JP2585217B2 - Filter device for adaptive subsample - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a filter device for sub-Nyquist sampling for digitally encoding a television signal and reducing its sampling frequency, and in particular, to improve the image quality by switching the sub-Nyquist sampling device. It is designed to work.

[Conventional technology]

First, FIG. 3 shows a configuration diagram of the sub-Nyquist sampling. In the figure, reference numeral 1 denotes an A / D (Analog to Dig) image signal.
ital) a digital video input terminal for inputting the converted signal; 110, a two-dimensional pre-filter for reducing the diagonal component of the signal from the digital video input terminal 1; 11, a thinning out of the output signal of the two-dimensional pre-filter 110 for each pixel Is a sub-sample switch for sub-sampling. 12 is a communication path for transmitting a signal from the sub-sample switch 11, and 111 is a communication path 1.
A two-dimensional interpolation filter for interpolating the signal from 2 is a digital video signal output terminal for outputting the output signal of the two-dimensional interpolation filter 111 to the outside.

Next, an example of a conventional configuration of the two-dimensional pre-filter 110 and the two-dimensional interpolation filter 111 is shown in FIG.

First, in the two-dimensional pre-filter 110, reference numeral 2 denotes a one-line delay unit (hereinafter referred to as 1H delay unit) for delaying the signal input from the digital video input terminal 1 by one line, and reference numeral 3 denotes an output signal of the 1H delay unit 2 further. 1H delayer for delaying one line, 4 is a one-pixel delayer (hereinafter referred to as 1D delayer) for delaying the output signal of 1H delayer 1 by one pixel, and 5 is further delaying the output signal of 1D delayer 4 by one pixel. A 1D delay unit, 9 is a 1D delay unit for delaying the output signal of the 1H delay unit 3 by one pixel, and 10 is a 1D delay unit for delaying the signal input from the digital video input terminal 1 by one pixel. 101 is 1H delay 2 and 1D delay 5,
An adder for obtaining the sum of the output signals of 9, 10 is provided, 102 is a divider for dividing the output signal of the 1D delay unit 4 by 2, and 103 is an adder 101.
Is an adder that divides the output of the divider by 8, and 104 is an adder that adds the outputs of the dividers 102 and 103.

Further, in the two-dimensional interpolation filter 111, 13 is the communication path 1
A 1H delay unit for delaying the signal from 2 by one line, a 1H delay unit for further delaying the output signal of the 1H delay unit 13 by one line,
19 is a 1D delay unit that delays the output signal of the 1H delay unit 14 by one pixel, 20 is a 1D delay unit that delays the signal from the communication path 12 by one pixel, and 15 is a 1D delay unit that delays the output signal of the 1H delay unit 13 by one pixel. The delay unit 16 is a 1D delay unit that further delays the output signal of the 1D delay unit 15 by one pixel. 105 is 1H delay 13 and 1D delay 19,2
An adder for adding the output signals of 0 and 16; 106, a divider for dividing the output signal of the adder 105 by 4; 107, an output signal of the 1D delay unit 15 and an output signal of the divider 106 It is an adder.

 Next, the operation will be described.

First, sub-sampling will be described with reference to FIG. In FIG. 3, a signal 3a obtained by sampling an image at a sampling frequency of 2fs is input to a digital video input terminal 1. The signal 3a is as shown in FIG. 6 when represented by a pixel arrangement, and as shown in FIG. 8 when represented by a two-dimensional space spectrum sampled in the x and y directions. In FIG. 6, Tx indicates one pixel interval, and Ty indicates one line interval. In FIG. 8, since the sampling frequency exists on a lattice point having a basic period of 1 / Ty, 1 / Tx, a two-dimensional spatial spectrum region in which a video can be reproduced without return noise is a horizontal spatial frequency of 1 / 2Tx. , A rectangular region having a vertical spatial frequency of 1/2 Ty. Normally, in subsampling, a PASS (Phase Alternative) in which the sampling frequency is shifted by 180 ° for each line
Sub-Nyquist Sampling) is adopted. Third
The signal 3c after the sub-sampling in the figure is a staggered sample point in FIG. 7 when represented by the pixel arrangement, and is represented as in FIG. 9 in the two-dimensional spatial spectrum sampled in the x and y directions. In FIG. 9, the sampling frequency is represented by a staggered grid point, and the video can be reproduced without aliasing noise.
The dimensional space spectrum region is a triangular region formed by connecting a horizontal space frequency of 1 / 2Tx and a vertical space frequency of 1 / 2Ty by a straight line. When a thin oblique line exists on an image, aliasing noise occurs. For this reason, it is important to reduce oblique components in the sub-sample filter.

