JP2003230149A - High compression coding unit - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve a coding efficiency and to improve an image quality by reducing an amount of motion vector coding without deteriorating an image quality. <P>SOLUTION: The image quality is improved by introducing a cost function D+IR in a mode decision process of a 'frame compensation + VCL process' when a motion vector detected by a motion compensation predicting unit 1 is variable length coded (VLC) by a variable length coding unit 6, and a mode decision process of an 'interframe coding process' or an 'in-frame coding process' in a coding mode deciding unit 3; deciding a coding parameter of a motion vector, a coding mode, a coding unit or the like to be minimums; and controlling a quantizing step size in a quantizing unit 5. Further, scanning order selection at a DCT coefficient quantizing time in which a power distribution of a DCT coefficient in a DCT unit 4 is considered or a quantization step size control for realizing time change smoothing of an amount of generated information in the unit 5 is conducted. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high compression encoding apparatus for obtaining a motion vector by motion compensation prediction and encoding a moving image.

[0002]

2. Description of the Related Art An encoding method (Test Model Version 5, abbreviated as TM) which is a reference model created by the standardization work of MPEG-2
In 5), when detecting a motion vector, the motion vector is determined by a simple minimum distortion condition based on the sum of absolute difference, and when the DCT coefficient is quantized, only a so-called zigzag scan is performed to scan intra / macroblock. When determining the coding mode of the inter mode, the coding mode was selected with the coding distortion as the evaluation scale.

[0003]

However, in the motion vector determination method based on the sum of absolute difference, since the code length of the motion vector is almost proportional to the difference with the vector of the adjacent macroblock, the motion unrelated to the original motion. There is a case where the motion search accuracy is low such that a vector is selected, and the coding efficiency is deteriorated. In particular, the motion vector prediction model involves parallel movement in units of macroblocks, so in images that do not conform to this model, the motion vector calculated from time to time is independent of the physical motion, and It sometimes points to a position far away from, which causes a decrease in coding efficiency.

Further, for example, the DCT coefficient power distribution in an HDTV signal or the like statistically indicates that the horizontal pixel-to-pixel correlation is stronger than the vertical pixel-to-pixel correlation, so that in the DCT coefficient region. The vertical high-frequency coefficient statistically has higher power than the horizontal high-frequency coefficient, but the method of scanning the DCT coefficient weighted horizontally or vertically based on the inter-pixel correlation is not used. The conversion efficiency was not good.

Further, when the intra / inter mode coding mode is selected with only the coding distortion as an evaluation measure for each macroblock, the coding efficiency is actually better when the coding mode is reversed. May be
In this respect as well, the coding efficiency was reduced.

Therefore, the present invention aims to solve the problems by reducing the motion vector code amount that does not cause image quality deterioration, optimizing coefficient scan according to video, optimizing intra / inter mode, and the like. Thus, it is an object of the present invention to provide a high compression encoding device capable of improving encoding efficiency and image quality.

[0007]

In order to solve the above-mentioned problems, according to the present invention, there is provided a high compression coding apparatus for coding a video signal of a moving image, which control can be dynamically determined in the coding process. When determining an item, that is, a control item such as adoption or non-use of inter-frame prediction or determination of a motion vector in motion-compensated prediction, mode selection from a plurality of defined prediction modes, or determination of a quantization step size, encoding is performed. It is characterized in that one or a plurality of control items are determined based on a cost function obtained by combining the distortion due to and the code amount required for encoding.

Further, in the high compression coding apparatus for coding the image signal of the moving image, when the motion vector is obtained by the motion compensation prediction, the matching distortion of the image signal and the variable length coding of the motion vector are performed. The feature is that the motion vector is obtained based on the cost function based on the code amount.

In particular, when a motion vector is calculated based on a cost function, a macro block having a low motion search accuracy is determined and a motion vector whose cost function is close to the origin is calculated from macro blocks having a low motion search accuracy. And

When the motion vector is calculated based on the cost function, the motion vector is calculated based on the cost function while determining the motion search range according to the code length of the motion vector.

Further, when the difference between the motion compensation prediction signal based on the motion vector obtained based on the cost function and the input image signal is inter-frame coded, the distortion of the image signal due to the inter-frame coding and the inter-frame code It is characterized in that inter-frame coding is performed based on a cost function based on the code amount of the coded output.

