JPH0225105A - Digital filter - Google Patents

Abstract

PURPOSE:To reduce number of delay devices by using a delay device (delay memory) and a tap in common between a low pass filter for luminance signal and a high frequency emphasis filter. CONSTITUTION:A switch 10 selects R, G, B signals synchronously with a clock CLK and gives the result to a filter comprising 4 delay devices 11, 12, 13 and 14. Each coefficient of constant multiple devices 15-19 is set properly, and signals are added by an adder 20 to act the filter as a low pass filter for a luminance signal. Each coefficient of constant multiple devices 21-25 is set properly, and signals are added by an adder 26 to act the filter as a high frequency band pass filter. The output of the adder 20 and the output of the adder 26 through a variable constant multiple device 28 are added by the adder 26 to obtain a luminance signal Y. An output of R, G, B colors subject to low pass filtering is obtained by constant multiple devices 29-33, adders 34, 35 and a switch 36.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a digital filter used in color video cameras and the like.

(Prior art) In a single-chip color solid-state image sensor,
Each pixel is equipped with a specific color filter.
By subjecting these color signals to appropriate signal processing, a luminance signal Y and two color difference signals R-Y and B-Y are finally obtained. For example, a block diagram of a sensor equipped with RGB pure color filters in stripes is shown in FIG. The R, G, and B color signals from the sensor 4oh) are sent to the amplifier 41.4.
At 2°43, the gain is adjusted to give an equal response to white. Then A/D converter 44°45.4
Rγ,
Gγ and BY. Each color signal is filtered by a low-pass filter 48
After band-limiting, the process circuit 49 outputs the two color difference signals R-YL.

Output as BYL. On the other hand, the output Rγ after γ conversion,
Gγ and Bγ are switched by the switch 50 at the timing of each pixel.
is selected and band-limited by a low-pass filter 51. Thereafter, the high-frequency components are emphasized by the aperture compensation filter 52 and finally output as a luminance signal.

(Problem to be solved by the invention) Now, a luminance low-pass filter 51, an aperture compensation filter 52, and
Consider configuring the color low-pass filters 48 individually using FIR type digital filters. The number of taps for the brightness low-pass filter is N11.The number of taps for the color low-pass filter is N2. Set the number of taps of the aperture compensation filter to N
3, the total is (N+-1)+ (N3-1)+3
(N2-1) delays are required, leading to an increase in circuit scale.

The above describes the problems in the case of horizontal filters, but similar problems also apply to vertical or two-dimensional filters, and especially in that case, the delay is significant because it includes IH memory. This led to an increase in equipment costs.

(Means for Solving the Problems) The present invention was made by focusing on the following two points. 1) Since the luminance signal and the color signal are originally multiplexed into one signal,
The delays and taps used in the low-pass filters for the luminance signal and the three color signals can be made common.

2) Furthermore, the low-pass filter and aperture compensation filter for the luminance signal can also use the same tap and delay.

Taking note of these two points, the digital filter of the present invention is characterized in that a delay and a tap are made common between the horizontal and/or vertical luminance low-pass filter and the high-frequency emphasis filter.

(Function) As a result, the required delay can be significantly reduced in the present invention.

(Embodiment) FIG. 1 is an embodiment of a low-pass filter for brightness and color according to the present invention, and FIG. 2 is a waveform diagram of the main part of FIG. 1. Considering the case of a stripe sensor as shown in Figure 3, switch 1
Input signals R, G to 0.

B has a phase of 1 as shown in Figure 2 a), b), and c).
The clocks are out of alignment, and there is one signal component every 3 clocks.
The other periods are O. The clock here is a read clock for each pixel.

The switch 10 selects the R, G, and B signals in synchronization with the clock and outputs an output as shown in FIG. 2e) as a luminance signal. This output has four delays 11, 12, 13
．． It is input to a common filter consisting of 14 filters.

Also, to explain the brightness, constant multiplier 1516.17
, 18.19 are, for example, each 0. l/4.1/2.
It is set to 1/4.0 and together with the adder 20 constitutes a low pass filter for brightness. Also, constant multiplier 21,2
The coefficient of 2.23.24.25 is, for example, -1/4-1/
41-1/4-1. '4, and together with the adder 26 constitutes a high-frequency bandpass filter for luminance. This may of course be a bypass filter. The high-frequency component of the luminance, which is the output of the adder 26, is multiplied by an appropriate coefficient in a variable constant multiplier 27, and added to the low-pass filtered output of the luminance in an adder 28 to form a final luminance signal.

