JP2001245297A - Moving image encoder, moving image decoder, moving image coding method and moving image decoding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a moving image coding system employing motion compensation prediction(MC) and discrete cosine transform(DCT) by which high-speed coding processing can be conducted, without incurring drastic deterioration in image quality. SOLUTION: In the moving image coder of this invention where motion compensation prediction(MC) is conducted, on the basis of a motion vector(VT) obtained from a received image and an image resulting from reproducing image data that are coding-processed, the obtained MC signal is used for in-frame coding processing, and the received image is processed sequentially in combination use with inter-frame coding processing which does not use the MC signal, a conversion coefficient obtained by applying the discrete cosine transform(DCT) to the processed received image is quantized for the variable length coding. A means 105A detecting the VT has a means that detects the VT with precision of an integral number of pixels by each block consisting at least of 2N×2N (N is a natural number) pixels and a means, that detects the VT with a precision of a half pixel and has a configuration employing a discrimination means discriminating a quantity width used for quantizing the conversion coefficient that adopts the VT, obtained through the detection of the VT with the precision of an integer number pixels without detecting the VT with the precision of a half pixel, when the discrimination result denotes the quantization with rough quantization in excess of a prescribed level, and that adopts the VT by the detection of the VT with the precision of half a pixel in other cases.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an international standard system for video coding, for example, ISO / IEC JTC1 / SC2.
9WG11 MPEG1 (Motion Picture Experts Group
1), MPEG2, (Motion Picture Experts Group 2)
And MPEG4 (Motion Picture Experts Group 4)
And H. of ITU-T. In the hybrid coding system using motion compensated prediction (MC) and discrete cosine transform (DCT) adopted in H.263, the encoding and decoding processes are speeded up without significant deterioration in image quality. And a moving picture decoding apparatus.

[0002]

2. Description of the Related Art MP which is an international standard system of moving picture coding
In EG1, MPEG2 and MPEG4, an MC + DCT method is adopted as a basic coding method. Hereinafter, with reference to a reference document (ed. By Miki, "All about MPEG-4", Chapter 3, Industrial Research Institute, 1998), MPEG4
The description will be given according to a verification model.

<Outline of MC + DCT method> First, an outline of an encoding method of the MC + DCT method will be described with reference to FIGS.
This will be described with reference to FIG.

A moving picture coding apparatus of the MC + DCT system is:
As shown in FIG. 8, the difference value calculation unit 101, the motion compensation prediction unit (MC) 102, the intra-frame / inter-frame (In
(tra / Inter) switching unit 103, frame memory (FM) 104, motion vector detection unit (ME) 10
5, discrete cosine transform section (DCT) 106, quantization section (Q) 107, variable length coding section (VLC) 108, inverse quantization section (IQ) 109, inverse discrete cosine transform section (IDC)
T) 110, an adder 111, an output buffer (Buffe)
r) 112 and a rate control unit (Rate Control) 113.

[0005] As shown in FIG. 9A, an image signal converted into a macro block (MB) is supplied to the image signal input line 11.

In the example shown in FIG. 9, the MB includes four blocks each composed of 8 × 8 pixels, and the luminance signal (Y) includes four blocks and the color difference signal (U or V) includes one block. The difference value calculation unit 101 calculates a difference between the image signal supplied via the image signal input line 11 and the motion compensation prediction signal supplied via the signal line 12, and this difference signal is transmitted through the signal line 13. The signal is supplied to the discrete cosine transform unit 106 via

[0007] Intra-frame encoding (In
ter) mode, the motion compensation prediction unit (MC) 1
02 is supplied via the intra / inter-frame switching unit 103, but is not supplied in the case of the inter-frame coding (Intra) mode.

That is, the signal of the image signal input line 11 is supplied to the signal line 13 instead of the difference signal.
The switching between the Intra / Inter modes is determined by the motion vector detection unit 105 as described later, and the determination result is supplied to the intra-frame / inter-frame switching unit 103 via the signal line 14 as a switching signal. This is performed by switching the output of the motion compensation prediction unit 102 by the intra-frame / inter-frame switching unit 103 corresponding to the determination result.

A motion compensation prediction signal generated by a motion compensation prediction unit (MC) 102 is a motion vector detection unit that detects a motion vector detected by a motion vector detection unit 105 from a signal of an already encoded frame stored in a frame memory 104. Generated according to the information.

In a discrete cosine transform unit (DCT) 106, a signal supplied via the signal line 13 is subjected to a discrete cosine transform, and a signal obtained by the discrete cosine transform is supplied to a quantization unit 107. The discrete cosine transform coefficients quantized by the quantization unit 107 are supplied to a variable length coding unit 108, where they are subjected to variable length coding and an inverse quantization unit (I
Q) 109, where it is dequantized. The transform coefficient obtained by the inverse quantization by the inverse quantization unit 109 is supplied to an inverse discrete cosine transform unit (IDCT) 110 to generate a reproduction signal for the signal line 13, and then to the addition unit 111. .

The adder 111 adds the signal supplied from the inverse discrete cosine transformer 110 and the signal supplied via the signal line 12 to reproduce the image signal, and then sends the signal to the frame memory 104. Lever frame memory 104
To accumulate.

The variable-length encoding unit 108 encodes discrete cosine transform coefficients and motion vector information (not shown) quantized by the quantization unit (Q) 107, and multiplexes to generate a bit stream. And the output buffer 11
Feed to 2. The output buffer 112 outputs a bit stream according to the characteristics to a network or a storage medium via the signal line 15.

From the output buffer 112, the rate control unit 1
13 is provided with information on the bitstream storage amount via a signal line 16, and the rate control unit 113 determines a quantization parameter in accordance with the bitstream storage amount of this buffer, and via a signal line 17. This is supplied to the quantization unit 107 and the inverse quantization unit 109.

Here, when the accumulation amount in the buffer increases, the quantization parameter is increased to reduce the generated code amount, and when the accumulation amount in the buffer decreases, the quantization parameter is decreased. By making it smaller, the generated code amount is controlled to be constant.

The outline of the MC + DCT video coding apparatus has been described above.

Next, a motion vector detecting unit (ME), which is an important component of the moving picture coding apparatus of the MC + DCT system,
105 will be described.

<Details of Motion Vector Detector> A motion vector detector (ME) will be described with reference to FIGS. 10, 11 and 12.
105 will be described.

FIG. 10 is a general flowchart showing the processing contents of the motion vector detecting section 105. The processing according to the flow is executed for the luminance signal of each MB.
First, in step S201, a motion vector (M
V) is initialized. This is a process of setting the initial value of the motion vector to zero vector.

Next, in step S202, the absolute value sum (SAD) of the motion compensation prediction error signal at zero vector
0) is larger than a preset threshold value (TH0). Then, this step S20
As a result of the determination in step 2, if the absolute value sum (SAD0) of the motion compensation prediction error signal at zero vector is smaller than the threshold value (TH0), the process ends as that zero vector is detected. .

On the other hand, as a result of the determination in step S202,
Sum of absolute values of motion compensation prediction error signal at zero vector (S
AD0) is larger than the threshold value (TH0), a motion vector (MVint) with integer pixel precision is detected in step S203, and the next step S2 is performed.
At 04, the intra-block activity (ACT) of the luminance signal of the MB is calculated.

Here, the activity within the block (AC
T) is, for example, the sum of absolute values of the differences between the average value in the block and each pixel value in the block.

Next, the process proceeds to step S205, where the sum of absolute values of the motion compensation prediction error signal at MVint (S
ADi) is smaller than ACT. Then, as a result of the determination in step S205, SADi
Is larger than ACT, step S2
06, where the encoding mode of the MB is changed to In.
The motion vector detection ends as tra.

Here, if the encoding mode of the MB is not set to Intra in S206, the encoding mode of the MB becomes Inter.

On the other hand, as a result of the determination in step S205,
If SADi is smaller than ACT, the process proceeds to step S207, where a motion vector (MVhalf) with half-pixel accuracy is detected, and in step S208, a motion vector (MV4mv) for each 8 × 8 pixel is calculated. To detect.

In some cases, a motion vector at an integer pixel position is detected as a result of motion vector detection with half-pixel accuracy (O in FIG. 12).

Further, as shown in FIG.
The detection of a motion vector every 6 pixels is called a 1MV mode, and the detection of a motion vector every 8 × 8 pixels is called a 4MV mode.
In the mode, the motion vector detection is performed with half-pixel accuracy.

