
It is also possible to perform class classification by compressing data using a method using a nificant bit) or a method such as binarization or discrete cosine transform (DCT).

In the above embodiment, the case where the class classification unit 12 executes the class classification method in two steps or three steps has been described, but the present invention is not limited to this, and the rough classification of the first step is performed. Subsequently, if the adaptability is sequentially increased, it is also possible to execute a further multi-stage class classification. As a result, more accurate class classification can be achieved.

Further, in the above-mentioned embodiment, the case where the two-dimensional non-separable filter is used as the up converter has been described, but the present invention is not limited to this, and FIG.
An up-converter 110 having a vertical / horizontal separable filter configuration as shown in FIG.

In this up converter 110,
First, the SD signal S1 input through the input terminal IN is
It is supplied to the class classification unit 12 and the prediction calculation unit 111, respectively. The prediction calculation unit 111 uses the scanning line positions mode1 and mode.
2 of the vertical prediction calculation unit 111A and the horizontal prediction calculation unit 111B corresponding to 2 and the vertical prediction calculation unit 111C and the horizontal prediction calculation unit 111D corresponding to the scanning line positions mode3 and mode4.
Divided into types. In the class classification unit 12, a class code d0 corresponding to the input SD signal S1 is generated by applying the class classification unit of the above-described embodiment, and the prediction coefficient ROM1 which is a storage unit that stores the tap prediction coefficient in advance.
12 is sent. The prediction coefficient ROM 112 is a vertical coefficient ROM 1 that stores vertical and horizontal components of tap prediction coefficients.
12A and a horizontal coefficient ROM 112B. Class code d0 is vertical coefficient ROM 112A and horizontal coefficient R
It is supplied to each of the OM 112B.

First, the vertical prediction coefficient d6 output from the vertical coefficient ROM 112A is the vertical prediction calculation unit 111A and 1
Supplied to 11C. Then, the vertical prediction calculation unit 111A
And 111C generate vertical estimated values d7 and d8 by a product-sum operation using the input SD signal S1 and the vertical prediction coefficient d6. The vertical estimated values d7 and d8 are supplied to the horizontal prediction calculation units 111B and 111D at the next stage.

The horizontal prediction coefficient d9 generated from the horizontal coefficient ROM 112B is the horizontal prediction calculation unit 111B and 111D.
Is supplied to. Then, the horizontal prediction calculation units 111B and 1
The 11D generates HD pixel d10 and d11 signals by multiply-add operation of the vertical estimated values d7 and d8 and the horizontal prediction coefficient d9. This H
The D pixel d10 and d11 signals are supplied to the selection unit, and the selection unit 1
In step 5, the HD signal S2, which is the final output, is output from the output terminal OUT by appropriate rearrangement.

Further, in the above embodiment, the case has been described in which the HD pixel around the pixel of interest is generated from the SD pixel by using the prediction coefficient representing the correlation between the HD pixel of interest and the transmission pixel around the pixel of interest. The present invention is not limited to this,
Instead of the prediction coefficient, the predicted value of the HD interpolation pixel for each class may be set in advance and stored in the storage means. The signal conversion of the SD image signal into the HD image signal based on the predicted value uses an up converter 115 as shown in FIG.

This up converter 115 has an input terminal
The SD image signal S1 is sent to the class classification unit 12 through IN. The class classification unit 12 generates a class code d0 based on the characteristics of SD pixels around HD interpolation pixels, which are newly generated interpolation pixels, and predicts the predicted values ROMs 116A to 116A.
Send to 16D. Prediction values, which are prediction data of HD interpolation pixels obtained by learning in advance, are stored in the prediction value ROMs 116A to 116D in association with the class code d0 for each class. The prediction value ROM 116 reads the prediction values d20 to d23 of the HD interpolation pixels using the class code d0 as the address data, and outputs the output end OU through the selector 15.
Output from T. This makes it possible to obtain a high-resolution image signal in which the predicted values d20 to d23 are HD interpolation pixels of the signal pixels of the input image SD signal S1.

Here, there is a learning method using the weighted average method as the first method for obtaining the predicted value. In the weighted average method, the target pixel is classified into classes using SD pixels around the target pixel, and the target pixel integrated in each class (that is, HD pixel)
When the pixel value of is divided by the frequency incremented according to the number of target pixels, a prediction value is obtained by performing various processes on various images.