The constant of the variable constant multiplier 27 is such that the degree of high frequency emphasis can be controlled from the outside using an operation switch or the like.

Next, a color low-pass filter will be explained. The coefficients of constant multipliers 29, 30, 31, 32, and 33 are each 1/4
, 3/4, 1, 3/4, and 1/4.

For example, assume that the output of the switch 10 is g2 at a certain time t. Then, delay 1112.13. 14
The outputs of are b 2 + r 1g+ b+, respectively.

Therefore, the output of adder 34 is 1/4 g2i - 3/4
g, and the output of the adder 35 is 1/4 b2i-3
/4b, and the output of the coefficient multiplier 31 is rl.

At the next clock, when the output of the switch 10 becomes r2, the adder 34 receives 1/4 r 2 +3/43/4 r 1
．． 3/4 g2i-174 gl appears in the adder 35, and 3/4 g2i-174gl appears in the output of the coefficient multiplier 31.

Therefore, when the same operation is performed thereafter, the R, G, and B low-pass filtered outputs appear alternately at the outputs of the adders 34 and 35 and the coefficient unit 31. Therefore, if these are switched in synchronization with the clock using the switch 36, R, G
, B are obtained separately.

By configuring it in this way, you can prepare three sets of low-pass filters with HC = [1/43/413/41/4] and produce exactly the same output as when the above filter processing is performed independently on each of RlG and B. Obtainable.

As described above, in the present invention, a low-pass filter for luminance signals, a band-pass filter for aperture compensation, or
Since the bypass filter and the low-pass filter for color signals all share a delay and a tap, the number of required delays is greatly reduced.

In the above explanation, the coefficient of the constant multiplier 15.19 is set to 0.
Although the case where the luminance low-pass filter is set is described above, other constants such as negative constants may be set to set the characteristics of the luminance low-pass filter to desired characteristics. Also, although we have explained the case where the color signal filter has 5 taps, even if it is configured with a larger number of taps, if the outputs of every 3 or so taps are grouped together and added by one adder, Common low-pass filters for three colors can be realized.

Also, in the above explanation, the case where the output from the sensor is obtained independently for RlG and B is explained, but if the output can be directly read out as one signal in R-G-B order, the switch 10 is simply unnecessary. The present invention is effective.

(Second Embodiment) The above concept can be developed not only in one dimension in the horizontal direction but also in two dimensions in the horizontal and vertical directions. In that case, the 10 outputs of the switches shown in Figure 1 can be combined into two common outputs as shown in Figure 4.
All you need to do is manually create the dimensional filter. 20 in Figure 4
1 to 204, 206 to 209 and 211 to 214 are delays for one pixel, respectively, and 205 and 210 are IH memories each consisting of a (number of pixels in the horizontal direction - 4) stages of shift registers. In other words, 211 manpower is 2
01 human power to 2H.

There is an IH delay from the input of 206. Coefficient units 211-21
5.216 to 220 and 221 to 225 are, for example, (0, -1/8.-1/4.-1/8.0) (-1
/8, -1/4.7/4, -1/4, -1/8) (
0, -1/8, -1/4, -1/8.

0) and together with the adder 226 the following 2
Configure a dimensional bypass filter.

Coefficient units 227-231, 232-236 and 237-2
41, for example, in order (1/:121/321/8
1/32 1/:12 ) (1/16 1/16
1/4 1/16 1/+6 ) (1/321/3
21/81/321/32), and together with the adder 242, constitutes a two-dimensional low-pass filter for luminance as shown below.

The output of the adder 226 is externally set to a suitable constant by a variable constant multiplier 243, and added to the output of the adder 242 by an adder 244. Also, three sets of coefficients 245 to 249.2
50-254 and 255-259 are, for example, (
1/163/161/43/161/18 ) (1
/83/81/23/81/8 ) (1/163/
161/43/181/16). In each set of coefficient multipliers, the outputs of three taps are grouped together as in the first embodiment, and then the corresponding outputs of each set are grouped together, and a single adder will be added. For example, the output of the coefficient multiplier 245.248 is combined with the output of the coefficient multiplier 250.253.255.258, and added by the adder 260. With this configuration, the following color low-pass filters can be performed.