Next, in step S209, MV4mv
The absolute value sum (SAD4mv) of the prediction error signal at
It is determined whether or not the absolute value of the prediction error signal in the half is greater than the sum of the absolute values (SADhalf). Then, this step S20
As a result of the determination in S9, if SAD4mv is larger than SADhalf, the process proceeds to step S210, and MVha
Let lf be the detected MV, and end the motion vector detection.

On the other hand, as a result of the determination in step S209,
If SAD4mv is smaller than SADhalf, the process proceeds to step S211 and MV4mv is detected as MV
And ends the motion vector detection.

FIG. 12 is a diagram for explaining half-pixel precision motion compensation. In FIG. 12, the predicted value at the half pixel position (△ or ×) is obtained using the pixel value at the integer pixel position (O). For example, the value of twice the pixel e and the value of the pixel f are
It is obtained by the following equation.

E = (a + b + c + d + 2-rc) / 2 f = (c + d + 1-rc) / 2 Here, as shown in the partial enlarged view of FIG. 12 (a) in FIG. Where a, b, c, and d are integer pixel positions around e, f is a pixel position between the integer pixel positions, and indicates a predicted value at that pixel position, and rc = 0. or 1. The value of rc may be a fixed value or may be switched periodically. That is, the above calculation is required to obtain the pixel value at the half pixel position in the motion vector detection, and there is a problem that processing time is required. Therefore, it is necessary to make a trade-off between the amount of processing for detecting a half-pixel accuracy motion vector and the improvement in prediction efficiency by half-pixel accuracy motion compensation prediction.

The above is the processing content of the motion vector detection unit (ME) 105 in the MC + DCT video coding apparatus.

[0032]

By the way, when a moving picture is coded by the above-described MC + DCT method, if the quantization width is coarse, the reproduced picture will be distorted in a block-like manner, resulting in a poor appearance. Is well known. In order to reduce the influence of the distortion, a decoder (moving image decoding device) often performs a filtering process on a reproduced image, which is a post filter.

However, when the filtering process is performed, the appearance is improved, but the processing amount for the filtering process is increased, so that there is a problem that the load on the reproduction processing system is increased.

If the quantization width is made smaller, the block distortion becomes less noticeable, but on the other hand, the processing load in the system including the MC and DCT transforms, which leads to the encoding process, becomes heavy, and smooth moving image encoding becomes possible. There is a worry that it will not be possible.

In the moving picture coding apparatus, the discrete cosine transform coefficients and motion vector information quantized by the quantization section are coded by the variable length coding section, and then multiplexed to generate a bit stream. This is temporarily stored in an output buffer as compressed moving image data. Then, the output buffer outputs a bit stream of the compressed moving image data to a network or a storage medium according to the characteristics thereof.

At this time, the rate control unit determines the quantization parameter Qp according to the bit stream storage amount of the output buffer, and supplies the quantization parameter Qp to the quantization unit and the inverse quantization unit to adjust the code amount.

That is, when the accumulation amount in the buffer increases, the quantization parameter Qp is increased to reduce the generated code amount, and when the accumulation amount in the output buffer decreases, the quantization parameter By reducing Qp, control is performed so that the generated code amount becomes constant.

Therefore, the quantization width used for the quantization process changes depending on the situation. This means that the image quality at the encoding stage originally differs. That is, when the quantization width is small, the image quality is high, but when the quantization width is large, the image quality is low. This image quality varies,
Unless the accuracy of the motion vector detection used for the motion compensation prediction is appropriately reflected, the processing load is unnecessarily burdened, which is wasteful.

In order to optimize the processing load, it becomes necessary to increase the performance of a device such as a processor used in a moving picture encoding / decoding processing system more than necessary. However, smooth coding and reproduction of moving images is a problem.

Accordingly, it is possible to reduce the processing load in the moving picture encoding processing system and the decoding processing system, to minimize the load of the filter processing during reproduction, and to reduce the occurrence of block-like distortion. There is a demand for the development of a moving image encoding / decoding technique capable of suppressing such a phenomenon.

Therefore, an object of the present invention is to provide a hybrid coding system using motion compensation prediction (MC) and discrete cosine transform (DCT), which allows high-speed coding processing without significant deterioration in image quality. It is an object of the present invention to provide a moving picture coding apparatus, a moving picture decoding apparatus, and a moving picture coding / decoding method capable of realizing video coding.

[0042]

In order to achieve the above object, the present invention is configured as follows. That is, first, [1] motion vector detecting means in a video coding method using motion compensated prediction and discrete cosine transform, wherein the motion vector detecting means includes at least 2N × 2N pixels (N: natural number ) Has means for detecting a motion vector with an integer pixel precision without a block and means for detecting a motion vector with a half-pixel precision, and a quantization parameter for quantizing a discrete cosine transform coefficient is smaller than a certain threshold value. And a motion vector obtained by the motion vector detection means of integer pixel precision without performing the motion vector detection of half pixel precision when the result of the determination means is false (NO). As a result, and if the result of the determination means is true (YES), a motion vector obtained by a half-pixel accuracy motion vector detection means is output as a result. Configured to have a means for.

In the present invention, the amount of motion vector detection is determined based on the quantization width and the code amount to be applied, and the motion vector detection is rounded up at the stage to reduce the load of the motion vector detection process. .

The quantization width is successively changed in order to adjust the generated code amount, and the image quality is closely related to the used quantization width. Therefore, the accuracy of motion vector detection is optimized in accordance with the quantization width and the code amount. Thus, the processing load can be appropriately reduced without deteriorating the image quality.

In order to achieve the above object, the present invention provides
Secondly, [2] a moving picture encoding means using motion compensation prediction and discrete cosine transform, a means for calculating block activity from a motion compensation prediction error signal, and a means for changing according to the internal capacity of an output buffer. Means for setting a threshold value, means for determining whether the block activity is greater than the threshold value, and determination means false (NO)
In the case of (1), the block is forcibly coded as an insignificant block to have a means for omitting processing such as discrete cosine transform.

According to this configuration, the prediction error signal is analyzed, and if the block to be processed is presumed to be an insignificant block, the block is forcibly changed to an insignificant block. As a result, the processing of DCT and quantization can be omitted without deteriorating the image quality, and the load can be reduced.

As described above, in the second specific example, unintended blocks, which are blocks that do not need to be processed, are not processed from the beginning, so that unnecessary processing is omitted, and image quality is degraded. Without DCT and quantization processing, the load can be reduced.

Further, according to the present invention, in order to achieve the above object,
Third, [3] moving picture decoding means for reproducing data coded by the moving picture coding means using motion compensation prediction and discrete cosine transform, wherein a post-filter is used to output a reproduced picture. , A means for separately turning on / off the post filter for the luminance signal and the color difference signal, and a first determination means for determining whether the quantization parameter is greater than a first threshold value. , The first determination means is false
In the case of (NO), the means for turning off the post-filter processing for both the luminance signal and the chrominance signal and the first determination means are true (YES).
In the case of, the second determination means for determining whether or not the quantization parameter is greater than the second threshold value, and if the result of the second determination means is true (YES), the luminance signal and the color difference The configuration includes means for performing post-filter processing on both signals and means for performing post-filter processing only on the luminance signal when the result of the determination means is false (NO).

The quantization width used for the quantization process is given by the quantization parameter. When the quantization parameter is small, that is, when the quantization width is small, the reference image is less affected by the coding distortion. Since it can be estimated that it has not been received, when the quantization parameter is small, the effect of the filter processing is weak.

Therefore, in the present invention, filtering is performed on an object that can be expected to have an effect in performing the filtering process. However, for an object with a small effect or an effect that cannot be expected, the filtering process is omitted to reduce the load due to the filtering process. In order to reduce the noise, it is determined whether or not the quantization parameter is larger than predetermined threshold values THp1 and THp2, and control of execution / non-execution of the filter process is performed according to the determination result. It is.

Thus, in the moving picture decoding apparatus,
ON / OFF of post-filter processing adaptively according to quantization width
By performing the turning off, the amount of post-filter processing can be reduced without deteriorating the image quality.

[0052]

Embodiments of the present invention will be described below with reference to the drawings.

<First Specific Example> A first specific example of the present invention will be described with reference to FIG. 1, FIG. 2, FIG. 3, and FIG.