As a second method for obtaining the prediction coefficient, there is a learning method by normalization. In this learning method, first, a block composed of a plurality of pixels including a target pixel is formed, and a value obtained by subtracting the reference value of the block from the pixel value of the target pixel is normalized by the dynamic range in the block. Next, the predicted value is obtained by dividing the cumulative value of this normalized value by the cumulative frequency.

Further, in the above-mentioned embodiment, the spatial activity of the input SD image signal is evaluated, and the tap pattern for executing the class classification is selected based on the evaluation result. Is not limited to this, the spatial activity of the input SD image signal may be evaluated, and a prediction tap pattern calculated by the prediction coefficient and linear linear combination in the prediction calculation unit may be selected based on the evaluation result. . In this case, for example, the control signal CT from the class classification unit 50A of FIG. 11 is supplied to the prediction tap pattern unit 60 of the calculation unit 50C.

Further, in the above-described embodiment, the prediction coefficient of the prediction coefficient memory is read based on both the class code from the first-stage class classification section and the class code from the next-stage class classification section. However, the present invention is not limited to this, and the prediction coefficient of the prediction coefficient memory is read using either one of the class code from the class classification unit in the first stage and the class code from the class classification unit in the next stage. You may make it a structure like this.

Further, in the above-mentioned embodiment, the case where the SD image signal is converted into the HD image signal has been described, but the present invention is not limited to this, and it is used to generate an interpolation pixel when enlarging an image. May be. In addition, PAL (Phase Alternation by Line is the origin of NTSC signals.
The present invention may be applied to a signal conversion device that converts a signal with a small number of pixels into a signal with a larger number of pixels, such as a converter that converts a signal of a system. Further, it may be applied to a YC separator or the like that generates a signal with higher accuracy than the original signal.

Various modifications and application examples can be considered without departing from the gist of the present invention. Therefore, the gist of the present invention is not limited to the embodiments.

[0147]

As described above, according to the first to fourth aspects of the invention, the spatial activity of the input image signal for each block unit is evaluated and classified, and the block unit of the block unit is selected according to the activity code obtained as a result. Gradual classification is applied to the input image signal. As a result, the class code of the next and subsequent stages reflects the activity code of the first stage and the result of the class classification of the previous stage, so that highly accurate class classification can be performed, and as a result, the prediction coefficient or prediction based on this class classification result is obtained. It is possible to realize a signal conversion device and a signal conversion method that can predictively generate a highly accurate interpolated pixel for a low-resolution input image signal using a value and obtain a high-resolution image signal.

[Brief description of drawings]

FIG. 1 is a block diagram showing an upconverter composed of a two-dimensional non-separable filter according to the present invention.

FIG. 2 is a block diagram showing a configuration of a class classification unit shown in FIG.

FIG. 3 is a block diagram showing a configuration of an activity classification section.

FIG. 4 is a schematic diagram showing an arrangement example of SD pixels.

FIG. 5 is a schematic diagram showing a cluster pattern of a class classification unit.

FIG. 6 is a schematic diagram showing a prediction tap of learning data.

FIG. 7 is a flow chart showing a prediction coefficient learning procedure.

FIG. 8 is a schematic diagram for explaining the hierarchization of prediction coefficient ROM.

FIG. 9 is a block diagram for explaining an activity classification unit of the up converter of the second embodiment.

FIG. 10 is a graph showing a frequency distribution according to levels of ADRC code values.

FIG. 11 is a block diagram for explaining an up converter of a sixth embodiment.

FIG. 12 is a schematic diagram for explaining a one-dimensional Laplacian filter.

FIG. 13 is a schematic diagram for explaining an index that is a class code of a first-stage class classification unit.

FIG. 14 is a schematic diagram for explaining a prediction tap pattern having a horizontal characteristic as a classification result of a class classification unit.

FIG. 15 is a schematic diagram for explaining a prediction tap pattern having a vertical characteristic as a classification result of a class classification unit.

FIG. 16 is a schematic diagram for explaining a prediction tap pattern having a characteristic in an oblique direction as a classification result of a class classification unit.

FIG. 17 is a schematic diagram for explaining a prediction tap pattern having a characteristic in an oblique direction as a classification result of the class classification unit.

FIG. 18 is a block diagram for explaining an up converter of a seventh embodiment.

