JPH05219498A - Moving picture encoder - Google Patents

Abstract

PURPOSE:To attain the on/off control of an in-loop filter by an arithmetic operation which is approximate to the in-loop filter arithmetic operation for the on/off judgement of the in-loop filter, and to obtain a satisfactory characteristic without increasing an arithmetic circuit or the like by appropriately combining a 1/2 picture element movement compensation prediction system with an in-loop filter system. CONSTITUTION:This device is equipped with a 1/2 picture element movement compensation predicting part 28 which operates a 1/2 picture element movement compensation prediction and searches the moving vector of a size notched by a 1/2 picture element, in-loop filter 27 which puts a filter on a prediction screen, and judging part 29 which outputs an on/off control signal for turning-on the in-loop filter 27 to the in-loop filter 27 when the moving vector searched by the 1/2 picture movement compensation predicting part 28 is set to the 1/2 picture vector.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding apparatus having a motion compensation predictor and an in-loop filter, and more particularly to a technique for controlling on / off of an in-loop filter.

[0002]

2. Description of the Related Art In high-efficiency coding of moving image signals,
A motion-compensated interframe predictive coding (hereinafter, simply referred to as motion-compensated prediction) system is known as a system for increasing the compression rate. In this motion compensation prediction, the screen is divided into blocks of a certain size (for example, blocks of 16 pixels × 16 lines), and for each block, the block with the smallest prediction error is searched for in the prediction screen. The resulting motion vector is sent to the receiving side as information indicating the motion of the object on the screen.

This motion-compensated prediction is usually performed on a prediction screen composed of real pixels, and an integer pixel vector (that is, a vector without a decimal part) is obtained as a motion vector. Such a method of performing prediction in units of integer pixels is called integer pixel motion compensation prediction.

Further, as will be described later, a virtual 1/2 pixel (half-pel) between real pixels is obtained by interpolation from surrounding real pixels, and the virtual pixel (interpolated pixel) is also used as a prediction screen. There is also motion compensation to use, which is called 1/2 pixel (half-pel) motion compensated prediction.

FIG. 7 shows CCITT Recommendation H.264. A H.261 compliant moving picture coding device is shown. This moving picture coding apparatus uses integer pixel motion compensation prediction as a prediction method, and in addition to this function, it further includes an in-loop filter.

In FIG. 7, reference numeral 1 is a subtracter for taking a difference between an input screen and a prediction screen and outputting a prediction error signal, 2 is a DCT circuit for performing a discrete cosine transform on the prediction error signal, and 3 is a discrete cosine transform. Quantizer for quantizing the prediction error signal, 4 is an inverse quantizer for inverse quantizing the output signal of the quantizer 3, 5 is an IDCT circuit for further performing discrete cosine inverse transform of the dequantized signal, and 6 is an IDCT An adder for adding a prediction screen to the output signal of the circuit 5 and outputting a locally decoded signal, a frame memory 7 for holding the locally decoded screen in frame units, and a variable delay circuit 8 for variably delaying the output signal of the frame memory , 9 is an in-loop filter for filtering the prediction value output from the variable delay circuit 8, 13 is a comparison between the input screen and the locally decoded screen of the frame memory 7, and the operation is performed based on the result. A motion compensation prediction unit that controls the delay amount of the variable delay circuit 8 generates a vector MV.

The in-loop filter 9 is for reducing a pulse-like signal that is difficult to code as a prediction error signal by applying a low-pass filter to the locally decoded signal to which the quantization error is added. The coding efficiency can be improved by controlling the on / off of the in-loop filter.

FIG. 8 shows an example of filter characteristics of the in-loop filter 9. Here, filtering is performed in units of blocks of 8 pixels × 8 lines. Explaining this filter processing, the areas (a) at the four corners of the block
The pixels of are used as they are. Each pixel in the area (b) of the middle 6 pixels on the 1st line and the 8th line is weighted by ½ to its own pixel and is divided into 1 / n by the pixels on the left and right.
The weighted values of 4 are summed to obtain the value of its own pixel after the filtering process. For each pixel in the area (c) of the middle 6 lines of the 1st pixel row and the 8th pixel row, the value obtained by weighting its own pixel by 1/2 and the weighted pixels by 1/4 respectively is summed up. Then, the pixel value of the pixel after filtering is set. Each pixel in the area (d) of 6 pixels × 6 lines inside the block has its own pixel weighted by 1/4, the pixels on the left, right, top, and bottom respectively weighted by 1/8, and the pixels adjacent to the four diagonal corners. 1/16 weighted values are summed to obtain the pixel value of the pixel after filtering.