In this example, the processing amount of the motion vector detection is reduced without deteriorating the image quality by adaptively omitting the motion vector detection with half-pixel accuracy.
Motion vector detection has integer pixel accuracy and half-pixel accuracy, but the latter is more accurate than the former. The effect of the half-pixel accuracy motion compensation obtained by the half-pixel accuracy motion vector detection is not only the effect of generating a motion compensation predicted value at the half-pixel position, but also the effect of the reference image signal stored in the frame memory. When the signal is deteriorated due to the coding distortion, the effect of reducing the influence of the coding distortion on the prediction signal by performing the filtering process is great.

On the other hand, when the applied quantization parameter Qp is small (the quantization width is small), it can be estimated that the reference image is not significantly affected by the coding distortion.
When the quantization parameter Qp is small, the effect of reducing the coding distortion is hardly obtained.

Therefore, in this case, the precision of the motion vector detection is set to an appropriate precision corresponding to the quantization width so as not to be over-specified (excessive quality), so that the processing load is insignificantly increased. An embodiment will be described in which the processing load is reduced and the image quality is prevented from deteriorating by setting the accuracy to an appropriate level.

(First Specific Example 1) In the present invention, the extent to which motion vector detection is to be performed is determined from the quantization width and the code amount to be applied, and the motion vector detection is rounded up at the stage, so that the motion vector detection is performed. Reduce the processing load.

The simplest concrete example will be described. In this specific example, a motion vector detection process (4MV
This is an example in which the detection processing is also omitted together with the half-pixel precision motion vector detection processing, and thus the minimum processing is required.

The moving picture coding apparatus according to the present invention is shown in FIG.
, The difference value calculation unit 101, the motion compensation prediction unit (MC) 102, the intra-frame / inter-frame (Intr
a / Inter) switching unit 103, frame memory (FM) 104, motion vector detection unit (ME) 105
A, discrete cosine transform unit (DCT) 106, quantization unit (Q) 107, variable length coding unit (VLC) 108, inverse quantization unit (IQ) 109, inverse discrete cosine transform unit (IDC)
T) 110, an adder 111, an output buffer (Buffe)
r) 112 and a rate control unit (Rate Control) 113.

The difference value calculation unit 101 subtracts the motion prediction signal output from the motion compensation prediction unit (MC) 102 from the video signal to be coded,
The motion vector detection unit (ME) 105A detects the motion of each block of the image from the signal of the moving image to be encoded, and outputs the obtained difference signal (motion compensation prediction error signal). The frame memory 104 has a function of obtaining a vector and determining switching of the Intra / Inter mode from the motion vector.
Is for accumulating the encoded frame signal in frame units.

The motion compensation prediction unit (MC) 102
Based on the signal on the frame memory 104 and the motion vector from the motion vector detection unit 105, the motion of each block of the moving image to be encoded is predicted and output as a motion compensated prediction signal. .

Intra-frame / inter-frame (Intra / I
nter) The switching unit 103 is switched based on the determination result of the Intra / Inter mode switching output from the motion vector detection unit (ME) 105, and a motion compensation prediction unit (MC) corresponding to the determination result. 102
Is supplied to the motion compensation prediction signal 101.

The discrete cosine transform unit (DCT) 10
Numeral 6 denotes a discrete cosine transform of the difference value (motion compensation prediction error signal) output from the difference value calculation unit 101. The transform coefficient obtained by the discrete cosine transform is quantized by the quantization unit (Q) 107. The output is quantized. Note that the discrete cosine transform is used as a standard in MPEG, but this is one type of orthogonal transform for decomposing an image into spatial frequency components, and various other orthogonal transform methods are known. Is well known, and it goes without saying that other orthogonal transform means may be used as necessary, for example, when applied to other systems.

A variable length coding unit (VLC) 108 performs variable length coding on the quantized output, and an output buffer (Buffer) 112 buffers the variable length coded signal to output a signal. 15 to send it out.

An inverse quantization unit (IQ) 109 is for returning the quantized output of the quantization unit 107 to the original discrete cosine transform coefficient after the inverse quantization, and is performed by an inverse discrete cosine transform unit (IDCT). ) 110 generates a reproduced signal by performing an inverse discrete cosine transform on the output of the inverse quantization unit 109.

Further, the adding section 111 includes the motion compensation prediction section 1
02 and the reproduction signal output of the inverse discrete cosine transform unit 110 are added to be stored in the frame memory (FM) 104.

A rate control unit (Rate Control) 11
3 determines the quantization parameter Qp corresponding to the bit stream storage amount of this buffer using the information of the bit stream storage amount provided from the output buffer 112, and provides the quantization parameter Qp to the quantization unit 107 and the inverse quantization unit 109. It has a function of adjusting the quantization width.

The MC + DCT type moving picture coding apparatus having the above-described configuration uses the video signal input line 11 to
As shown in (a), an image signal converted into a macroblock (MB) is supplied.

In the example shown in FIG. 9, the MB includes four blocks each composed of 8 × 8 pixels, and the luminance signal (Y) includes four blocks and the color difference signal (U or V) includes one block. The difference value calculation unit 101 calculates a difference between the image signal supplied via the image signal input line 11 and the motion compensation prediction signal supplied via the signal line 12, and this difference signal is transmitted through the signal line 13. The signal is supplied to the discrete cosine transform unit 106 via

The signal line 12 has an intra-frame encoding (In
ter) mode, the motion compensation prediction unit (MC) 1
02 is supplied via the intra / inter-frame switching unit 103, but is not supplied in the case of the inter-frame coding (Intra) mode.

That is, the signal of the image signal input line 11 is supplied to the signal line 13 as it is, not the difference signal.
Switching between the Intra / Inter modes is determined by the motion vector detection unit 105, and the determination result is supplied to the intra-frame / inter-frame switching unit 103 via the signal line 14 as a switching signal. This is performed by the intra-frame / inter-frame switching unit 103 switching the output of the motion compensation prediction unit 102.

The motion compensation prediction unit (MC) 102 generates a motion compensation prediction signal from the reproduced image signal of the previous frame stored in the frame memory 104 according to the motion vector information detected by the motion vector detection unit 105A.

In the discrete cosine transform unit (DCT) 106, the difference value calculating unit 101 supplied via the signal line 13
Is subjected to discrete cosine transform, and the obtained coefficient is supplied to the quantization unit 107. The quantization unit 107 quantizes the coefficient (DCT coefficient) obtained by the discrete cosine transform with a quantization width determined by a quantization parameter Qp given from the rate control unit 113, and performs a variable length coding unit 108
And to the inverse quantization unit (IQ) 109 as well.

The variable-length coding unit 108 performs variable-length coding on the supplied quantized discrete cosine transform coefficients, and sends the resultant to the buffer 112 as a compressed moving image signal.
Specifically, the variable-length encoding unit 108 encodes discrete cosine transform coefficients and motion vector information quantized by the quantization unit (Q) 107, and multiplexes to generate a bit stream. The data is supplied to the buffer 112.

Then, the output buffer 112 outputs to the network or the storage medium via the signal line 15 as a bit stream according to the characteristics.

Although the multiplexing unit is not shown, in actuality, the moving image encoding unit quantizes the DCT coefficients and uses the moving image data compressed by variable-length coding. It is multiplexed with the information of the quantization width (specifically, the quantization parameter Qp) and output.

On the other hand, the inverse quantizer (IQ) 109 supplies the supplied quantized discrete cosine transform transform coefficient (DCT).
) Is inversely quantized with a quantization width determined by a quantization parameter Qp given from the rate control unit 113, and supplied to an inverse discrete cosine transform unit (IDCT) 110 to restore the original image signal. Then, the reproduced image signal is supplied to the adder 111.

The addition section 111 adds the restored image signal supplied from the inverse discrete cosine transform section 110 and the motion compensation prediction signal from the motion compensation prediction section 102 supplied via the signal line 12 to generate an image signal. Is reproduced and transmitted to the frame memory 104 to be stored in the frame memory 104.

From the output buffer 112, the rate control unit 1
13 is provided with information on the bitstream storage amount via the signal line 16, and the rate control unit 113 determines the quantization parameter Qp according to the bitstream storage amount in this buffer. Then, the determined quantization parameter Qp is supplied to the quantization unit 107 and the inverse quantization unit 109 via the signal line 17.

Here, when the amount of accumulation in the buffer increases, the rate control unit 113 increases the quantization parameter Qp to reduce the amount of generated codes, and the amount of accumulation in the buffer decreases. In such a case, the quantization parameter Qp is reduced so that the generated code amount is controlled to be constant.