FIG. 19 is a block diagram for explaining a class classification unit according to the eighth embodiment.

FIG. 20 is a schematic diagram for explaining a cluster pattern of a class classification unit.

FIG. 21 is a schematic diagram for explaining a cluster pattern of a class classification unit.

FIG. 22 is a block diagram for explaining the up converter of the ninth embodiment.

FIG. 23 is a schematic diagram for explaining class generation block data used for generating a class code in the class classification unit of the ninth embodiment.

FIG. 24 is a schematic diagram for explaining class code hierarchization in the ninth embodiment.

FIG. 25 is a schematic diagram for explaining a prediction tap pattern of a prediction calculation unit of the ninth embodiment.

FIG. 26 is a block diagram showing an up-converter using a vertical / horizontal separable filter according to the present invention.

FIG. 27 is a block diagram showing an up converter that generates an HD pixel signal according to a prediction value of the present invention.

FIG. 28 is a schematic diagram for explaining the relationship between an SD image signal and an HD image signal.

FIG. 29 is a block diagram showing a conventional two-dimensional non-separable interpolation filter.

FIG. 30 is a block diagram showing a conventional vertical / horizontal separable interpolation filter.

[Explanation of symbols]

1 ... HD image, 2, 3 ... Interpolation filter, 4A to 4D
...... 2-dimensional filter, 5, 8, 15, 73 ...... Selector,
6A, 6B ... Vertical interpolation filter, 7A, 7B ... Horizontal interpolation filter 10, 50, 65, 90, 110, 11
5 ... Up converter, 12 ... Class classification unit, 1
3, 100 ... Prediction calculation unit, 14 ... Prediction coefficient ROM,
21, 35 ... Activity classification section, 22 ... Wide area tap selection section, 23 ... Standard tap selection section, 24 ... Narrow area tap selection section, 26,36,58,71 ... ADR
C class classification unit, 31, 38 ... Threshold value determination unit, 51A to
51E ... Laplacian filter, 52A-52E ...
Absolute value circuit, 53 ... Maximum value detector, 54 ... Comparator,
55A to 55I ... selector, 56A, 56B, 93
A, 93B, 97 ... Delay circuit, 57 ... Register array, 59, 99 ... Prediction coefficient RAM, 66A to 66E ...
... Quantizer, 72A to 72G ... Cluster-up selector.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Kenji Takahashi 6-735 Kita-Shinagawa, Shinagawa-ku, Tokyo Sony Corporation

Claims (44)