The operation of switch 272 is the same as switch 26 in FIG. In this case, IH along with one pixel delay
Since the memory delay is also shared among the three filters, the circuit scale can be significantly reduced. Also, of course, by replacing the one-pixel delay in Figure 1 with IH memory, we first perform common vertical filtering, and then perform horizontal filtering independently for brightness and color. You may do so.

(Third Embodiment) Next, a case where the present invention is applied to a sensor having an offset sampling structure as shown in FIG. 5 will be described.

642 pixels in the horizontal direction and 480 pixels in the vertical direction are arranged at intervals of half a pixel in every other row. It is assumed that the color filter is as shown in FIG. In such a sensor, filters of the same color are arranged vertically, so first,
It is preferable to interpolate information between pixels as shown by ■ in FIG. 5 using several pixels in the vertical direction. - If interpolation is performed in the vertical direction, R-G-B
Since 1284 pixels are arranged in a row, the subsequent processing is the same as in the first embodiment. Interpolation is a kind of low-pass filter, and as shown in Figure 6, where information is present is indicated by an O mark and where information is not present by an X mark, a zero is inserted at the X mark and a certain vertical low-pass filter is Just let it work. For example, for brightness, it is best to use a low-pass filter process such as . for example,
When YLI is used, A becomes 1/2 (B+C), and B and C remain B and C as they are. Furthermore, the hue may be determined by low-pass filter processing such as, for example.

However, if you select an odd-numbered tap like YLI for brightness, then select an odd-numbered tap like CLI for hue, and conversely, if you select an even-numbered tap for luminance, the hue will also change. Choose the one with even number of taps and vertical phase difference of brightness and color! ♂ I need to smoke. Further, when an even number of taps is used, the overall phase is shifted by 0.5 vertical lines, but there is no problem as long as the luminance and color phases match.

Now, even in the case of such a sensor, the vertical aperture correction filter can share the vertical luminance and color low-pass filters, delay, IH memory, and tap, as in the previous embodiment. FIG. 7 is a diagram showing an embodiment in which the present invention is applied to a sensor as shown in FIG. 5. After being read out in a zigzag manner from the sensor as shown at 31 in FIG.
The switch 90 is operated manually. switch 90
The signal in the Ai column is sent to the 642-stage shift register 93, and the signal in the Bi column is sent to the 642-stage shift register 91.
Output to. Therefore, the shift register transfer lock is CLKI, which is half CLKO. The switch 94 is
It is a switch with a (2x4) configuration that operates synchronously for each field.

Now, assuming that the output of the shift register 92 is 81 columns, the manpower and output of the shift register 91 are B3.
Bl, the manual power and output of the shift register 93 are A3, respectively.
．． It is A2. Therefore, if switch 94 is set to the upper side for the first field, switch 94 is set to the upper side.
4 outputs P1, P2 . P3. P4 has B3 respectively.
゜A3. B2. Data in column A2 is output. Also, the second
In the field, the switch 94 is on the lower side, so Pi, P2 . P3. P4 includes A3. B2. A2.
Data in column B1 is output. The switch 95 outputs data PI, P2 . P3. P4
and zero alternately, Ql, Q2 . Q3. Output Q4.

Therefore, the outputs Ql, Q2. Q3. If Q4 is subjected to a low-pass filter for brightness, an aperture correction filter, and a low-pass filter for color, it will be equivalent to inserting a zero at the X mark in FIG. 6 and performing each filtering.

Constant multipliers 96, 97, 98. 'J9. The coefficients are set to, for example, (0, 1, 1, 0), and together with the adder 100, form a low-pass filter in the vertical direction of brightness. With such coefficient settings, the response at a vertical resolution of 480 TV lines becomes zero. If you later record the field onto a still video floppy, use the coefficients (1/2.1/2.1/2.1/2) to remove vertical moiré.
It is recommended to set it to . In this way, the vertical resolution is 24
The response at 0 TV lines becomes zero, and the occurrence of moiré becomes extremely small.

Of course, an external switch may be provided to switch the coefficients depending on these two cases. Constant multiplier 101,1
For example, the coefficients of 02, 103, and 104 are (-1, 1,
1, -1), and together with the adder 105, forms a bandpass filter for high frequency components of luminance. This output is multiplied by an appropriate constant in a variable constant multiplier 106.