That is, the quantization width changes according to the code accumulation amount in the buffer. If the quantization width is large (coarse), it does not make sense to reduce the accuracy of the motion vector, so it is meaningless. If the quantization width is small (fine), the accuracy of the motion vector should be increased to detect the motion vector. The processing load can be adjusted corresponding to the quantization width, and unnecessary processing can be eliminated.

For this purpose, in the image coding apparatus of the present invention, the motion vector detecting section 105A has a function of varying the motion vector detection processing accuracy in accordance with the quantization width. Specifically, the function is to determine which precision of the motion vector is to be used as the detection vector with reference to the quantization parameter Qp and the motion vector detection result from the rate control unit 113.

The accuracy of motion vector detection includes motion vector detection with integer value accuracy and motion vector detection with half-pixel accuracy, and the latter is finer than the former.

FIG. 2 is a flow chart showing the processing contents of the motion vector detecting section 105A, which is the first specific example 1 of the present invention. The flow is executed for the luminance signal of each MB (macro block). Is done. First, step S1
At 01, the initial value of the motion vector (MV) is set to a zero vector.

Next, in step S102, it is determined whether or not the absolute value sum (SAD0) of the prediction error signal at the zero vector is larger than a predetermined zero vector threshold value (TH0). If the result of the determination in step S102 is that SAD0 is smaller than the threshold value (TH0), the motion vector detection process ends, assuming that a zero vector has been detected.

On the other hand, as a result of the determination in step S102,
If SAD0 is greater than the zero vector threshold value (TH0), a motion vector (MVint) with integer pixel precision is detected in step S103, and
In step S104, the in-block activity (ACT) of the luminance signal of the MB is calculated. Here, the intra-block activity is, for example, the absolute value sum of the difference between the average value in the block and each pixel value in the block.

Next, in step S105, it is determined whether or not the absolute value sum (SADi) of the prediction error signal in the motion vector (MVint) with integer pixel precision is smaller than ACT. As a result of the determination in step S105, SADi
Is larger than ACT, Step S106
Then, the encoding mode of the MB is set to “Intra” (inter-frame encoding), and the motion vector detection processing ends.

Here, in step S106, the MB
Is not set to "Intra", the coding mode of the MB becomes "Inter" (intra-frame coding).

On the other hand, as a result of the determination in step S105,
If SADi is smaller than ACT, the process proceeds to step S107, where Qp is a predetermined integer pixel accuracy determination threshold for determining whether or not to perform motion vector detection with integer pixel accuracy. It is determined whether it is greater than THint.

Here, Qp is a quantization parameter (not shown) supplied from the rate control unit 113 to the motion vector detection unit 105 via the signal line 17, and is a parameter for determining the quantization width. For example, “1” to “3”
An integer value of 1 "is adopted.

That is, in this example, the quantization parameter Qp is divided into 31 levels, and the smaller the quantization parameter Qp, the smaller (fine) the quantization width.

As described above, the quantization parameter Qp
Is supplied from the rate control unit 113 based on the information of the bit stream storage amount in the output buffer 112 which plays a role of sending out the compression-encoded and multiplexed moving image signal. Then, when the bit stream storage amount of the output buffer 11 increases, the rate control unit 113 increases the quantization parameter Qp to reduce the generated code amount, and the storage amount in the output buffer 112 decreases. The quantization parameter Q
By making p smaller so that the amount of generated codes can be increased, control is performed so that the buffer does not overflow while the quantization width is kept as small as possible.

When the quantization width is large, the use of a half-pixel precision motion vector is insignificant. Therefore, in the present invention, in the case of detecting a motion vector, for those which cannot be expected to be effective, the half-pixel precision motion vector detection is omitted. And substitute the motion vector with the motion vector detection result with integer pixel accuracy,
When the effect can be expected, perform motion vector detection with half-pixel accuracy, and use the detection result as a motion vector, so as to determine how far the motion vector detection process is performed depending on the situation, The load spent on the motion vector detection processing is reduced according to the situation.

Therefore, in the present invention, the rate control unit 1
It is determined whether or not the quantization parameter Qp from 13 is larger than the threshold value THint. According to this determination result, a half-pixel precision motion vector is not detected, and a motion vector with integer pixel precision is used as a motion vector. Is to do so.

Therefore, if Qp is smaller than THint as a result of the determination in step S107, the motion vector (MVint) with integer pixel precision is adopted as the detected motion vector, and the motion vector detection is completed.
On the other hand, as a result of the determination in step S107, Qp becomes THin
If it is larger than t, a motion vector (MVhalf) with half-pixel accuracy is detected (step S109). Then, after that, a motion vector (MV
4mv) is detected (step S110).

Next, at S111, the absolute value sum (SAD4mv) of the prediction error signal at MV4mv is calculated as the sum of absolute values (SAD4) of the prediction error signal at the half-pixel precision motion vector (MVhalf).
ADhalf) is determined. If the result of determination in step S111 is that SAD4mv is greater than SADhalf, MVhalf is used as the detected MV (step S112), and the motion vector detection process ends.

On the other hand, as a result of the judgment in step S111,
If SAD4mv is smaller than SADhalf, MV4m
v is used as the detected MV (step S
113), and ends the motion vector detection processing.

As described above, an appropriate motion vector detection accuracy is selected according to the quantization width used for encoding, and the motion vector detection processing is performed according to the selected precision. Motion vector detection accuracy can be optimized, and the load of processing can be reduced to an appropriate level according to the situation, in combination with the effect of eliminating waste in DCT and quantization processing in the subsequent stage. .

The quantization width is determined by the rate control unit 113 on the basis of the information of the bit stream storage amount in the output buffer 112 which plays the role of sending out the compressed and coded moving picture signal. When the bit stream storage amount of the output buffer 11 increases, the rate control unit 113 increases the quantization parameter Qp to reduce the generated code amount, and the storage amount in the output buffer 112 decreases. If you have
By reducing the quantization parameter Qp so as to increase the amount of generated code, it is provided to control so that the buffer does not overflow while keeping the quantization width as small as possible, and the quantization width is large. At this time, since it is insignificant to use a half-pixel precision motion vector, in the present invention, in the case of detecting a motion vector, those which cannot be expected to have an effect are omitted from the half-pixel precision motion vector detection and the integer-valued pixel precision motion vector detection result. Substitute the motion vector with
When the effect can be expected, motion vector detection with half-pixel accuracy is performed, and the detection result is used as a motion vector, so that it is optimized for the quantization width, which may adversely affect the image quality. Therefore, the processing load can be appropriately reduced without deteriorating the image quality.

In the processing method of the first specific example 1 described above, in the motion vector detection processing, if the quantization width is within the applicable range of the integer pixel precision, the motion vector detection processing must be performed with the integer pixel precision. This is a configuration in which motion vector detection is applied, and the processing of 4MV detection is omitted. However, the effect of 4MV is large for an image sequence including a complicated motion. Problems remain with images of image sequences that include complex motion. Therefore, next, in the motion vector detection process, a specific example which leaves the possibility of applying the 4MV detection even when the quantization width is within the applicable range of the integer pixel precision is described below.
This will be described as a specific example 2.

(First Specific Example 2) FIG. 3 shows a motion vector detecting section as a second specific example 2 of the first embodiment of the present invention which also includes 4MV detection (detection of a motion vector for every 8 × 8 pixels). It is a process flowchart of 105A. In the flowchart of FIG. 2, when the quantization width is within the applicable range of the integer pixel precision (S107), the motion vector detection with the integer pixel precision is uniquely applied. Also in this case, the point (S308 to S311) in which the possibility of 4MV detection application is searched for is different.
Others are basically the same as the processing of the flowchart of FIG.

The specific contents of the processing will be described with reference to FIG.

Here, the processing according to the flow is executed for the luminance signal of each MB.

First, in step S301, the initial value of the motion vector (MV) is set to a zero vector. next,
In step S302, it is determined whether or not the absolute value sum (SAD0) of the prediction error signal at zero vector is larger than a threshold value (TH0). Then, step S30
If the result of determination in step 2 is that SAD0 is smaller than TH0, the process ends assuming that a zero vector has been detected.

On the other hand, as a result of the determination in step S302,
If SAD0 is greater than TH0, step S
At 303, a motion vector (MVint
) Is detected, and in step S304, the M
The intra-block activity (ACT) of the B luminance signal is calculated. Here, the intra-block activity is, for example, the absolute value sum of the difference between the average value in the block and each pixel value in the block.