[Claims] 1. A signal conversion apparatus for converting an input first image signal into a second image signal different from the first image signal, wherein the spatial activity of the first image signal is evaluated and the activity is evaluated. A first class classification means for generating a code; a second class classification means for performing a stepwise class classification based on the activity code and generating a class code based on the classification result; Predicting coefficient storage means for storing the predictive coefficient for predictively generating the second image signal using the image signal, and the predictive coefficient storing means for reading the predictive coefficient according to the activity code and / or the class code. A prediction calculation unit that generates a second image signal by performing a prediction calculation on the first input image signal using a prediction coefficient. A signal conversion device comprising: 2. The first image signal is a low resolution image signal, and the second image signal is a high resolution signal having a higher resolution than the low resolution image signal. Item 1. The signal conversion device according to item 1. 3. The signal conversion device according to claim 1, wherein the second image signal is an image signal having more pixels than the first image signal. 4. The signal conversion according to claim 1, wherein the first class classification means evaluates the in-spatial activity and temporal activity of the first image signal to generate an activity code. apparatus. 5. The second class classification means sets a plurality of different pixel patterns for the first image signal,
Selecting a pixel pattern from the set plurality of pixel patterns according to the activity code, classifying the first image signal using the selected pixel pattern, and generating a class code. The signal conversion device according to claim 1, which is characterized in that. 6. The first class classification means evaluates in-spatial activity by using a dynamic range of each pixel in a region near the pixel of interest of the first image signal to generate an activity code. Claim 1 characterized by the above-mentioned.
The signal conversion device described in 1. 7. The first class classification means is responsive to a level distribution of quantized values obtained based on a dynamic range defined by pixels in a region near a pixel of interest of the first image signal. 2. The activity code is generated by evaluating the activity in the space, and the activity code is generated.
The signal conversion device described in 1. 8. The activity code is evaluated by the first class classification means for evaluating spatial activity by using a standard deviation obtained from a signal distribution of each pixel in a region near the pixel of interest of the first image signal. The signal conversion device according to claim 1, wherein 9. The first class classification means responds to a frequency distribution of quantized values obtained based on a dynamic range defined by pixels in a region near a pixel of interest of the first image signal. 2. The signal conversion device according to claim 1, wherein the activity code in the space is evaluated to generate an activity code. 10. The first class classification means is the first class.
2. The signal conversion according to claim 1, wherein the in-spatial activity is evaluated and an activity code is generated according to the frequency distribution of the adjacent pixel difference values in each pixel in the vicinity of the pixel of interest of the image signal of FIG. apparatus. 11. The first class classification means uses a Laplacian filter to evaluate the in-space activity on the basis of the Laplacian values obtained for different directions in the space, and to generate an activity code. The signal conversion device according to claim 1, which is characterized in that. 12. The first class classification means uses a Laplacian filter to evaluate the in-spatial activity and the in-spatial activity on the basis of Laplacian values obtained for different directions in the space and the time direction, respectively. The signal conversion device according to claim 4, wherein the signal conversion device generates an activity code. 13. The first class classification means is the first class.
2. The signal conversion apparatus according to claim 1, wherein the spatial activity is evaluated and an activity code is generated in accordance with the level distribution of each pixel in the area near the pixel of interest of the image signal. 14. The second class classification means is the first class.
By quantizing the first image signal on the basis of the dynamic range defined by each pixel in the vicinity of the pixel of interest of the image signal, the first image signal is classified into a class code. The signal conversion device according to claim 1, wherein the signal conversion device is generated. 15. The method according to claim 14, wherein the second class classification means adaptively switches the level resolution when quantizing the first image signal according to the activity code. Signal converter. 16. The second class classification means is the first class.
A pixel pattern over a wide area, a pixel pattern within a narrow area as compared with the wide area, and a standard pixel pattern corresponding to the wide area and the narrow area for the image signal The signal conversion device according to claim 4, which is characterized in that. 17. The first class classification means is the first class
Each pixel in the vicinity area of the pixel of interest of the image signal of the data is compressed by quantization, the spatial activity is evaluated according to the distribution characteristic of the level of the quantized value, and the activity code is generated, The second class classification means sets an appropriate pixel pattern by the activity code, sets a predetermined number of bits for each pixel of the pixel pattern by the activity code, and quantizes the pixel pattern. The signal conversion device according to claim 1, which is characterized in that. 18. A signal conversion device for converting an input first image signal into a second image signal different from the first image signal, wherein the spatial activity of the first image signal is evaluated and the activity is evaluated. A first class classification means for generating a code; a second class classification means for performing a stepwise class classification based on the activity code and generating a class code based on the classification result; A predicted value generated as an interpolated pixel signal of the image signal is stored, and a predicted value storing / reading unit that reads and outputs a predicted value corresponding to the activity code and / or the class code is output. A signal conversion device characterized by. 19. The first image signal is a low resolution image signal, and the second image signal is a high resolution signal having a higher resolution than the low resolution image signal. Item 18. The signal conversion device according to item 18. 20. The signal conversion device according to claim 18, wherein the second image signal is an image signal having a larger number of pixels than the first image signal. 21. The first class classification means is the first class.