There are several methods for on / off control of the in-loop filter when obtaining a prediction value in the moving picture coding apparatus of FIG. As one of the methods for obtaining the sub-optimal prediction value, first, the motion compensation prediction unit 13 obtains an optimum integer pixel vector (that is, integer pixel motion compensation prediction) as a motion vector, and a prediction block having this motion vector is obtained. In contrast, the motion compensation speed unit 13 creates prediction values both when the filter in the loop is turned on (when the filter calculation is performed) and when it is turned off (when the filter calculation is not performed). For each of the prediction errors, the prediction error of each is calculated, the one with the smaller prediction error is selected, and if the selected prediction error is the one when the filter in the loop is on, the filter 9 in the loop is actually selected. It is controlled to be turned on, and if it is off, the in-loop filter 9 is controlled to be turned off.

In the case of the "motion-compensated prediction + in-loop filter" method, when the motion-compensated prediction unit 13 detects a motion vector of an integer pixel and then a in-loop filter is applied to a prediction block having the motion vector. It is necessary to perform a calculation, and a calculation circuit therefor is required.

Therefore, there is the RM8 method as a method for simplifying the on / off control of the in-loop filter. In this method, the in-loop filter is turned on when the motion vector (MV _x , MV _y ) obtained by the motion compensation prediction unit 13 is not (0, 0) (that is, when there is motion). .. As a result, the in-loop filter calculation in the motion compensation prediction unit 13 for determining the on / off of the in-loop filter is unnecessary, but the on / off determination is performed without actually performing the filter calculation. As compared with the method of performing the on-off determination by performing the in-loop filter calculation described above, good characteristics such as S / N ratio cannot be obtained.

The motion-compensated prediction method is not limited to the integer pixel motion-compensated prediction method described above, but a virtual half pixel (half pel) between real pixels is used as a surrounding real pixel. There is also motion compensation in which the virtual pixel (interpolation pixel) obtained by interpolation is used as a prediction screen, and this is called 1/2 pixel (half-pel) motion compensation prediction. In this 1/2 pixel motion compensation prediction, the size of the motion vector is obtained not only in integer pixel steps but also in 0.5 pixel steps. This 1/2 pixel motion compensation prediction is MPEG
It is adopted in the method.

FIG. 9 shows a schematic configuration example of a moving picture coding apparatus in the 1/2 pixel motion compensation prediction method. In the figure, a subtractor 1, a quantizer 3, an inverse quantizer 4, an adder 6,
The frame memory 7 and the variable delay circuit 8 are the same as those in the embodiment of FIG. 7 described above, and the difference is that the motion compensation prediction unit is a 1/2 pixel motion compensation prediction unit that performs motion compensation prediction with 1/2 pixel. 13 'and does not have the in-loop filter 9.

FIG. 10 (I) shows the concept of the prediction screen of this 1/2 pixel motion compensation prediction. In the figure, the circle marks are actual integer pixels, and the cross marks are 1/2 pixels obtained by interpolation. (Interpolation pixel). As shown in (II) of FIG. 10, when the value of adjacent real integer pixels is A, B, C, and D, the calculation method for obtaining the 1/2 pixel is 1 between them.
The values a, b, c, d, and e of the / 2 pixel (interpolation pixel) are obtained according to the following equations. a = (A + B) / 2 b = (A + c) / 2 c = (A + B + C + D) / 4 d = (B + D) / 2 e = (C + D) / 2

[0015]

As described above, the conventional method for controlling the on / off of the in-loop filter in the moving picture coding apparatus is such that when the in-loop filter calculation is performed for the on / off determination. Therefore, there is a problem in that the device scale is increased because of the need for the arithmetic circuit, and the characteristics are deteriorated when the on / off determination is performed only based on whether or not there is a motion with the motion vector.