Next, in step S305, it is determined whether or not the absolute value sum (SADint) of the prediction error signal at MVint is smaller than ACT.

As a result of the determination in step S305, S
If ADint is larger than ACT, the encoding mode of the MB is set to "Intra" in step S306, and the motion vector detection processing ends. Here, if the coding mode of the MB is not set to “Intra” in step S306, the coding mode of the MB is set to “Intra”.
ter ”.

On the other hand, as a result of the determination in step S305,
If SADint is smaller than ACT, it is determined in step S307 whether the value of Qp is larger than threshold value THint.

If the value of Qp is smaller than the threshold value THint as a result of the determination in step S307, a motion vector (MVint4mv) with integer pixel precision is detected from the 4MV detection in step S308. .

Next, in step S309, the absolute value sum (SADint) of the prediction error signal at MVint is calculated as MVint4
It is determined whether or not the absolute value of the prediction error signal in mv is larger than the sum of the absolute values (SADint4mv). Then, this step S30
If the result of determination in step 9 is that SADint is smaller than SADint4mv, MVint is used as the detected MV (step S310), and the motion vector detection process ends.

On the other hand, as a result of the determination in step S309,
If SADint is greater than SADint4mv, M
Vint4mv is used as the detected MV (step S311), and the motion vector detection processing ends.

As a result of the determination in step S307,
If Qp is greater than THint, a motion vector (MVhalf) with half-pixel accuracy is detected (step S3).
12) A motion vector (MV4mv) for each 8 × 8 pixel is detected (step S313). And then, MV4mv
Is the sum of the absolute values of the prediction error signals (SAD4mv) at MVha
It is determined whether or not it is greater than the absolute value sum (SADhalf) of the prediction error signal at lf (step S314).

As a result of the determination in step S314, the SAD
If 4mv is greater than SADhalf, MVha
lf is used as the detected MV (step S
315), the motion vector detection processing ends.

On the other hand, as a result of the determination in step S314,
If SAD4mv is smaller than SADhalf, MV
4mv is used as the detected MV (step S316), and the motion vector detection processing ends.

In the processing method of the first specific example, the processing function of the first specific example has the same function as that of the first specific example. , The motion vector detection with integer pixel accuracy is not applied unambiguously.
It is possible to provide a moving image encoding device that can perform MV detection processing and can optimize processing of an image sequence including a complicated motion without deteriorating image quality.

The above is a technique for reducing the processing without deteriorating the image by focusing on the detection of the motion vector and adaptively changing the extent to which the motion vector detection is performed.

Next, paying attention to macroblocks, among these, insignificant blocks, which do not need to be processed, are not processed from the beginning, and the waste of processing is eliminated, thereby deteriorating the image quality. In the second example, it is possible to reduce the load by making it possible to eliminate the waste of DCT and quantization processing without accompanying
Will be described as a specific example.

<Second Specific Example> Here, of the macro blocks, a significant block is a block in which all DCT coefficients (discrete cosine transform coefficients) are "0", and the other macro blocks are significant. Block.

In FIG. 2, when the quantized discrete cosine transform coefficients output from the quantization unit (Q) 107 to the inverse quantization unit (IQ) 109 via the signal line 18 are all “0” , All signals output via the signal line 19 without going through the inverse quantization unit 109 and the inverse discrete cosine transform unit (IDCT) 110 become “0”.

In such a case, information indicating that the block is insignificant is sent to the decoder side according to the pattern illustrated in FIG. 9B, so that both the encoder / decoder perform inverse quantization and inverse discrete cosine. No conversion processing is required. In other words, there is no need to process the inexplicable blocks.

In some cases, a block can be clearly determined as an insignificant block without performing discrete cosine transform and quantization. For example, when the motion compensated prediction signals match perfectly and the prediction error signal supplied via the signal line 13 becomes “0”, it is apparent that the discrete cosine transform processing and the quantization processing are not performed. It can be determined that the block is a meaningless block. Also,
When the prediction error signal is small and the quantization parameter is large, it can be determined that the block becomes a meaningless block. In such a case, it is useless to perform the discrete cosine transform process and the quantization process.

Thus, the insignificant block is a block that does not need to be processed. Therefore, the second specific example is to reduce the load by not processing the insignificant block from the beginning and eliminating waste, and FIG. 4, FIG. 5 and FIG. An example will be described with reference to FIG.

FIG. 4 is a block diagram showing the system configuration.

The moving picture coding apparatus according to the present invention has the structure shown in FIG.
, The difference value calculation unit 101, the motion compensation prediction unit (MC) 102, the intra-frame / inter-frame (Intr
a / Inter) switching unit 103, frame memory (FM) 104, motion vector detection unit (ME) 105,
Discrete cosine transform unit (DCT) 106, quantization unit (Q)
107, a variable length coding unit (VLC) 108, an inverse quantization unit (IQ) 109, an inverse discrete cosine transform unit (IDCT) 1
10, adder 111, output buffer (Buffer) 1
12, a rate control unit (Rate Control) 113, a significant / insignificant block determination unit 214, and switching units 215 and 216.

Of these, those having the same reference numerals and the same names as those in FIG. 1 are basically the same as those described in FIG. 1, and will not be described again here.

In this specific example, a significant / insignificant block determination unit 214 and switching units 215 and 216 are newly added to FIG. 1, and the motion vector detection unit (ME) is replaced with a conventional motion vector detection unit instead of 105A. (ME) 105 is provided.

In this example, a prediction error signal is analyzed, and a block which is presumed to be an insignificant block is forcibly set as an insignificant block in advance, so that DCT or quantization without deteriorating image quality is performed. The purpose of this is to omit the processing of. Therefore, the effect of reducing the processing can be enjoyed by omitting the processing for the insignificant block without using the technique for reducing the processing without deteriorating the image, but the effect of reducing the processing can be enjoyed. , The use of the motion vector detecting unit 105A, whereby the method of the first specific example that can adaptively change the extent to which the motion vector detection is performed, can be used to reduce the processing effect. It will be even bigger.

As shown in FIG. 4, the significant / insignificant block determining section 214 sends the difference value calculating section 101
, And the quantization parameter Qp provided from the rate control unit 113, and when the motion compensated prediction error signal supplied from the difference value calculation unit 101 is “0”, and When the motion compensation prediction error signal is small and the quantization parameter Qp is large, it is determined that the block is an insignificant block.

The switching section 216 is arranged between the inverse discrete cosine transform section (IDCT) 110 and the adding section 111. When the significant / insignificant block determination section 214 determines that the block is insignificant, the switching section 216 performs the inverse operation. Discrete cosine transform unit 110
Is a switch function unit that shuts off the supply of the output of (1) to the adding unit 111 and supplies the output if the block is a significant block.

In this apparatus having such a configuration, an image signal converted into a macro block (MB) as shown in FIG. 9A is supplied from the image signal input line 11. This is input to the motion vector detection unit 105 and the difference value calculation unit 101.

Then, in the difference value calculation unit 101,
The difference (motion compensation prediction error signal) between the supplied image signal and the motion compensation prediction signal supplied via the signal line 12 is calculated. Then, the difference signal (motion compensation prediction error signal) calculated by the difference value calculation unit 101 is supplied to the significant / insignificant block determination unit 214 and the switching unit 115 via the signal line 13.

[0132] The intra-frame encoding (In
ter) mode, the motion-compensated prediction signal generated by the motion-compensated prediction unit 102 is supplied via the intra-frame / inter-frame switching unit 103, and in the inter-frame coding (Intra) mode, No signal is supplied. In other words, the moving image signal from the image input line 11 is supplied to the signal line 13 as it is, instead of the difference signal (motion compensation prediction error signal) from the difference value calculation unit 101.

The switching between the Intra / Inter modes is determined by the motion vector detection unit 105, and the signal line 14
Is supplied to the intra-frame / inter-frame switching unit 103 via the. The motion-compensated prediction signal generated by the motion-compensated prediction unit 102 is generated from a reproduction signal of an already encoded image frame stored in the frame memory 104 according to the motion vector information detected by the motion vector detection unit 105. Is done.

In the significant / insignificant block determination section 214, the sum of absolute values (SAE: Sum of Absolute Er) of the motion compensation prediction error signal supplied via the signal line 13 is provided.
rr) is calculated as a block activity, and if the block activity is small, the block is forcibly determined to be an insignificant block.