A high compression encoding apparatus for encoding an image signal of a moving image, wherein when an input image signal is intra-frame encoded, distortion of the image signal due to intra-frame encoding and intra-frame encoding output It is characterized in that the intra-frame coding is performed based on the cost function based on the code amount of.

Further, in the high compression coding apparatus for coding the image signal of the moving image, when the input image signal is subjected to intra-frame coding or inter-frame coding and the coding coefficient is scanned and quantized, The scanning is performed by weighting in the horizontal or vertical direction based on the cost function based on the distortion of the image signal due to the quantization and the code amount of the quantized output.

A high compression encoding apparatus for encoding an image signal of a moving image, wherein an input image signal is provisionally encoded in both intra-frame encoding and inter-frame encoding modes, and an image obtained by intra-frame encoding Based on the size of the cost function by the signal distortion and the code amount of the intra-frame coding output, and the cost function by the distortion of the image signal by the inter-frame coding and the code amount of the inter-frame coding output, the input image signal is It is characterized by determining whether to perform intra-frame coding or inter-frame coding.

Furthermore, it is possible to control the quantization step size when quantizing the encoded output in accordance with the efficiency of the case of using the cost function compared to the case of not using the cost function. Characterize.

In particular, it is characterized in that the processing based on the cost function is performed for every one or a plurality of macroblocks.

Further, it is characterized in that the processing based on the cost function is performed for every one or a plurality of frames.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1. FIG. 1 shows a configuration example of a high compression coding apparatus according to the present invention. In the figure, 1
Is a motion compensation prediction unit, 2 is a difference unit, 3 is a coding mode determination unit, 4 is a DCT unit, 5 is a quantization unit, 6 is a variable length coding unit, 7 is a buffer, 8 is an inverse quantization unit, 9 Is the inverse DCT part,
10 is an adder, 11 is a video memory, 12 is a switch, 1
Reference numeral 3 is an encoding control unit, and the basic device configuration is the same as the TM5 encoding system.

The feature of the present invention is that, as indicated by a balloon in FIG. 1, "motion" when the motion vector detected by the motion compensation prediction unit 1 is variable length coded (VLC) by the variable length coding unit 6. "Compensation + VLC processing" and the coding mode determination processing of "interframe coding processing" or "intraframe coding processing" in the coding mode determination unit 3, where the cost function D + lR is introduced and The motion compensation prediction unit 1 and the coding control unit 13 determine the coding parameters such as the following motion vector, coding mode, and coding unit, and the quantization in the quantizing unit 5 is performed by the amount corresponding to the improved coding efficiency. The image quality is improved by controlling the step size by the encoding control unit 13. Further, the encoding control unit 13 also realizes the scan order selection at the time of DCT coefficient quantization in consideration of the power distribution of DCT coefficients in the DCT unit 4, and the time-varying smoothing of the information generation amount in the quantization unit 5. The quantization step size is controlled. Next, a basic concept for solving the improvement problem in the present invention will be described.

[Basic Concept] The basic method for realizing the high compression adopted is the optimum setting of the coding parameters based on the rate / distortion criterion. The optimum coding parameter is a parameter that can obtain the minimum distortion (code amount) at a given rate (code amount) or the minimum rate (code amount) within the allowable distortion range. To determine the parameters, the parameters may be selected so that the cost function D + lR obtained by combining the rate (R) and the distortion (D) with l becomes small.

This will be described with reference to FIG. Figure 2
It is a figure explaining how to set a parameter based on a rate and a distortion standard. When l is given in advance, the value of the cost function is the point having the magnitude l of the slope and the line passing through the points of the rate and distortion as the coding result intersects the vertical axis. Clearly, the closer this point is to the origin 0, the better the coding efficiency. The basic principle of high compression coding is to determine the parameter group so that the following function is small when the state vector representing the setting of all control parameters related to coding is L.

Si [Di (L) + l Ri (L)]

Here, l corresponds to an undetermined multiplier of Lagrange, and has a value of 0 or more.