Adder 107 adds the outputs of 100 and 106 and outputs a vertically filtered luminance signal Y. constant multiplier 1
For example, the coefficients of 08, 109, 110, 111 are (1
/21/21/21/2) and adder 1
Together with 12, it forms a low-pass filter in the hue vertical direction. With such coefficient settings, the vertical resolution is 240TV.
Since the response of the book becomes 0, it is possible to effectively suppress the occurrence of color moiré in the vertical direction when, for example, sequential color difference line recording is performed on a still video floppy.

Also, in the case of movies, etc., the vertical color band is set to 24.
There is no need to set it to 0 in the 0TV tree, so for example, (Q, 1,
1. A coefficient setting such as O) may also be used. In this case, the color vertical low-pass filter is the same as the brightness vertical low-pass filter, so even the coefficients can be shared, resulting in constant multipliers 108°, 109.110, 111, . and adder 11
2 becomes unnecessary.

Note that, if necessary, an external switch may be provided to switch the optimum hue filter coefficient in the above two cases.

Further, this switch may be linked with the coefficient changeover switch of the luminance low-pass filter described above.

After vertical filtering as described above, the luminance and color signals may be processed by performing horizontal filtering processing independently.

(Fourth Embodiment) FIG. 8 shows another embodiment in which the present invention is applied to a sensor having an offset sampling structure as shown in FIG. In this case, unlike the case shown in FIG. 7, the signals from the sensors are read out in the order indicated by Sl in the first field, but are read out in the order indicated by Sl in the second field. Similarly to the third embodiment, in the first field, the switch 90 shifts the column A to the shift register 93.
and outputs the 81st column to the shift register 91. Therefore,
When the scan of Sl is scanning on the A2 and 82 columns, Pi, P2 . P3. P4 includes B2. A2. Bl,A
One column of data is output. In addition, the switch 9o transfers the Bi column to the shift register 93 in the second field.
+1 column is output to shift register 91. Therefore, Sl
When scanning on columns 82 and A3, Pi
, P2. P3. P4. The outputs of A3. B2. A
2. Data in column B1 is output. Therefore, by setting the method of reading from the sensor to S1 for the first field and Sl for the second field, the same output as in the case of FIG. P3. P4. Get above. Therefore, the scale of the circuit can be reduced by one tree of shift registers. The subsequent operation is the same as in the case of FIG. 7 described above.

The above explained the case where the output from the sensor is read out in real time in the order of 31 or S2 zigzag scan, but of course it is also possible to use memory to store information in -H memory and then read it out. good.

(Effects of the Invention) As described above, according to the present invention, the delay and tap of the low-pass filter for horizontal and/or vertical brightness are
By making the delays and taps of the low-pass filter for color signals and the delays and taps of the brightness aperture compensation filter common, it is possible to significantly reduce the circuit scale.

Note that the present invention is effective with both a stripe filter and a single-plate sensor having a so-called offset sampling structure.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the configuration of the present invention in a stripe filter type single-chip color camera. FIG. 2 is a diagram showing a timing chart of the signal in FIG. 1. FIG. 3 is a diagram showing a stripe filter type single-plate color camera. A block diagram of signal processing of a single-chip color camera. FIG. 4 is a diagram showing the configuration when the present invention is implemented with a two-dimensional filter. FIG. 5 is a diagram showing a color sensor with an offset sampling structure. These are two examples when applied to a sensor having a structure illustrating an offset sampling structure. 10: Switch 11.12, 13, 14: Delay 31.32, 33: Coefficient multiplier 20.26, 2
8, 34, 35: Adder 36: Switch 27: Variable constant multiplier 91.92.93: IH delay Fig. 2

Claims (4)

[Claims] (1) A digital filter characterized in that a delay and a tap are shared between a horizontal and/or vertical luminance low-pass filter and a high-frequency emphasis filter. (2) The digital filter according to claim 1, wherein the output of the high-frequency emphasis filter is multiplied by a variable constant in a variable constant multiplier and then added to the output of the low-pass filter. (3) The digital filter according to claim 1, wherein the delay and tap of the digital filter are common to the delay and tap of a horizontal and/or vertical color low-pass filter. (4) The luminance signal is a signal from a sensor having an offset sampling structure, and the first field is the (2_i−
1), (2_i) columns of pixels are read out in a zigzag pattern,
2. The digital filter according to claim 1, wherein the second field is a signal obtained by reading out pixels in columns (2_i) and (2_i+1) in a zigzag manner.