The result of this determination is supplied to the switching units 215 and 216 via the signal line 30, respectively. If it is determined that the block is insignificant, no signal is supplied via the signal line 31, so that the discrete cosine transform unit 106, quantization unit 107, inverse quantization unit 109, inverse discrete cosine transform unit 11
0 is turned off and the signal line 19 is turned off.
The signal supplied to the switching unit 216 via the
Since it is not supplied to 1, the processing by the discrete cosine transform unit 106, the quantization unit 107, the inverse quantization unit 109, and the inverse discrete cosine transform unit 110 is not required.

If it is determined that the block is an insignificant block, the flag in FIG. 9B is set to "0" in this example.

The significant / insignificant block determining section 214 for determining whether the macroblock to be processed is a significant block or an insignificant block can be realized by executing the following processing procedure. It is.

Referring to FIG. 5, significant / insignificant block determination unit 2
The contents of the process of No. 14 will be specifically described. First, in step S401, the sum of absolute values (SAE) of motion compensation prediction error signals is calculated for each block of an image to be processed. Next, in step S402, it is determined whether the SAE obtained by this calculation is larger than the threshold value TH (B).

TH (B) is a function that is supplied to the rate control unit 113 and whose value changes according to the bit stream storage amount (B) of the output buffer 112, and is given by the rate control unit 113. . Then, the rate control unit 213 increases the value of TH (B) when the amount of accumulation in the buffer increases, and decreases the value of TH (B) when the amount of accumulation in the buffer decreases. When the output buffer 112 is about to overflow due to the adjustment control, the number of insignificant blocks is increased and the generated code amount is suppressed.

As a result of the determination in step S402, the SAE
Is smaller than TH (B), Step S4
In 03, the block is forcibly changed to an insignificant block,
The process ends.

On the other hand, as a result of the determination in step S402,
If the SAE is larger than TH (B), the process ends. Here, an example in which the significance / insignificance determination is performed for each block has been described, but the significance / insignificance determination may be performed for each macroblock.

In the discrete cosine transform unit 106, the signal line 3
After the signal supplied via the signal 1 is subjected to the discrete cosine transform, the signal is supplied to the quantization unit 107 and is quantized there.
The discrete cosine transform coefficients quantized by the quantization unit 107 are supplied to the variable-length encoding unit 108, are subjected to variable-length encoding, and are supplied to the inverse quantization unit 109 and are inversely quantized.

The inversely quantized transform coefficients are supplied to an inverse discrete cosine transform unit 110, where a reproduced signal for the signal line 23 is generated and supplied to an adder unit 111. Adder 11
In step 1, the signal supplied from the inverse discrete cosine transform unit 110 and the signal supplied via the signal line 12 are added to reproduce an image signal, and then stored in the frame memory 104.

In the variable length coding unit 108, the quantization unit 10
After encoding the discrete cosine transform coefficients and motion vector information (not shown) quantized in step 7, a bit stream is generated by multiplexing and supplied to the output buffer 112. The output buffer 112 outputs a billet stream to the network or the storage medium via the signal line 25 according to the characteristics.

The rate control unit 113 determines the quantization parameter Qp according to the bit stream storage amount of the output buffer 112 supplied via the signal line 26, and inversely determines the quantization parameter Qp via the signal line 27. This is supplied to the quantization unit 109. Here, when the accumulation amount in the buffer increases, the quantization parameter Qp is increased to reduce the generated code amount, and when the accumulation amount in the output buffer 112 decreases, the quantization parameter By reducing the size, control is performed so that the generated code amount becomes constant.

The multiplexing unit is not shown, but actually,
The moving image coding unit quantizes the DCT coefficients and compresses the moving image data by performing variable length coding on the DCT coefficients. The moving image data is multiplexed with information on the used quantization width (specifically, the quantization parameter Qp). Is output.

As described above, according to the second specific example, the prediction error signal is analyzed, and when the block to be processed is presumed to be an insignificant block, the block is forcibly set as an insignificant block. I did it. As a result, the processing of DCT and quantization can be omitted without deteriorating the image quality, and the load can be reduced.

The above are all the embodiments on the moving picture coding apparatus side. Next, a description will be given as a third specific example of an embodiment on the moving picture decoding device side capable of reducing the processing load.

<Third Specific Example> The quantization width applied at the time of encoding is changed in accordance with the situation in order to adjust the generated code amount. In the decoding apparatus, post-filter processing is performed to remove block distortion. However, when the quantization width is small, there is little concern about the influence of block distortion, and when the quantization width is coarse, the post-filter processing is performed. The meaning of the processing comes out.

Therefore, in this case, the filter processing is applied to an object which can be expected to have an effect in performing the filter processing. However, the object having a small effect or the one in which the effect cannot be expected is omitted so that the filtering processing is omitted. An example will be described in which the number is reduced.

A third embodiment of the present invention will be described with reference to FIGS.

This specific example relates to a technique relating to a moving picture decoding apparatus, in which the post-filter processing is adaptively turned on / off at the time of decoding, so that the processing amount of the post-filter processing can be reduced without deteriorating the image quality. It is to reduce. Specifically, on / off control of the post-filter processing is performed according to the quantization parameter Qp given to the inverse quantization unit.

FIG. 6 is a block diagram of a moving picture decoding apparatus for explaining a third specific example of the present invention. In the drawing, reference numeral 401 denotes a variable length code decoding unit for decoding received compressed moving picture data. , 402 are inverse quantization units that perform inverse quantization of the data decoded by the variable-length code decoding unit 401 and restore the original discrete cosine transform coefficients (DCT coefficients).

An inverse discrete cosine transform unit 403 performs an inverse discrete cosine transform of the DCT coefficient obtained by the inverse quantization unit 402 to return to the original motion compensation prediction error signal. A restoring unit for restoring a frame image by adding the restored motion compensation prediction error signal and the motion compensation prediction signal from the motion compensation prediction unit 406. Reference numeral 405 denotes a frame memory. This is a memory for holding the frame image restored by addition processing at 404.

The motion compensation prediction unit 406 performs motion compensation prediction from a frame image in the frame memory 405 to obtain a motion compensation prediction signal. Reference numeral 407 denotes intra-frame coding (Inter) / inter-frame coding. (Intra) switching section for turning on / off the supply of the motion compensation prediction signal of the motion compensation prediction section 406 to the addition section 404. Since the mode of the intra-frame coding (Inter) or the mode of the inter-frame coding (Intra) is determined by multiplexing the compressed image data during processing on the moving picture coding apparatus side, the The coding / inter-frame coding switching unit 407 performs the above-mentioned on / off control corresponding to the mode. Reference numeral 408 denotes a post-filter unit that performs post-filter processing on the motion-compensated prediction signal restored by the addition processing in the addition unit 404, and outputs moving image data without distortion.

The moving picture coding section quantizes the DCT coefficients and compresses the moving picture data by performing variable length coding to obtain the information of the used quantization width (specifically, the quantization parameter Qp). And multiplexed with the output.

Next, the operation of the present apparatus having such a configuration will be described.

The moving picture decoding apparatus separates the data into quantization width information and compressed moving picture data and processes them. The compressed moving image data is sent to the variable length decoding unit 401 via the signal line 41.
, And the information of the quantization width is supplied to the inverse quantization unit 402 and the post-filter unit 408 via the signal line 42.

In the variable length code decoding section 401, the signal line 41
The decoding process is performed on the variable-length code of the DCT coefficient supplied via the DCT coefficient to return it to the DCT coefficient, and the DCT coefficient is supplied to the inverse quantization unit 402.
The inverse quantization unit 402 inversely quantizes the DCT coefficient according to the quantization parameter Qp supplied via the signal line 42.

DCT inversely quantized by inverse quantization section 402
The coefficients are subjected to inverse DCT in the inverse discrete cosine transform unit 403 and are reproduced as a motion compensation prediction error signal. The reproduced motion-compensated prediction error signal is supplied to the adder 404, and is added to the motion-compensated predicted signal supplied via the signal line 43 to become a reproduced signal. The reproduced signal is supplied to the frame memory 405 and the post filter unit 408 via the signal line 44.

Here, in the case of the intra-frame encoding (Inter) mode, the motion compensation prediction unit 4
The motion compensated prediction signal generated in 06 is supplied via the intra / interframe switching unit 407, and no signal is supplied in the case of the interframe coding (Intra) mode.