In FIG. 3, a signal 3a input from the digital video input terminal 1 has a basic clock 2fs to reduce oblique components.
Is input to the two-dimensional pre-filter 110 that operates. The signal 3b that has passed through the two-dimensional pre-filter 110 becomes a signal with a diagonal component dropped, and is sub-sampled by the sub-sample switch 11 to become a signal 3c. Signal 3c is the sample clock fs
Since the signal is resampled every time, the image information is reduced by half. The signal 3c is transmitted using the communication path 12, and the transmitted signal is the sample clock fs.
It is a signal for each. Next, the sample clock is set to 2fs on the receiving side.
In FIG. 7, the missing pixels indicated by crosses in FIG. 7 are interpolated by the two-dimensional interpolation filter 111 and the oblique components are reduced. Then, the interpolated signal 3d is output to the digital video output terminal 48 as a signal whose sample clock has become 2 fs.

The importance of the filtering in the sub-sampling has been described above with reference to FIG. 3. Next, a specific example of the conventional filtering will be described with reference to FIG. By the time the signal 10a input from the video input terminal 1 becomes the input signal 10b of the sub-sample switch 11, the two-dimensional pre-filter 110 realizing the transfer characteristic of the following equation (1) is obtained.
As a result, the oblique component is dropped.

Z ^-L : One line delay on the image Z ^-1 : One pixel delay on the image Since the signal 10b is processed by the sampling clock of 2fs, the sub-sample switch 11 performs the sub-switching by the clock of fs which inverts the phase by 180 ° for each line. Sampled, and when this is represented by pixel arrangement, the staggered grid sampling shown in FIG. 7 is obtained.
The sub-sampled signal 10c is transmitted by the communication path 12 at the transmission clock fs. The signal transmitted in this manner is a 2 fs clock signal with 0 inserted at the missing sample point in FIG. On the receiving side to which a signal from the communication path 12 is input, the input signal is converted to a digital video output terminal.
Missing sample points are interpolated by the interpolation filter 111 having realized the transfer characteristic of the above equation (1) until the output video signal 10d of 48 is obtained.

The above filters are both filters for reducing the diagonal components in the two-dimensional pre-filter 110 and the two-dimensional interpolation filter 111.

[Problems to be solved by the invention]

The conventional sub-sample filter is configured as described above, and performs filtering in the oblique direction unconditionally even when there is no oblique high-frequency component in the image information. If a component is included, the image quality of that part will be degraded. Therefore, in order to prevent this, it is necessary to use a high-order filter, that is, a hardware-complex filter, and there is a disadvantage that the hardware scale is increased.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and can achieve an image quality having a higher horizontal and vertical resolution than the conventional one with the same hardware scale as the conventional one, and furthermore, has an adaptive type with less erroneous detection. It is intended to obtain a filter device for sub-sampling.

[Means for solving the problem]

An adaptive sub-sample filter device according to the present invention includes a post-filter (interpolation filter) including a horizontal low-pass filter (LPF), a vertical LPF, a two-dimensional LPF, and a local horizontal direction of an image. A comparison means for detecting a change and a change in the vertical direction and comparing them with each other;
Switching means for selecting any one of the output values of the LPF, the vertical direction LPF, and the two-dimensional LPF is provided.

[Action]

According to the present invention, a post-filter detects a local horizontal change and a vertical change of an image at a pixel of interest and its adjacent pixels, and according to the detection result, an image having a large amount of high-frequency components in the horizontal direction is vertical. By selecting the directional LPF, the horizontal LPF for an image having many vertical high-frequency components, and the two-dimensional LPF for an image that is neither of them, it is possible to obtain a high-resolution image with less erroneous detection.