Further, when the device is made of LSI or the like, the above H. It is desirable to make a general-purpose one that can be applied to both of the H.261 system and the MPEG system separately, rather than making them separately.

The present invention has been made in view of the above problems, and an object thereof is to turn on the in-loop filter by properly combining the 1/2 pixel motion compensation prediction method and the in-loop filter method. Even if the in-loop filter calculation for the on / off judgment is not specially performed, it is possible to perform on / off control of the in-loop filter by an approximate calculation and obtain good characteristics without increasing the number of calculation circuits. Is to

[0018]

FIG. 1 is an explanatory view of the principle of the present invention. In FIG. 1, reference numeral 21 denotes a subtracter that takes the difference between the input screen and the prediction screen and outputs a prediction error signal, 22
Is a quantizer for quantizing the prediction error signal, 23 is an inverse quantizer for dequantizing the output signal of the quantizer 22, 24 is a locally decoded signal by adding a prediction screen to the dequantized prediction error signal , 25 is a frame memory that holds the locally decoded signal in frame units, 26 is a variable delay circuit that variably delays the output signal of the frame memory, and 27 is a predictive value that is output from the variable delay circuit 8. The filter in the loop to apply the input, 28 is the input screen and the frame memory 25
1 based on the result of comparison with the locally decoded screen of
A 1 / 2-pixel motion compensation prediction unit that generates a motion vector MV in 1/2 pixel increments and controls the variable delay circuit 26, and 29 is an in-loop filter based on the motion vector from the 1 / 2-pixel motion compensation prediction unit 28. 27 is a determination unit that determines ON / OFF of 27 and outputs an ON / OFF control signal.

As one form, the moving picture coding apparatus of the present invention performs a motion-compensated prediction with 1/2 pixels to obtain 1/2.
The 1/2 pixel motion compensation prediction unit 28 that obtains a motion vector having a size of a pixel, the in-loop filter 27 that filters the prediction screen, and the motion vector obtained by the 1/2 pixel motion compensation prediction unit 28 are 1 The determination unit 29 outputs an on / off control signal for turning on the in-loop filter 27 to the in-loop filter 27 at the time of the / 2 pixel vector.

As another form, the moving picture coding apparatus of the present invention performs a motion-compensated inter-frame prediction with 1/2 pixels to obtain a motion vector with a size of 1/2 pixel. The compensation prediction unit, an in-loop filter for filtering the prediction screen, the motion vector obtained by the 1/2 pixel motion compensation prediction unit is a 1/2 pixel vector, and the norm of the motion vector is a predetermined threshold value. ON / Control the filter in the loop ON when it is larger than
And a determination unit that outputs an off control signal to the in-loop filter.

In the above moving picture coding device, 1/2
The motion vector calculated by the pixel motion compensation prediction unit is 1/2
When the pixel vector is a pixel vector, the 1/2 pixel vector is converted into an integer pixel vector and transmitted, and when the integer is converted, the fractional part is truncated to be an integer so as to approach the (0, 0) vector. can do.

[0022]

According to the present invention, the calculation for obtaining 1/2 pixel in the 1/2 pixel motion compensation prediction unit 28 is regarded as an in-loop filter calculation for on / off determination, and the filter on / off control is performed. I do. Considered in this way, 1 /
When the 1/2 pixel vector is obtained as the motion vector in the 2-pixel motion compensation prediction unit 28, this is a loop when the block in which the intra-loop filter operation is performed on the prediction block by the conventional method is compared with the block that is not performed. Since it can be considered that it is approximately the same as the block in which the inner filter calculation is performed having a smaller prediction error, in this case, the determination unit 29 determines that and the loop inner filter 27 Control to turn on.

The condition for turning on the in-loop filter 27 depends not only on whether the above-described motion vector is a 1/2 pixel vector or an integer pixel vector, but in addition to that condition, the norm of the motion vector is predetermined. The in-loop filter may be controlled to turn on if it is above a threshold (i.e. only if there is motion), which allows for better on / off control.

When an integer pixel vector is required as a motion vector to be sent to the receiving side, the fractional part of the motion vector in 1/2 pixel increments obtained by the 1/2 pixel motion compensation prediction unit 28 is truncated to form an integer vector. To do. At this time, if the integerization is performed so as to approach the (0,0) vector, the amount of information to be transmitted can be further reduced.