The post-filter unit 408 has an MC + DCT
This is to provide a user with a good-looking reproduced image by reducing block distortion caused by an encoding method. Here, when the quantization parameter Qp is small, the visual distortion of the reproduced image is not so large because the block distortion is small.

Therefore, even if the filter processing is performed by the post-filter unit 408 in such a case, the appearance is not particularly improved, but rather, the effect of the blur due to the filter processing is greater, so that the filter processing has the opposite effect. In some cases.

Since the color difference signal (U, V) has lower sensitivity than the luminance signal (Y), the influence of block distortion is smaller for the color difference signal than for the luminance signal. Therefore, even if the post-filter processing is performed on the color difference signal similarly to the luminance signal, the appearance is hardly improved and the processing amount is wasted.

Therefore, in this specific example, a function of performing or not performing post-filter processing according to the value of the quantization parameter Qp is provided by focusing on this point. That is, in order to secure this function, in this specific example, the post filter unit 408 is a software filter configured to be processed in processing steps as shown in FIG.

The processing steps in FIG. 7 will be described.

First, in step S501, Qp
Is larger than a first threshold value THp1. Then, as a result of the determination in step S501, Qp
Is smaller than THp1, the process proceeds to step S502, where the reproduced image supplied from the signal line 44 is output to the signal line 45 without being subjected to post-filter processing, and is provided to the user.

If Qp is greater than THp1 as a result of the determination in step S501, the process proceeds to step S503, where Qp is compared with a second threshold THp2, whereby Qp is equal to the second threshold. It is determined whether the value is greater than the value THp2.

If Qp is smaller than THp2 as a result of the determination in step S503, the process proceeds to step S503.
Moving to the process of 504, post-filter processing is performed only on the luminance signal (Y) of the reproduced image supplied from the signal line 44, and is output to the signal line 45 to be a final reproduced image signal.
Provide to users.

If the result of determination in step S 503 is that Qp is greater than THp 2, the processing moves to step S 505, where post-filter processing is performed on the reproduced image supplied from the signal line 44, and Output and provide to user.

Here, the relationship between the threshold values THp1 and THp2 is usually "THp1 <THp2". Note that "TH
By setting p1 = THp2 ″, only the flow of step S501 and step S502 is performed,
Only the flow of step 504 and step S505 may be performed.

Here, Qp is a quantization parameter supplied from the rate control unit 113 to the motion vector detection unit 105A via the signal line 17, and is a parameter for determining the quantization width. An integer value of 1 "to" 31 "is adopted.

That is, in this example, the quantization parameter Qp is classified into 31 levels, and the smaller the value of the quantization parameter Qp, the smaller (fine) the quantization width. And the quantization parameter Q
When the value of p is small, that is, when the quantization width is small, it can be estimated that the reference image is not significantly affected by the coding distortion. Therefore, when the quantization parameter Qp is small, the effect of the filtering process is reduced. Is thin.

Therefore, in the present invention, filtering is performed on an object that can be expected to have an effect upon performing the filtering process. However, for an object with a small effect or an effect that cannot be expected, the filtering process is omitted to reduce the load of the filtering process. In order to reduce this, it is determined whether or not Qp is larger than the threshold values THp1 and THp2, and the execution / non-execution of the filter processing is controlled according to the result of this determination.

As described above, in this specific example, the moving picture decoding apparatus adaptively turns on / off the post-filter processing according to the quantization width, so that the picture quality does not deteriorate. This makes it possible to reduce the amount of post-filter processing.

Although various specific examples have been described above, the present invention is not limited to the above examples, and can be implemented with various modifications.

[0177]

As described in detail above, according to the present invention,
In a hybrid coding scheme using motion compensated prediction (MC) and discrete cosine transform (DCT), adaptively omitting motion vector detection with half-pixel accuracy, the coding process can be performed without significant image quality degradation. Higher speed can be achieved.

Further, by analyzing the prediction error signal and forcibly setting a block which is presumed to be an insignificant block in advance as an insignificant block, high-speed encoding can be performed without significant deterioration in image quality. Can be achieved.

Further, by performing the on / off operation of the post-filter processing adaptively, the speed of the post-filter processing can be increased without significantly deteriorating the image quality.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of a moving picture encoding device according to the present invention.

FIG. 2 is a flowchart for explaining an operation of detecting a motion vector in the first specific example of the present invention.

FIG. 3 is a second flowchart for explaining an operation of motion vector detection in the first specific example of the present invention.

FIG. 4 is a diagram illustrating a second specific example of the present invention.

FIG. 5 is a flowchart illustrating an operation of a significant / insignificant block determination unit.

FIG. 6 is a diagram illustrating a third specific example of the present invention.

FIG. 7 is a flowchart illustrating an operation of a post filter unit according to a third specific example of the present invention.

FIG. 8 is an encoder block diagram of a verification model.

FIG. 9 is a diagram illustrating a macroblock.

FIG. 10 is a flowchart illustrating an operation of detecting a motion vector in a verification model.

FIG. 11 shows a motion vector in units of 16 × 16 pixels and 8 × 8 pixels.
FIG. 3 is a diagram illustrating a motion vector in pixel units.

FIG. 12 is a diagram illustrating half-pixel accuracy motion compensation.

[Explanation of symbols]

101: difference value calculation unit 102, 406: motion compensation prediction unit (MC) 103: intra-frame / inter-frame (Intra / Int)
er) switching sections 104, 405 frame memory (FM) 105 motion vector detecting section (ME) 106 discrete cosine transform section (DCT) 107 quantizing section (Q) 108 variable length coding section (VLC) 109 , 402: inverse quantization unit (IQ) 110, 403: inverse discrete cosine transform unit (IDCT) 111, 404: addition unit 112: output buffer (Buffer) 113: rate control unit (Rate Control) 214: significant / insignificant block Determination unit 215, 216, 407 ... Switching unit 408 ... Post filter unit.

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5C059 KK03 KK15 MA00 MA04 MA05 MA23 MC11 ME01 NN01 NN15 PP04 TA41 TA46 TA62 TA68 TB07 TB08 TC10 TC18 TD12 UA02 UA05 UA11 UA32 UA38 5J064 AA03 BA09 BA13 BA16 BC14 BC22

Claims (15)