Next, the operation of the first embodiment will be described.
FIG. 3 is a flowchart showing the basic operation of the high compression coding apparatus according to the first embodiment. In the high compression encoding apparatus according to the first embodiment, first, in step 110, the input image signal is used together with the target code amount and f_
A parameter indicating the search range of motion compensation prediction called code, and a indicating the scan order of quantized DCT coefficients
Set a flag called lternate_scan. Note that f_c
The larger the ode, the wider the search range for motion-compensated prediction, but the larger the code amount of the motion vector, on the contrary. Then, each picture is encoded using the parameters set in step 100. Since one picture is composed of a plurality of slices and each slice is composed of a plurality of macroblocks, first, step 200 is performed. In step 300, a picture coding process is performed based on a parameter indicating a search range of motion compensation prediction called f_code. In step 300, a slice process is performed.
Performs macroblock encoding processing based on a flag called nate_scan.

FIG. 4 is a flowchart showing the encoding process of one macroblock in the macroblock encoding process of step 400 in FIG. The input image signal is subjected to motion vector detection processing by the frame MC and field MC at steps 410 and 420, and subsequently, determination processing for selecting one of the modes is performed at step 430. And the next step 440
Thus, motion compensation prediction is performed using the motion vector obtained in the mode selected in step 430 to obtain a differential signal. Next, at step 450, a process of determining which mode of intra-frame predictive coding or inter-frame predictive coding is selected is performed using the difference signal and the input image signal. The selected signal is processed in the next step 460.
DCT by the following, quantized by the following step 470 and the alte set in the step 100 after quantization.
Scan processing according to the rnate_scan flag is performed, and VLC (variable length coding) is performed in the next step 480.

In the first embodiment, the above-mentioned FIG.
Among the processing of FIG. 4, “motion compensation processing” in steps 410, 420, and 430, “DCT coefficient quantization processing” in step 470, and “information generation amount smoothing processing” by setting the target code amount in step 100 and the like. , And "the coding mode determination process for interframe coding / intraframe coding" in step 450, and the other processes are the same as the conventional ones.
The two processes will be described separately in detail.

[Motion Compensation] Usually, as shown in the statistical characteristics of motion vectors, motion vectors are concentrated in the vicinity of zero motion amount. In the TM5 coding method, since a motion vector is determined by a simple minimum distortion condition based on the sum of absolute difference, a vector different from the original motion of the image may be selected. Since code allocation to a motion vector assumes correlation with an adjacent macroblock, the code length is almost proportional to the difference from the vector of the adjacent macroblock. For this reason, when the motion search accuracy is low, such as when a motion vector unrelated to the original motion is selected or a redundant motion vector code is generated, the coding efficiency deteriorates.

Therefore, in the motion compensation prediction unit 1 of the first embodiment, the code amount R1 of the motion vector and the matching distortion D1 between the input image signal and the prediction signal having the motion vector are used in accordance with the rate / distortion criterion. , The cost function [D1 + λ1 · R1] by the motion vector code amount R1 and the matching distortion D1 is controlled to be the minimum, and the motion vector code amount R1 is reduced.

Further, the motion compensation prediction unit 1 of the first embodiment
Then, not only the reduction of the motion vector code amount R1 based on the cost function [D1 + λ1 · R1] by the motion vector code amount R1 and the matching distortion D1, but also the motion search in the motion compensation prediction unit 1 is performed as follows. When the accuracy is low, the motion vector code amount is reduced and the motion search range is optimized to reduce the redundant motion vector code amount.

(1) Motion vector when motion search accuracy is low
Reduction of Le code amount First, the motion search accuracy described method of determining the lower macroblock. ． Macroblock determination method with low motion search accuracy Here, the macroblock that the searched motion vector does not match the actual motion on the screen and may select a long codeword away from the base point of the search. It is expressed as "macro block with low motion search accuracy". In order to specify such a macroblock, the motion compensation prediction unit 1 according to the first embodiment does not make the determination by the cost function based on the rate / distortion criterion according to the present invention, and the obtained motion vector matches the motion. Determine if there is no vector. Based on the following conditions, it is determined whether or not the motion vector does not match the motion for each macroblock, that is, the macroblock with low motion search accuracy.