[0025]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 2 shows the configuration of a moving picture coding apparatus as an embodiment of the present invention. This example is based on H.264. "Motion Compensation Prediction + In-loop Filter" so that it can be applied to H.
Has the function of. In FIG. 2, the subtractor 1 and the DCT
Circuit 2, quantizer 3, inverse quantizer 4, IDCT circuit 5,
The adder 6, the frame memory 7, the variable delay circuit 8, and the in-loop filter 9 have the same functions as those described in the above-mentioned conventional example.

The difference from the conventional example is that the motion compensation prediction unit calculates a motion vector in 1/2 pixel increments to perform 1/2 motion compensation prediction, and a 1/2 motion compensation prediction unit 10.
/ 2 ON / OFF of the in-loop filter 9 is determined based on the motion vector output from the motion compensation prediction unit 10 (motion vector having a size of 1/2 pixel)
The filter determination unit 11 sends an off control signal to the in-loop filter 9.

In the filter determination unit 11, the motion vector from the 1/2 motion compensation prediction unit 10 is a 1/2 pixel vector (that is, a vector having a size of integer + 0.5).
If it is, the in-loop filter 9 is controlled to be turned on, while if it is an integer pixel vector (that is, a vector having an integer size), it is controlled to be turned off. FIG. 4 shows a configuration example of the filter determination unit 11. Here, the motion vector (MV _x , MV _y ) is composed of m bits, and the least significant bit LSB is the decimal part. The filter determination unit 11 determines the fractional part (MV of the X-direction motion vector MV _x).
x -LSB) and fractional part of the Y-direction motion vector MV _y (M
OR gate V _y -LSB) are respectively input 111
The output signal of the OR gate 111 serves as an ON / OFF control signal for the in-loop filter 9.

The filter determining unit 11 further includes an integer circuit for converting the input motion vector MV in 1/2 pixel steps into an integer pixel motion vector (IMV). This circuit is required by CCITT Recommendation H.264. Two
This is because at 61, the motion vector sent to the receiving side is an integer pixel vector. FIG. 4 shows a configuration example of this integer circuit. This integer conversion circuit performs integer conversion by truncating the decimal part, and this integer conversion circuit is prepared for each of the X-direction and Y-direction motion vectors.

As shown in the figure, this integer conversion circuit comprises an adder 112 and a selector 113. The adder 112 is
The least significant bit L of the input motion vector (m bits)
"1" is added to SB (decimal part) and output. The selector 113 receives the input motion vector (m bits) and the output signal from the adder 112 as input terminals IN0 and IN0, respectively.
It is input to IN1, and by the most significant bit MSB (sign bit) of the motion vector MV, switching is performed to select the input terminal IN0 side when it is "0" and the input terminal IN1 side when it is "1", and output As the signal, the upper (m-1) bits of the motion vector MV (that is, the integer part excluding the decimal part) is output.

As a result, for example, the input motion vector is the binary number "0000001" corresponding to the decimal number +3.5.
When the value is 1.1 ", the value is selected by the selector 113 and the value obtained by removing the LSB is" 00000011 "(=
+3) is output, on the other hand, when the binary number “11111100.1” corresponding to the decimal number −3.5, the LS
The value "11111" obtained by adding "1" to the B in the adder 112
101.0 "is selected by the selector 113 and its LSB is selected.
The value “11111101” (= −3) obtained by removing is output. As described above, in this integer circuit, the number of bits of the motion vector to be transmitted is reduced as much as possible, that is, the amount of transmission information can be reduced as much as possible, because the integer conversion is performed in the direction closer to the (0,0) vector. I am trying.

As described above, the operation of the apparatus of this embodiment is such that the motion vector MV calculated by the 1/2 motion compensation predicting unit 10 at intervals of 1/2 pixel is 1 / X in at least one of the X direction and the Y direction. If it is a two-pixel vector, the filter determination unit 11 determines it and outputs a filter-on control signal to turn on the in-loop filter 9, while if the motion vector is an integer pixel vector, the filter is turned off. The control signal is output to turn off the in-loop filter 9.