[Claims] 1. A motion compensation prediction is performed from a motion vector obtained from an input image and an image obtained by reproducing encoded image data. It is used for performing the encoding process, sequentially processes the input image in combination with the inter-frame encoding process that does not use the motion compensation prediction signal, quantizes the transform coefficients obtained by orthogonally transforming the input image, and performs variable-length encoding. By doing so, in the moving picture coding apparatus adapted to perform compression coding, the means for detecting the motion vector is at least 2N ×
It has means for detecting a motion vector with integer pixel accuracy and means for detecting a motion vector with half pixel accuracy for each block of 2N pixels (N: natural number), and determines a quantization width used for quantization processing of the transform coefficient. A determination unit that performs quantization processing with a quantization width finer than a predetermined level, and performs motion vector detection with integer pixel precision motion vector detection without performing half pixel precision motion vector detection. Is a motion vector, and when the determination unit determines that the quantization process is performed using a coarser quantization width, a motion vector obtained by detecting a motion vector with half-pixel accuracy is output as a motion vector. Moving image encoding apparatus. 2. A motion compensation prediction is performed based on a motion vector obtained from an input image and an image obtained by reproducing encoded image data. It is used for performing the encoding process, sequentially processes the input image in combination with the inter-frame encoding process that does not use the motion compensation prediction signal, quantizes the transform coefficients obtained by orthogonally transforming the input image, and performs variable-length encoding. By doing so, in the moving picture coding apparatus adapted to perform compression coding, the means for detecting the motion vector is at least 2N ×
Means for detecting a motion vector with integer pixel precision for each block of 2N pixels (N: natural number), means for detecting a motion vector with integer pixel precision and half-pixel precision for each block of N × N pixels, and quantum of the transform coefficient Determination means for determining the quantization width used in the quantization process and when the determination means determines that the quantization process is a quantization width coarser than a predetermined level,
A motion vector obtained by detecting a motion vector with half-pixel accuracy is output as a motion vector, and when the determination unit determines that the quantization process is performed with a quantization width finer than a predetermined level, an integer pixel accuracy of 2N × 2N pixels is used. The magnitude of the error evaluation value for the motion vector obtained by the motion vector detection means and the magnitude of the error evaluation value for the motion vector obtained by the integer pixel precision motion vector detection for every N × N pixels are compared. When the former error evaluation value is smaller, a motion vector obtained by detecting an integer pixel precision motion vector for every 2N × 2N pixels is output as a motion vector, and as a result of the comparison, the former error evaluation value is larger. In this case, a motion vector obtained by detecting an integer pixel precision motion vector for each N × N pixel is output as a motion vector. That means the moving picture encoding apparatus characterized by a configuration with. 3. A motion compensation prediction is performed from a motion vector obtained from an input image and an image obtained by reproducing encoded image data. The motion compensation prediction error signal, which is the difference between the two, is used for performing the intra-frame encoding process, and the input image is sequentially processed in combination with the inter-frame encoding process using only the input image, and the orthogonal transform is performed. In the moving picture encoding apparatus which performs compression encoding by quantizing the transform coefficients obtained in the above manner and performs variable length encoding, an output buffer that stores the compression encoded data and sequentially outputs the data, Means for calculating a block activity which is a sum of absolute values from the compensation prediction error signal; means for setting a threshold value corresponding to the internal capacity of the output buffer; Means for determining whether the activity is greater than the threshold value. If the determination means determines that the block activity is less than the threshold value, the block is forcibly reset to zero. A means for coding as an insignificant block. 4. A moving picture coding apparatus according to claim 3, further comprising means for changing a quantization parameter for determining a quantization width to be applied to a quantization process in correspondence with a remaining capacity of said output buffer. A moving image encoding apparatus, wherein the threshold value setting means sets a threshold value according to a value of a quantization parameter. 5. A moving picture decoding means for decoding and reproducing data obtained by coding a moving picture by a moving picture coding method using motion compensation prediction and discrete cosine transform, and comprising: On the other hand, filter means for performing post-filter processing, and means for controlling execution of post-filter processing by the filter means on the decoded image signal according to a quantization width used in quantization processing at the time of encoding, An image decoding device comprising: 6. The moving picture decoding means according to claim 5, wherein said means for controlling the execution of said post-filter processing comprises: said post-filter processing when the quantization width used in the quantization processing at the time of encoding is small. An image decoding apparatus characterized by performing control so as not to perform image decoding. 7. The moving picture decoding means according to claim 5, wherein said filter means separates said post-filter into a luminance signal and a chrominance signal, and said means for controlling the execution of said post-filter processing includes an encoding process. When the used quantization width in the quantization process at the time is coarser than a predetermined level, control is performed so that both the luminance signal and the chrominance signal are post-filtered, and when the used quantization width is smaller than the predetermined level, only the luminance signal is used. A moving picture decoding apparatus, wherein the moving picture decoding apparatus is controlled to perform post-filter processing. 8. A moving picture decoding means for decoding and reproducing data obtained by coding a moving picture by a moving picture coding method using motion compensated prediction and discrete cosine transform, comprising the steps of: Filter means for performing post-filter processing, the filter means for performing post-filter processing on the decoded image signal, and the decoded image signal according to a quantization width used in quantization processing at the time of encoding. Controlling the execution of the post-filter processing by the filter means for the,
When the used quantization width in the quantization process at the time of encoding is smaller than a predetermined first level, control is performed so that the post-filter process is not performed, and the used quantization width in the quantization process at the time of encoding is Means for controlling to post-filter both the luminance signal and the color-difference signal when coarser than the predetermined second level when coarser than the predetermined first level; , A moving image decoding device. 9. A motion compensation prediction is performed based on a motion vector obtained from an input image and an image obtained by reproducing encoded image data. It is used for performing the encoding process, sequentially processes the input image in combination with the inter-frame encoding process that does not use the motion compensation prediction signal, quantizes the transform coefficients obtained by orthogonally transforming the input image, and performs variable-length encoding. In the moving picture coding method adapted to perform compression coding, the motion vector may be detected by detecting a motion vector with integer pixel precision for each block of at least 2N × 2N pixels (N: natural number). And a detection method for detecting a motion vector with half-pixel accuracy, and determines a quantization width used for the quantization process of the transform coefficient, and the determination result is a quantization width finer than a predetermined level. If it is determined that the quantization process according to, the motion vector obtained by the motion vector detection of integer pixel accuracy without performing the motion vector detection of half pixel accuracy is used as a detected motion vector, the determination result is the A moving picture encoding method characterized by using a motion vector obtained by detecting a motion vector with half-pixel accuracy as a motion vector when it is determined that the quantization processing is performed using a quantization width coarser than a predetermined level. 10. A motion compensation prediction is performed from a motion vector obtained from an input image and an image obtained by reproducing encoded image data. It is used for performing the encoding process, sequentially processes the input image in combination with the inter-frame encoding process that does not use the motion compensation prediction signal, quantizes the transform coefficients obtained by orthogonally transforming the input image, and performs variable-length encoding. In the moving picture coding method adapted to perform compression coding, the motion vector may be detected by detecting a motion vector with integer pixel precision for each block of at least 2N × 2N pixels (N: natural number). Using a detection method of No. 1 and a second detection method of detecting a motion vector with integer pixel precision and half pixel precision for each block of N × N pixels, a quantization width used for quantization processing of the transform coefficient is determined. If the determination result is a quantization process with a quantization width coarser than a predetermined level, a motion vector obtained by detecting a motion vector with half-pixel accuracy is adopted as a motion vector, and the determination result is higher than a predetermined level. When it is determined that the quantization process is performed by a fine quantization width, the evaluation value of the error with respect to the motion vector obtained by the first detection method and the motion vector obtained by detecting the integer pixel precision motion vector for each N × N pixel are calculated. The magnitude is compared with the error evaluation value, and as a result of the comparison, when the former error evaluation value is smaller, 2N ×
A motion vector obtained by detecting an integer pixel precision motion vector for every 2N pixels is adopted as a motion vector, and as a result of the comparison, if the former error evaluation value is larger, N ×
A moving image encoding method, wherein a motion vector obtained by detecting an integer pixel precision motion vector for every N pixels is adopted as a motion vector. 11. A motion compensation prediction is performed from a motion vector obtained from an input image and an image obtained by reproducing encoded image data. The motion compensation prediction error signal, which is the difference between the two, is used for performing the intra-frame encoding process, and the input image is sequentially processed in combination with the inter-frame encoding process using only the input image, and the orthogonal transform is performed. In the moving picture coding method in which the obtained transform coefficients are quantized and subjected to variable-length coding to perform compression coding, the compression-coded data is stored in a buffer and sequentially output, and A threshold value is set corresponding to the capacity of the buffer, and a block activity, which is a sum of absolute values thereof, is calculated from the motion compensation prediction error signal, and the block activity is calculated. Determining whether the activity is greater than the threshold, and if the block activity is determined to be less than the threshold, forcibly encoding the block as an insignificant block with zero content A moving picture coding method characterized by the following. 12. A moving picture decoding method for decoding and reproducing data obtained by coding a moving picture by a moving picture coding method using motion compensated prediction and discrete cosine transform. An image decoding method, comprising: controlling the execution of a post-filter process on the decoded image signal according to a quantization width used in the process. 13. A moving picture decoding method according to claim 12, wherein said post-filter processing is not performed when a quantization width used in quantization processing during encoding is small. Decryption method. 14. A moving picture decoding method according to claim 12, wherein said post-filter processing is performed for each of a luminance signal and a chrominance signal, and a quantization width used in a quantization processing at the time of encoding is smaller than a predetermined level. A moving picture decoding method comprising controlling the luminance signal and the chrominance signal to be post-filtered when coarse, and performing post-filtering only on the luminance signal when the used quantization width is smaller than a predetermined level. 15. A moving picture decoding method for decoding and reproducing data obtained by coding a moving picture by a moving picture coding method using motion compensated prediction and discrete cosine transform, comprising the steps of: On the other hand, a filter means for performing post-filter processing for each of the luminance signal and the chrominance signal is provided, and when the quantization width used in the quantization processing at the time of encoding is smaller than a predetermined first level, control is performed so that the post-filter processing is not performed. When the used quantization width in the quantization process at the time of encoding is coarser than the predetermined first level, the predetermined second level is used.
A moving picture decoding method, characterized in that when the level is coarse compared to the level, control is performed so that both the luminance signal and the color difference signal are post-filtered, and in the other cases, control is performed so that only the luminance signal is post-filtered.