19. The signal conversion device according to claim 18, wherein the activity code in the space and the temporal direction activity of the image signal are evaluated to generate an activity code. 22. The second class classification means is the first class.
A plurality of different pixel patterns are set for the image signal, the pixel pattern is selected from the set plurality of pixel patterns according to the activity code, and the first pixel pattern is selected using the selected pixel pattern. 19. The signal conversion apparatus according to claim 18, wherein the image signal of is classified into a class to generate a class code. 23. A signal conversion method for converting an input first image signal into a second image signal different from the first image signal, wherein the activity in space of the first image signal is evaluated to evaluate the activity. A code is generated, a stepwise class classification is executed based on the activity code, a class code is generated based on the class classification result, and the first code is generated according to the activity code and / or the class code. The prediction coefficient stored in the prediction coefficient storage unit for predictively generating the second image signal using the image signal is read, and the read prediction coefficient is used as the first input image signal. A signal conversion method, characterized in that the second image signal is generated by performing a prediction calculation. 24. The first image signal is a low resolution image signal, and the second image signal is a high resolution signal having a higher resolution than the low resolution image signal. Item 23. The signal conversion method according to Item 23. 25. The signal conversion method according to claim 23, wherein the second image signal is an image signal having more pixels than the first image signal. 26. The activity code is generated by evaluating in-space activity and temporal activity of the first image signal.
The signal conversion method described in. 27. A plurality of different pixel patterns are set for the first image signal, a pixel pattern is selected from the set plurality of pixel patterns according to the activity code, and the selected pixel pattern is selected. 24. The signal conversion method according to claim 23, wherein the first image signal is classified into classes using a pixel pattern to generate a class code. 28. The activity code is generated by evaluating the spatial activity by using the dynamic range of each pixel in the area near the pixel of interest of the first image signal. The signal conversion method described. 29. The spatial activity is evaluated according to the level distribution of the quantized value obtained on the basis of the dynamic range defined by the pixels in the region near the pixel of interest of the first image signal. 24. The signal conversion method according to claim 23, wherein the activity code is generated. 30. The spatial activity is evaluated by using the standard deviation obtained from the signal distribution of each pixel in the neighborhood area of the pixel of interest of the first image signal, and the activity code is generated. The signal conversion method according to claim 23. 31. The spatial activity is evaluated according to a frequency distribution of quantized values obtained based on a dynamic range defined by pixels in a region near a pixel of interest of the first image signal. 24. The signal conversion method according to claim 23, wherein the activity code is generated. 32. An activity in space is evaluated to generate the activity code according to a frequency distribution of adjacent pixel difference values in each pixel in a region in the vicinity of a pixel of interest of the first image signal. The signal conversion method according to claim 23. 33. The activity code is generated by using a Laplacian filter to evaluate the in-space activity based on the Laplacian values obtained for different directions in the space, and to generate the activity code. Signal conversion method. 34. The Laplacian filter is used to evaluate the in-spatial activity and the in-spatial activity on the basis of the Laplacian values obtained in different directions in the space and in the time direction, and generate the activity code. The signal conversion method according to claim 26. 35. The activity code is generated by evaluating the spatial activity in accordance with the level distribution of each pixel in the area near the pixel of interest of the first image signal. The signal conversion method described. 36. The first image signal is obtained by quantizing the first image signal on the basis of a dynamic range defined by pixels in a region near a pixel of interest of the first image signal. 24. The signal conversion method according to claim 23, wherein the signal is classified into classes to generate the class code. 37. The signal conversion method according to claim 36, wherein the level resolution at the time of quantizing the first image signal is adaptively switched according to the activity code. 38. A standard pattern corresponding to the first image signal, a pixel pattern over a wide area, a pixel pattern within a narrow area as compared with the wide area, and between the wide area and the narrow area. 27. The signal conversion method according to claim 26, wherein different pixel patterns are set. 39. Each pixel in a region near the pixel of interest of the first image signal is data-compressed by quantization, and the spatial activity is evaluated according to the distribution characteristic of the level of the quantized value. , Generating the activity code, setting an appropriate pixel pattern by the activity code, setting a predetermined number of bits for each pixel of the pixel pattern by the activity code, and quantizing The signal conversion method according to claim 23. 40. A signal conversion method for converting an input first image signal into a second image signal different from the first image signal, wherein the spatial activity of the first image signal is evaluated and the activity is evaluated. A code is generated, a stepwise class classification is executed based on the activity code, a class code is generated based on the classification result, and a predicted value storage means is generated according to the activity code and / or the class code. A signal conversion method comprising: reading out and outputting a prediction value generated as an interpolated pixel signal of the first image signal stored in. 41. The first image signal is a low-resolution image signal, and the second image signal is a high-resolution signal having a higher resolution than the low-resolution image signal. Item 40. The signal conversion method according to Item 40. 42. The signal conversion method according to claim 40, wherein the second image signal is an image signal having more pixels than the first image signal. 43. The activity code is generated by evaluating in-space activity and temporal activity of the first image signal.
The signal conversion method described in. 44. A plurality of different pixel patterns are set for the first image signal, a pixel pattern is selected from the set plurality of pixel patterns according to the activity code, and the selected pixel pattern is selected. The signal conversion method according to claim 40, wherein the first image signal is classified into classes using a pixel pattern to generate the class code.