Thus, the on / off control of the in-loop filter is performed depending on whether the motion vector is a 1/2 pixel vector or an integer pixel vector. By such control, the filter is appropriately turned on. The reason for turning on / off is that the function for calculating the filter in the loop (corresponding to FIG. 8) is also a filter function for obtaining 1/2 pixel (see FIG. 10).
Is also a low-pass filter function.
1 in place of the filter function for the in-loop filter operation of
By performing the on / off determination by using the filter function of the 1/2 pixel operation of 0 as the approximate function of the filter in the loop,
This is because it is possible to perform an approximately semi-equivalent operation.

That is, the fact that the motion vector obtained by the 1/2 motion compensation prediction unit 10 is a 1/2 pixel vector means that a prediction block of actual integer pixels (that is, a block in which no calculation is performed). It means that the prediction error of the 1/2 pixel prediction block obtained by interpolation (that is, the block to which the approximate in-loop filter function is applied by the 1/2 pixel calculation) was smaller than
This is considered to be approximately the same as the case where it is determined that the prediction error is smaller in the block to which the in-loop filter is applied by the conventional method of performing the in-loop filter calculation on the prediction block to determine ON / OFF. Therefore, if the on / off control is performed so that the in-loop filter is turned on when the obtained motion vector is a 1/2 pixel vector, it can be expected that the coding characteristic will be improved.

FIG. 5 shows a transmission rate S by computer simulation when the moving picture coding apparatus of this embodiment is used and when the moving picture coding apparatus of the RM system is used.
The N ratio characteristic is shown. In the figure, the horizontal axis is the transmission rate (× 64 kbps), the vertical axis is the S / N ratio (dB), and the characteristic marked with □ in the figure is the method of this embodiment and the characteristic marked with △ is (B). The RM8 system. As is clear from this figure, this method is about 0.2 compared to the RM8 method.
The S / N ratio is improved by .about.0.5 dB, and the improvement effect becomes greater as the transmission rate becomes higher. This is because the RM8 method does not perform the in-loop filter calculation at all to determine the on / off state of the filter, whereas this method approximately performs the 1/2 pixel calculation.
The S / N ratio is improved accordingly.

Incidentally, the apparatus of this embodiment is H.264. Although it has a circuit configuration applicable to the H.261 system, it can easily support the MPEG system by slightly changing the circuit configuration because it has a 1/2 motion compensation prediction unit. H. This is advantageous in that a general-purpose circuit that can support both the H.261 system and the MPEG system can be created with a single LSI.

Various modifications are possible in carrying out the present invention. FIG. 6 shows one of such modified examples. In this example, the configuration of the filter determination unit of the above-described embodiment is changed. As shown, the X-direction motion vector M
Comparator 114 for comparing V _x with a predetermined threshold TH1
And the Y direction motion vector MV _y is set to a predetermined threshold value TH2.
And a comparator 115 for comparing with the X direction motion vector MV _x
Fractional part (MV _x -LSB) of Y and motion vector MV in Y direction
_The OR gate 116 to which the fractional part (MV _y −LSB) of _y is input, the comparators 114 and 115, and the OR gate 1
The output signal of 16 is input and the AND gate 117 outputs the ON / OFF control signal.

This modification is the same as the above-mentioned R in the above-mentioned embodiment.
The concept of the M8 method is also incorporated, and the magnitude of the motion vector (MV _x , MV _y ) is a predetermined threshold value (TH
1, TH2) or less, the filter is not turned on. That is, the X direction of the motion vector MV, Y
At least one of the magnitudes in the directions is the respective threshold value T
When it is larger than H1 and TH2 and is a 1/2 pixel vector, the in-loop filter 9 is controlled to be turned on.

In this modification, it is determined whether or not the magnitude of the motion vector exceeds the threshold value in each X-direction component M.
V _x and Y direction component MV _y have respective threshold values TH ₁ ,
Although it is determined separately whether TH2 is exceeded, the present invention is not limited to this. For example, it may be determined whether or not the absolute value of the motion vector | MV | = (MV _x ² + MV _y ² ) ^1/2 exceeds a predetermined threshold value. That is, the condition for turning on the in-loop filter may be that the 1/2 pixel vector is obtained as the motion vector MV and the norm of the motion vector MV is larger than a predetermined threshold value.