Σ _SAD ² (k) <Th1 or Σ | ΔMV |> Th2

Here, the meaning of each symbol is as follows. σ _SAD ² (k): Sum of Absolute Difference of matching distortion for the target macroblock
Dispersion of distortion values calculated by extracting k matching distortions in order from the smallest value and calculating. Σ | ΔMV |: Sum of absolute differences between the magnitude of the motion vector obtained for the target macroblock and the magnitude of the motion vector obtained for the eight surrounding macroblocks. Th1: A threshold for determining the degree of variation in matching distortion. Th2: Threshold for judging the degree of variation in motion vector.

.. Detection of Motion Vector with Short Code Length Next, reduction of the motion vector code amount by detecting a motion vector with a short code length will be described. In the prediction mode using a plurality of motion vectors in a macroblock, the code amount is large. Therefore, in the present embodiment, the use is limited only when the prediction effect is sufficiently high, and when there is no significant difference in the prediction effect. , The vectors are collected in the vicinity of the motion amount zero so that the code amount becomes small. For macroblocks with low motion search accuracy obtained by the method of 1. ~ 3. The motion vector of the short code length is detected according to the procedure of.

[Short code length motion vector detection procedure] 1. In order from the motion vector that gives the minimum matching distortion
Select k candidates as candidates 2. The k motion vectors are classified into three groups by prioritizing [frame prediction> field prediction> bidirectional prediction]. 3. The motion vector of the group with the highest priority is searched in order, and the motion vector closest to the origin is selected.

In FIG. 5, the bit rate is 15 Mbps, the GOP structure is N value = 15, M value = 3, and the frame rate is 29.97 frames / sec. [Mbit] is shown by comparing the conventional TM5 system with the proposed method proposed in the embodiment of the present invention. As can be seen from the figure, in the conventional (TM5) coding method, many frames exceeding 10 Mbps lines are generated, but in the proposed method according to the embodiment of the present invention, the number of frames is less than 5 Mbps lines and the motion vector It can be seen that the code amount has been significantly reduced.

(2) Motion vector when motion search accuracy is low
According as reducing the aforementioned Le code amount to the statistical properties of the motion vectors, the motion vectors are concentrated in the vicinity of the movement amount zero. On the other hand, in the conventional (TM5) coding method, the motion search range is fixed and the motion search range that covers the maximum amount of motion is used. Therefore, a motion vector with a long code length may be generated. There is.

FIG. 6 shows adaptation of the motion search range (f-code).
And the code length. As shown in FIG. 6, the motion search range is an f-code that has set values in the horizontal and vertical directions.
It is set by a parameter that takes a value from 1 to 9 and the code length also changes depending on the f-code value.

FIG. 7 shows a correspondence example of the motion search range corresponding to the combination of the horizontal and vertical f-code values. Since the motion search range and the code length of the motion vector have the relationship shown in FIG. 6 described above, when the search range is wide, the motion followability improves and the inter-frame difference corresponding to the distortion decreases, The code amount of the motion vector increases. On the contrary, if the motion search range is narrowed down, the code amount of the motion vector is reduced, but the inter-frame difference corresponding to the distortion is increased. Therefore, there is a trade-off between the motion followability and the code amount. I have a relationship. Therefore, if both the amount of code and the amount of distortion are taken into consideration based on the rate / distortion criterion, and the optimal f-code value can be selected according to the amount of motion of each frame of the coded video, the coding efficiency will be improved. It will be possible.

Next, in the present embodiment for explaining an example of a method of determining a motion search range adapted to a motion amount, although the motion amount changes for each video scene, a motion search range suitable for each video scene is set. Consider making an adaptive decision. That is, first, the rate and distortion characteristics at that time are examined to see how much the motion vector obtained without limiting the search range is included in the search range corresponding to the value of f-code. The probability that a motion vector obtained without limit is included in a certain search range is the motion vector appearance probability n (horizontal fc
ode value, vertical f-code value) can be expressed by the following formula.

N (horizontal f-code value, vertical f-code value) = A / B

However, A: the amount of motion is horizontal f-code
Value, total number of macroblocks within vertical f-code value search range B: Total number of macroblocks.

Next, it is assumed that the target value N of the motion vector appearance probability is set and the minimum f-code satisfying the following conditions is used for encoding.