〔Example〕

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows a configuration of a post-filter, that is, a receiving side, of an adaptive sub-sampling filter device according to an embodiment of the present invention. In the figure, 12 is a communication path, 13 is a 1H delay unit for delaying the signal from the communication path 12 by one line, and 15 is a 1H delay unit
13 is a 1D delay device that delays the output signal by one pixel, 16 is a 1D delay device that further delays the output signal of the 1D delay device 15 by one pixel, 1
7 and 18 output the output signals of 1H delay unit 13 and 1D delay unit 16 respectively.
, 23 is an adder for adding the two output signals of the dividers 17 and 18, 33 is a 1H delay unit for delaying the output signal of the adder 23 by one line, and 38 is a 1H delay unit 33. This is a 1D delay unit that delays the output signal by one pixel. These components make up the horizontal digital filter (horizontal LPF).
Is configured.

Reference numeral 14 denotes a 1H delay unit that further delays the output signal of the 1H delay unit 13 by one line, 19 denotes a 1D delay unit that delays the output signal of the 1H delay unit 14 by one pixel, and 20 denotes a delay of the signal from the communication path 12 by one pixel. 1D delay devices, 21 and 22 are dividers for dividing the output signals of 1D delay devices 20 and 19 by 2, 25 is an adder for adding the two output signals of dividers 21 and 22, and 35 is an adder Reference numeral 40 denotes a 1H delay unit that delays the output signal of the 1H delay unit 35 by one pixel. And from channel 12
A vertical LPF is configured by each component that reaches the 1D delay device 40.

27 is an adder for adding the two output signals of the adders 23 and 25; 31 is a divider for dividing the output signal of the adder 27 by 2;
Is a 1H delay unit that delays the output signal of the divider 31 by one line,
39 delays the output signal of the 1H delay unit 34 by one pixel
1D delay device. Each component from the communication path 12 to the 1D delay device 39 forms a two-dimensional LFP.

Further, 24 is a subtractor for subtracting the output signal of the 1D delay unit 19 from the output signal of the 1D delay unit 20, 26 is a subtractor for subtracting the output signal of the 1D delay unit 16 from the output signal of the 1H delay unit 13, 28, Reference numeral 29 denotes an absolute value circuit which takes the absolute value of the output signal of the subtracters 26 and 24, respectively. 30 is a comparator for comparing the two output signals of the absolute value circuits 28 and 29, 36 is a 1H delay device for delaying the output signal of the comparator 30 by one line, and 41 is a delay for one pixel of the output signal of the 1H delay device 36. 1D delay device, 42 is a 2D delay device for delaying the output signal of the comparator 30 by 2 pixels, 43 is a 2H delay device for delaying the output signal of the comparator 30 by 2 lines, 44 is the 2H delay signal of the 2H delay device 43 for 2 pixels. A 2D delay device for delaying, 45 is a ROM that inputs five output signals of the comparator 30, a 1D delay device 41, a 2D delay device, 44, and a 2H delay device 43 and outputs a 2-bit signal. Depending on the output signal
It is a changeover switch for selecting any one of the output signals of the 1D delay units 38, 39, 40, that is, any one of the horizontal direction LPF, the two-dimensional LPF, and the vertical direction LPF.

32 is a 1H delay unit that delays the output signal of the 1D delay unit 15 by one line, and 37 is a 1D delay unit that delays the output signal of the 1H delay unit 32 by one pixel.
Delay device, 47 is the output signal of switch 46 and 1D delay device
An adder for adding the output signal of 37 is provided, and 48 is a digital video output terminal for outputting the output signal of the adder 47 to the outside.

Next, the operation of the two-dimensional post filter, that is, the operation of the receiving side will be described with reference to FIG.

The signal 1a input from the total path 12 is a signal in which 0 data is inserted at the missing sample point in FIG. This input signal
1a is delayed by one line by a 1H delay unit 13, and further delayed by one pixel by 1D delay units 15 and 16, respectively. The output signal of the 1H delay unit 13 is divided by 2 by the divider 17, and the output signal of the 1D delay unit 16 is divided by 2 by the divider 18. Dividers 17, 18
Are added by the adder 23, further delayed by one line by the 1H delay unit 33, and delayed by one pixel by the 1D delay unit 38 to become the output signal 1b. Where the input signals 1a to 1D
The transfer characteristic of the horizontal LPF up to the delay unit 38 is represented by H (Z) = Z ^-1 · (1 + Z ^-2 ) · Z ^-2L / 2. This transfer characteristic is equivalent to performing the calculation of E = (A + B) / 2 to obtain the point E in FIG. 2 as the calculation on the pixel arrangement. At this time, since the signal 1a is a signal in which 0 data is inserted for each pixel, the signal 1b is
When the point is 0 insertion data, an output value in the horizontal direction LPF is obtained, and when the point E is not 0 insertion data, it becomes 0.