[0039]

As described above, according to the present invention, since the 1/2 pixel motion compensation prediction circuit can be used as it is for the on / off control of the in-loop filter, the filter on / off determination can be performed. It becomes unnecessary to specially perform the in-loop filter calculation of. Further, the obtained characteristics are approximately those obtained by performing the in-loop filter calculation to perform the on / off determination, and therefore are improved as compared with the conventional RM8 system, and the improvement of the coding efficiency is large especially at a high transmission rate. Become.

Since the device of the present invention includes the 1/2 pixel motion compensation prediction circuit therein, the device of the present invention is compatible with the H.264 / AVC standard. Two
Not only can it be applied to the "motion compensation prediction + in-loop filter" method such as the 61 method, but the MP can be changed by a slight circuit change.
It can be easily applied to the EG system and the like, and the versatility of the device can be increased, which is advantageous when the LSI is formed.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a block diagram showing a moving picture coding apparatus as one embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of a filter determination unit in the apparatus of the embodiment.

FIG. 4 is a block diagram showing an integer circuit of a filter determination unit in the apparatus of the embodiment.

FIG. 5 is a diagram showing S / N ratio characteristics by computer simulation of filter control in a loop in comparison between the method of the present invention and the RM8 method.

FIG. 6 is a block diagram showing a configuration of a filter determination unit as another embodiment of the present invention.

FIG. 7 shows a conventional CCITT recommendation H.264. [Fig. 27] Fig. 27 is a block diagram illustrating a moving image encoding device according to the H.261 system.

FIG. 8 is a diagram showing an example of in-loop filter characteristics in a conventional device.

[Fig. 9] Fig. 9 is a diagram showing a moving image encoding apparatus based on the MPEG system.

FIG. 10 is a diagram showing a concept of a prediction screen of 1/2 pixel motion compensation prediction.

[Explanation of symbols]

 1 Subtractor 2 DCT (Discrete Cosine Transform) Circuit 3 Quantizer 4 Inverse Quantizer 5 IDCT (Inverse Discrete Cosine Transform) Circuit 6 Adder 7 Frame Memory 8 Variable Delay Circuit 9 In-Loop Filter 10 1/2 Pixel Motion Compensation Prediction unit 11 Filter determination unit 13 Motion compensation prediction unit 111, 116 OR gate 112 Adder 113 Selector 114, 115 Comparator 117 AND gate

Front page continuation (72) Katsuhiro Eguchi Inventor Katsuhiro Eguchi 3-22-8 Hakataekimae, Hakata-ku, Fukuoka-shi, Fukuoka Inside Fujitsu Kyushu Digital Technology Co., Ltd. Address within Fujitsu Limited (72) Inventor Kiichi Matsuda 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Within Fujitsu Limited

Claims (3)

[Claims] 1. A motion vector with a size of 1/2 pixel is obtained by performing motion compensation prediction with 1/2 pixel.
Pixel motion compensation prediction unit (28) and in-loop filter (27) for filtering the prediction screen
And a determination unit (29) for outputting an on / off control signal for turning on the in-loop filter to the in-loop filter when the motion vector obtained by the 1 / 2-pixel motion compensation prediction unit is a 1/2 pixel vector. ) And a moving picture coding device. 2. A 1 / 2-pixel motion compensation prediction unit for performing motion-compensated inter-frame prediction with 1/2 pixels to obtain a motion vector with a size of 1/2 pixel, and an in-loop filter for filtering a prediction screen. When the motion vector obtained by the 1/2 pixel motion compensation prediction unit is a 1/2 pixel vector and the norm of the motion vector is larger than a predetermined threshold value, the on-loop filter is controlled to be turned on. A moving image encoding apparatus, comprising: a determination unit that outputs an ON / OFF control signal to the in-loop filter. 3. When the motion vector obtained by the 1/2 pixel motion compensation prediction unit is a 1/2 pixel vector, the 1/2 pixel vector is converted into an integer pixel vector and transmitted, and the integer conversion is performed. 3. The moving picture coding apparatus according to claim 1, wherein the moving picture coding apparatus is configured to round down the fractional part so as to approach the (0,0) vector and convert it into an integer.