N (horizontal f-code value, vertical f-code value) ≧ N

As a result, a large f-code is selected for a scene with large motion, and a small f-code is selected for a scene with small motion.

[DCT Coefficient Quantization] Generally, the DCT coefficient power distribution in an HDTV signal is statistically stronger in the horizontal pixel-to-pixel correlation than in the vertical pixel-to-pixel correlation. Statistically, the high frequency coefficient in the vertical direction has larger power than the high frequency coefficient in the horizontal direction. Considering the two-dimensional VLC in the variable-length coding unit 6, it is appropriate to scan the DCT coefficients in order from a coefficient having a large power to a coefficient having a small power. Conventional TM
The coding method of 5 does not employ a method of scanning DCT coefficients weighted horizontally and vertically.

Therefore, in the first embodiment of the present invention, the quantizing unit 5 is designed to improve the coding efficiency by changing the scanning method of the DCT coefficient according to the image.

That is, the quantizing unit 5 according to the first embodiment of the present invention controls DCT in MPEG-2 under the control of the coding control unit 13.
The scanning order of coefficients is selected for each picture from two types shown in FIGS. 8 (a) and 8 (b).

FIGS. 8 (a) and 8 (b) show the scan order of the DCT coefficients in the quantizer 5 of the first embodiment, and FIG. 8 (a) shows the code of TM5 with alternate_scan = 0. This is the scan order used in the conversion method, and scans in a zigzag manner from the DCT coefficient on the low frequency side to the coefficient on the high frequency side.

FIG. 8B shows a scan order in which the DCT coefficient on the low frequency side of the horizontal component of alternate_scan = 1 is preferentially scanned, and is effective when the correlation between pixels in the horizontal direction is high.

FIG. 9 shows rate / distortion characteristics in the Whale Show video when each scanning method is used. Considering rate / distortion criterion, alt shown in Fig. 8 (a)
Figure 8 (b) from the case of TM5 encoding method with ernate_scan = 0
It can be seen that the encoding efficiency of alternate_scan = 1 shown in is high.

[Smoothing of Information Generation Amount] On the encoding device side, the transmission buffer 7 is provided to smooth the time variation of the generated code amount.
In the coding method of 5), if intra (intra-frame coding) frame insertion due to a scene change or the like is performed, buffer failure may occur. If the code amount of this intra frame can be controlled with high accuracy, it becomes possible to prevent buffer failure.

Therefore, in the first embodiment of the present invention, DC
When the T unit 4 performs intraframe coding / interframe coding, the coding control unit 13 controls the DCT unit 4 to perform coding based on a cost function based on the distortion and the code amount. That is, as shown in the balloon in FIG. 1, when the input image signal is intra-frame encoded, the encoding control unit 13 instructs the DCT unit 4 to distort the image signal D3 due to the intra-frame encoding and the intra-frame encoding. Based on the cost function [D3 + λ3R3] based on the code amount R3 of the encoded output, the intra-frame control is performed with the distortion D3 and the code amount R3 that minimize the cost function. When the input image signal is inter-frame encoded, the encoding control unit 13 instructs the DCT unit 4 to distort the image signal D due to the inter-frame encoding.
Based on the cost function [D2 + λ2 · R2] of 2 and the code amount R2 of the inter-frame coded output, the inter-frame coding is controlled by the distortion D2 and the code amount R2 that minimize the cost function. Next, the case of intraframe coding will be described as an example.

FIG. 10 shows "quantization step size" and "code amount of intra picture" in Whale Show video and the like.
Shows the relationship. The “quantization step size” is a parameter corresponding to distortion, that is, image quality, and the “code amount of intra picture” is a parameter corresponding to rate. That is, as shown in FIG. 10, when the quantization step size is small, the distortion is small and the image quality is improved, but instead, the generated code amount is increased. The characteristics of the three sequences such as the Whale Show video are similar to the rate / distortion curve, although the characteristics differ depending on the variance of the video signals. If this characteristic can be generalized by some index regardless of the sequence, the code amount can be controlled accurately.

Further, the generated code amount of the intra picture,
The DCT coefficient sum of absolute values has a linear characteristic as shown in FIG. 11 experimentally regardless of the scene, and can be used for generalization of the code amount estimation method.