On the other hand, the signal 1a input from the communication path 12 is delayed by one pixel by the 1D delay unit 20, and is further divided by 2 by the divider 21. The output signal of the 1H delay unit 13 is further
, And the output is delayed by one pixel by the 1D delay unit 19. The output signal of the 1D delay unit 19 is divided by a divider 22
The output signals of the dividers 21 and 22 are added by an adder 25, further delayed by one line by a 1H delay unit 35, delayed by one pixel by a 1D delay unit 40, and output signal 1d. Become. Here, the input signal 1a is converted to a 1D delay 40
The transfer characteristic of the vertical LPF up to the output signal 1d is expressed by H (Z) = Z− ^L · (1 + Z− ^2L ) · Z− ² / 2. This transfer characteristic is calculated as
In FIG. 2, calculating the point E is equivalent to performing the calculation of E = (C + D) / 2. At this time, since the signal 1a is a signal in which 0 data is inserted for each pixel, an output value in the vertical direction LPF is obtained when the point E is 0 insertion data, and when the point E is not 0 insertion data, the signal 1d is obtained. It becomes 0.

On the other hand, the output signals of the adders 23 and 25 are added by the adder 27 and divided by 2 by the divider 31. Further, the output signal of the divider 31 is delayed by one line by a 1H delay unit 34,
The output signal 1c is delayed by one image by the 1D delay device 39 and becomes an output signal 1c.
Here, the transfer characteristic of the two-dimensional LPF from the input signal 1a to the 1D delay unit 39 is as follows: H (Z) = Z ^-1 · Z ^-L · (1 + Z ^-1 · Z ^-L ) (Z ^-1 + Z ^-L ) It is represented by / 4. This transfer characteristic is equivalent to performing the calculation of E = (A + B + C + D) / 4 to obtain the point E in FIG. 2 as the calculation on the pixel arrangement. At this time, since the signal 1a is a signal in which 0 data is inserted for each pixel, the signal 1c is
When the point is 0 data insertion data, an output value of the two-dimensional LPF is obtained, and when the point E is not 0 insertion data, it becomes 0.

The output signals of the three LPFs described above are selected by the following logic.

First, the output signal of the 1D delay unit 16 is subtracted from the output signal of the 1H delay unit 13 by the subtracter 26, and the absolute value of this output signal is obtained as the signal 1e by the absolute value circuit 28. On the other hand, the output signal of the 1D delay unit 19 is subtracted from the output signal of the 1D delay unit 20 by the subtractor 24, and the absolute value of this output signal is signaled by the absolute value circuit 29.
Get as 1f. The comparator 30 compares the signal 1e with the signal 1f,
Outputs a 1-bit signal. The output signal of the comparator 30 is delayed by one line by a 1H delay unit 36, and further delayed by one pixel by a 1D delay unit 41. The output signal of the comparator 30 is delayed by two pixels by the 2D delay unit 42. The output signal of the comparator 30 is delayed by two lines by a 2H delay unit 43, and further a 2D delay unit
44 delays two pixels. ROM45 is 1D delay 41, comparator
An address is designated by five output signals of the 30, 2D delay units 42, 44, and 2H delay unit 43, and a 2-bit output 1g is obtained.

The signal flow from the output of the comparator 30 to the signal 1g is determined by the interpolation of the missing sample point of interest, as shown in FIG. When the missing sample point γ at the upper right of the pixel, the missing sample point δ at the lower left of one pixel on the space, and the missing sample point ε at the lower right of the pixel on the space are taken as the points of interest, either the change in the horizontal direction or the change in the vertical direction is large. Depending on the point of interest α
Is used to determine which of the horizontal direction LPF, the vertical direction LPF, and the two-dimensional LPF is used.