Next, the intra-frame code amount control method will be specifically described. It has been reported experimentally that the generated code amount is well approximated to a power curve of the quantization step size. Here, for generalization not depending on the video sequence, the characteristics shown in FIG. 11 are applied,
The "quantization step size" and "intra picture code amount / sum of DCT coefficient absolute values (AC component)" are used to approximate by the following formula.

Approximation formula: R = aQ ^b

Here, Q: quantization step size R: intra picture code amount / DCT coefficient absolute value sum (AC component) a, b: constants.

By using this approximate curve, the quantization step size of an intra picture can be derived from the target intra picture code amount and the sum of DCT coefficient absolute values, and a buffer failure due to an unexpected code amount is generated. Can be prevented.

In the above description, the case where the input image signal is intra-frame (intra) coded has been described.
The same description can be applied to the case where the input image signal is inter-frame (inter) coded. In this case, it is possible to reduce the amount of information generated by the inter-frame coding.

[Intra / Inter Mode Coding Mode Judgment] At the time of coding, the coding mode judgment unit 3 judges the intra / inter mode coding mode for each macroblock. The intra mode is a mode for coding without performing motion-compensated inter-frame prediction, and the inter mode is a coding mode for performing motion-compensated inter-frame prediction. Two pass coding was considered for this determination. The 2-pass encoding is a method of improving parameters used for actual encoding by using information obtained from the result of encoding once.

This is because in the conventional TM5 coding method, the intra / inter mode coding mode of each macroblock is E intra corresponding to the power of the current picture, and E corresponding to the motion compensation inter-frame prediction error power. It was judged as follows by comparing the magnitude of inter.

[0062]   Judgment condition:       Encoding mode = Inter mode E intra ≥ E inter       Encoding mode = intra mode E intra <E inter here,

[0063]

[Equation 1]

[0064] Is.

FIG. 12 is a plot of the evaluation values of the conventional (TM5) in one frame of video for each macroblock. Plot position is the reference line (Eintra = Einter)
The inter mode is selected above and the intra mode is selected below the reference line. Eintra corresponds to the distortion at the time of intra coding, and Einter corresponds to the distortion at the time of inter coding. In the conventional (TM5) coding system, the coding mode is selected using only the distortion as an evaluation measure.

In FIG. 12, the part surrounded by the ellipse is the conventional (TM
In fact, macroblocks with a smaller code amount are coded in the opposite mode to the one selected by the evaluation criterion in 5).

Therefore, in the first embodiment of the present invention,
The coding efficiency is improved by determining the intra / inter mode based on an evaluation scale (cost function) that takes into consideration not only the distortion but also the code amount based on the rate / distortion criterion. Specifically, intra / inter mode determination is performed as described below.

That is, in the intra / inter mode decision method of the first embodiment, the optimum mode decision based on the rate / distortion criterion is decided by the above cost function. Here, the effect when the selection of intra mode / inter mode is determined using this cost function will be seen.

Specifically, the cost functions E'intra and E shown below are tentatively encoded (2-pass encoding) in both the intra and inter modes, and based on the encoding distortion and the generated code amount obtained there. 'inter is used to determine the intra / inter mode coding mode.

[0070]   Judgment condition:       Encoding mode = inter mode E'intra ≥ E'inter       Coding mode = Intra mode E'intra <E'inter

Here, E'intra = D intra + λ · R intra (intra evaluation value) E'inter = D inter + λ · R inter (inter evaluation value) D intra: coding distortion at the time of intra coding ( MSE) R intra: Macroblock code amount (DCT at intra coding)
Coefficient) D inter: Coding distortion (MSE) at the time of inter coding R inter: Code amount of macroblock at the time of inter coding (motion vector + DCT coefficient) λ: A constant [MSE / code amount per macroblock].

In the first embodiment, the above-mentioned new "motion compensation process", "DCT coefficient quantization process",
The "information generation amount smoothing process" and the "interframe coding / intraframe coding coding mode determination process" have been described as being all executed, but the present invention is not limited to this.
From among the above-mentioned new “motion compensation processing”, “DCT coefficient quantization processing”, “information generation amount smoothing processing”, and “coding mode determination processing of interframe coding / intraframe coding” Only one process is executed, other processes are
The same processing as TM5 may be performed, or two or more processings may be arbitrarily selected from these new processings.

Further, in the first embodiment, the above-mentioned new "motion compensation processing", "DCT coefficient quantization processing",
The process based on the cost function of rate / distortion in the "information generation amount smoothing process" and the "inter-frame coding / intra-frame coding coding mode determination process" is applied to the correlation between macro blocks in the video to be coded. Accordingly, the processing may be performed for each one or a plurality of macro blocks, or for each one or a plurality of frames depending on the correlation between the frames.

As described above, according to the present invention,
When the motion vector is obtained by motion compensation prediction, the motion vector is obtained based on the matching distortion of the image signal and the cost function based on the code amount when the motion vector is variable length coded. The amount can be reduced, and the image quality can be improved by quantizing the coding coefficient in small steps.

Further, when the difference between the motion compensation prediction signal based on the motion vector obtained based on the above cost function and the input image signal is inter-coded, the distortion of the image signal due to the inter-frame coding and the inter-frame coding output Since the interframe coding is performed based on the cost function based on the code amount of, the code amount of the interframe coding output can be reduced, and the image quality can be reduced by quantizing the coding coefficient in small steps. Can be improved.

When the input image signal is intra-frame encoded, the intra-frame encoding is performed based on the cost function of the distortion of the image signal due to the intra-frame encoding and the code amount of the intra-frame encoded output. Therefore, the code amount of the intra-frame coded output can be reduced and the code amount of the intra-frame coded output can be smoothed, and the image quality can be improved by quantizing the coding coefficient in small steps. In addition to the improvement, it is possible to prevent the transmission buffer from being broken at the time of inserting the intra frame due to a scene change or the like.

When the input image signal is intra-frame encoded or inter-frame encoded and the encoding coefficient is scanned and quantized, the cost function based on the distortion of the image signal due to the quantization and the code amount of the quantized output is used. Since the scanning is performed by weighting in the horizontal or vertical direction based on, the correlation coefficient between pixels in the horizontal direction is statistically stronger than that between pixels in the vertical direction, such as in HDTV video. In the case of, the weighting is applied in the horizontal direction and the DCT coefficient on the low frequency side of the horizontal component is preferentially scanned to improve the variable length coding efficiency and reduce the code amount. The image quality can be improved by quantizing the quantization coefficient.

Further, the input image signal is provisionally coded in both intra-frame coding and inter-frame coding modes, and the cost function based on the distortion of the image signal due to intra-frame coding and the code amount of the intra-frame coding output, It is determined whether the input image signal is intraframe-coded or interframe-coded based on the size of the cost function based on the distortion of the image signal due to the interframe coding and the code amount of the interframe coding output. Therefore, it is possible to select a coding mode with a small code amount, and it is possible to improve the image quality by quantizing the coding coefficient in small steps.

[0078]

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration example of a first embodiment of a high compression encoding apparatus according to the present invention.

FIG. 2 is an explanatory diagram of parameter setting based on a rate / distortion criterion.

FIG. 3 is a flowchart showing the basic operation of the high compression encoding apparatus according to the first embodiment.

4 is a flowchart showing the encoding process of one macroblock in the macroblock encoding process of step 400 in FIG.

FIG. 5 is a diagram showing a comparison of motion vector code amounts (Whale
Show).

FIG. 6 is a diagram showing a relationship between adaptation (f-code) of a motion search range and code length.

FIG. 7 is a diagram showing an example of correspondence between horizontal / vertical f-code values and motion search ranges.

FIG. 8 is a diagram showing a scan order of DCT coefficients.

FIG. 9 is a diagram showing rate / distortion characteristics (Whale Show).

FIG. 10 is a diagram showing a quantization step size and the code amount of an intra picture.

FIG. 11 is a diagram showing the sum of absolute values of DCT coefficients measured in quantization step 8 and the amount of intra-picture generation code.

FIG. 12 is a diagram showing a distribution of macroblocks distributed according to a TM5 determination condition (one frame of Whale Show).