In FIG. 5, the signal when the change in the horizontal direction is larger is 1, and the signal when the change in the vertical direction is larger is 0.
Signal 1g when the horizontal LPF is used in 00, the vertical LPF
The selection criterion for the changeover switch 46 is shown assuming that the signal 1g in the case of using the signal is 01 and the signal 1g in the case of using the two-dimensional LPF is 10.
In the figure, α, β, γ, δ, and ε indicate output signals of the comparator 30 when the missing sample points α, β, γ, δ, and ε are set as the target pixels.

The signal 1h selected based on such a selection criterion is added to the adder 47.
Thus, the output of the 1D delay unit 15 is added to the 1H delay unit 32 and the 1D delay unit 37 and the signal 1i delayed by one line and one pixel, and output from the digital output terminal 48 as an interpolated digital video output signal 1j. . Where 1D delay from input signal 1a
The transfer characteristic up to the output signal 1i of 37 is represented by H (Z) = Z− ² · Z− ^2L . In addition, since 0 is inserted by the subsample, one of the signals 1h and 1i is a signal that becomes 0 alternately. Therefore, the transfer characteristic including the insertion of 0 from the input signal 1a to the output signal 1j of the digital video output terminal 48 (when ε is the pixel of interest) is as follows: When the horizontal LPF is selected, H (Z) = Z ^-1 · (1 + Z ^-1 ) ² · Z ^-2L / 2 When the vertical LPF is selected, H (Z) = Z ^-L · (1 + Z ^-L ) ² · Z ^-2 / 2 2D LPF When selected, H (Z) = Z ⁻¹ · Z− ^L ｛Z− ^L (1 + Z ⁻¹ ) ² + Z ⁻¹ (1 + Z ^−L ) ² ｝ / 4. According to the selection logic, as an operation on the pixel arrangement in FIG. 2, a signal 1e corresponds to | A−B | and a signal 1f corresponds to | C−D | Assuming that the outputs of the comparator 30 at points β, γ, δ, and ε are Sα, Sβ, Sγ, Sδ, and Sε, respectively, the case of | A−B | <| C−D | The case of B | ≧ | C−D | corresponds to Sα = 1. Therefore, when Sβ = 0, Sγ = 0, Sδ = 0, Sε = 0, or when Sα = 0 and only one of Sβ, Sγ, Sδ, Sε is not 0, E = (A + 2E + B) / 2 Sβ = 1, When Sγ = 1, Sδ = 1, Sε = 1 or when Sα = 1 and only one of Sβ, Sγ, Sδ, Sε is not 1, E = (C + 2E + D) / 2 Sα = 0 and Sβ, Sγ, Sδ, When two or more of Sε are not 0, or when Sα = 1 and two or more of Sβ, Sγ, Sδ, and Sε are not 0, a selection of E = (A + C + 4E + B + D) / 4 is made. As described above, there is a risk of erroneous detection if the determination is made based only on the target pixel α point. Therefore, when the LPF is switched based on the comprehensive determination of the target pixel and the peripheral pixels, the erroneous detection is reduced. With this, the horizontal LPF is selected for the image with little horizontal change depending on the image, and the vertical LPF is selected with little misjudgment for the image with little vertical change, and the missing pixels are interpolated, and the high precision adaptive Interpolation filtering can be realized.

In such an apparatus of the present embodiment, a vertical LPF is applied to an image having many high-frequency components in the horizontal direction, and a horizontal LPF is applied to an image having many high-frequency components in the vertical direction. With this circuit scale, it is possible to faithfully reproduce image quality with higher horizontal and vertical resolutions than before.

In the post-filter shown in FIG. 2 of the above embodiment, the output value of any of the horizontal LPF, the vertical LPF, and the two-dimensional LPF is obtained by using a ROM whose contents are defined as shown in FIG. Although the selection is made, this can also be realized by a gate, and the same effects as in the above embodiment can be obtained.

〔The invention's effect〕

As described above, according to the sub-sample filter device according to the present invention, the horizontal filter LPF, the vertical LPF, and the two-dimensional LPF are provided in the subsequent filter, and the local horizontal change and the vertical change of the image are reduced. Detect with few false detections,
The output of the vertical LPF is output for an image having a large change in the horizontal direction, the output of the horizontal LPF is output for an image having a large change in the vertical direction, and the output of the two-dimensional LPF is selected and output for an image that cannot be determined as either of them. Therefore, there is an effect that high resolution image quality can be obtained with a small amount of hardware.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a post-filter of an adaptive sub-sampling filter device according to an embodiment of the present invention.
FIG. 3 is a layout diagram on a pixel for explaining the present invention and the conventional example as an operation on the pixel, FIG. 3 is a block diagram of a PASS system for explaining a PASS system, and FIG. FIG. 5 is an arrangement diagram on a pixel for explaining a selection logic of three LPFs, FIG. 5 is a diagram showing selection criteria of three LPFs in a post-filter of the present invention, and FIG. FIG. 7 is a pixel arrangement diagram showing sampling points after sub-sampling, and FIG.
The figure shows the two-dimensional space spectrum of the sampling points shown in FIG. 6, FIG. 9 shows the two-dimensional space spectrum of the sampling points shown in FIG. 7, and FIG. It is a block diagram which shows a pre-filter and an interpolation filter. 11… Subsample switch, 12… Communication channel, 13,14,
32 to 36: 1-line delay unit, 15, 16, 19, 20, 37 to 41: 1
Pixel delay unit, 17, 18, 21, 22, 31… Divider, 24, 26… Subtractor, 23, 25, 27, 47… Adder, 28, 29… Absolute value circuit,
30 comparator, 42, 44 two-pixel delay device, 43 two-line delay device, 45 ROM, 46 switch, 110
... two-dimensional post-filter, 111 ... two-dimensional post-filter. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

(57) [Claims] A PASS system (Phase Alte) for reducing the sampling frequency of a digitized television signal on a communication channel.
A digital filter device used for rnative sub-nyquist sampling, wherein an interpolation filter provided on the receiving side for interpolating missing pixels by interpolation has a transfer characteristic of H (Z) = (1 + Z ⁻¹ ) ² • Z- ^L / 2, where Z- ^L : spatial one-line delay, Z- ¹ : spatial one-pixel delay, and the transfer characteristic is H (Z) = (1 + Z- ^L ) ^2. • A vertical digital filter that is Z ^-1/2 and the transfer characteristic is H (Z) = ｛(1 + Z ^-1 ) ^{2 .Z -L} + (1 + Z ^-L ) ^2.
A two-dimensional digital filter that satisfies Z ⁻¹ ｝ / 4, a vertical difference absolute value V ₀ between the pixel values of one line above and below the space of the pixel of interest to be interpolated, and one pixel on the space of the pixel of interest Horizontal difference absolute value H ₀ between pixel values of previous and subsequent pixels
And a comparing means for obtaining the difference absolute values V _{−1, −1} and H when the missing sample point at the upper left of one pixel in the space of the target pixel is set as the target pixel.
_{−1, −1} , each absolute difference value V _{1, −1} , H _{1, −1} when a missing sample point at the upper right of one pixel on the space of the noted pixel is set as the noted pixel, 1 pixel on the space of the noted pixel The absolute value of each difference V _−1,1 , H _−1,1 when the lower left missing sample point is set as the target pixel, and the difference absolute value V− _1,1 and H− _1,1 when the lower right missing sample point is set as the target pixel. From the absolute difference values V _1,1 and H _1,1 , Vx, y> Hx, y (x = 1, -1, y = 1, −1), or V ₀ > H ₀ and x = 1, − If only one of the four combinations of 1, y = 1, -1 satisfies Vx, y ≦ Hx, y, the output value of the horizontal digital filter is calculated as Vx, y ≦ Hx, y (x = 1, -1, y = 1, -1) or
If only one of the four combinations of V ₀ ≤H ₀ and x = 1, -1 and y = 1, -1 satisfies Vx, y> Hx, y, the output value of the above vertical digital filter V ₀ > H ₀ and x = −1,1 and y = −1,1 when two or more combinations satisfy Vx, y ≦ Hx, y, or V ₀ ≦ H ₀
And two of the four combinations x = −1,1 and y = −1,1
A switching means for selecting and outputting the output value of the two-dimensional digital filter when Vx, y> Hx, y.