[Explanation of symbols]

1 motion compensation prediction unit, 2 difference unit, 3 coding mode determination unit, 4 DCT unit, 5 quantization unit, 6 variable length coding unit, 7 buffer, 8 inverse quantization unit, 9 inverse DCT unit,
10 adder, 11 video memory, 12 coding controller.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Koichi Tanno             2-3 2-3 Marunouchi, Chiyoda-ku, Tokyo             Inside Ryo Electric Co., Ltd. (72) Inventor Yamada Etsuhisa             2-3 2-3 Marunouchi, Chiyoda-ku, Tokyo             Inside Ryo Electric Co., Ltd. (72) Inventor Kotaro Asai             2-3 2-3 Marunouchi, Chiyoda-ku, Tokyo             Inside Ryo Electric Co., Ltd. (72) Inventor Shigeo Nakajima             2-3 2-3 Marunouchi, Chiyoda-ku, Tokyo             Inside Ryo Electric Co., Ltd. F term (reference) 5C059 KK01 KK19 MA05 MA23 MC01                       MC11 ME01 NN01 PP05 TA00                       TA23 TA46 TA62 TB07 TC00                       TC04 TC18 UA02                 5J064 AA01 BA09 BA16 BB03 BB05                       BC01 BC08 BC16 BC28 BD01

Claims (11)

[Claims] 1. A high compression encoding apparatus for encoding an image signal of a moving image, wherein a control item that can be dynamically determined in the encoding process, that is, whether to use interframe prediction or motion in motion compensation prediction When determining control items such as vector determination, mode selection from multiple defined prediction modes, or quantization step size determination, combine distortion due to encoding with the code amount required for encoding. A high-compression encoding device, characterized in that one or more control items are determined based on the obtained cost function. 2. A high compression encoding apparatus for encoding an image signal of a moving image, wherein when a motion vector for motion compensation prediction is obtained, matching distortion of the image signal and the motion vector are variable length encoded. A high compression encoding device characterized in that a motion vector is obtained based on a cost function depending on the code amount at that time. 3. The high compression coding apparatus according to claim 1 or 2, wherein when a motion vector is obtained based on a cost function, a macro block having low motion search accuracy is determined,
A high compression encoding device characterized in that a cost function obtains a motion vector close to the origin from macroblocks having low motion search accuracy. 4. The high compression coding apparatus according to claim 1, wherein when a motion vector is obtained based on a cost function, a motion search range is determined based on a code length of the motion vector, A high compression encoding device characterized by obtaining a motion vector based on a cost function. 5. The high compression encoding apparatus according to any one of claims 1 to 4, further comprising: interframe coding a difference between a motion compensation prediction signal and an input image signal by a motion vector obtained based on the cost function. A high-compression encoding device characterized by performing inter-frame encoding based on a cost function based on a distortion of an image signal due to inter-frame encoding and a code amount of inter-frame encoded output. 6. A high compression encoding apparatus for encoding an image signal of a moving image, wherein when an input image signal is intra-frame encoded, distortion of the image signal due to intra-frame encoding and intra-frame encoding output A high-compression encoding device, which performs intra-frame encoding based on a cost function according to the code amount of. 7. A high compression encoding apparatus for encoding an image signal of a moving image, wherein an input image signal is scanned and quantized by encoding coefficients in a process of intra-frame encoding or inter-frame encoding. A high-compression encoding device which performs scanning by weighting in the horizontal or vertical direction based on a cost function based on the distortion of an image signal due to quantization and the code amount of a quantized output. 8. A high compression encoding apparatus for encoding an image signal of a moving image, wherein an input image signal is provisionally encoded in both intra-frame encoding and inter-frame encoding modes, and an image is produced by intra-frame encoding. Based on the size of the cost function by the signal distortion and the code amount of the intra-frame coding output, and the cost function by the distortion of the image signal by the inter-frame coding and the code amount of the inter-frame coding output, the input image signal is A high-compression encoding device characterized by determining whether to perform intra-frame encoding or inter-frame encoding. 9. The high compression coding apparatus according to any one of claims 1 to 8, further comprising: depending on an improved efficiency when the cost function is used as compared with the case where the cost function is not used. And a quantization step size when quantizing a coded output. 10. The high compression coding apparatus according to claim 1, wherein a processing based on a cost function is performed for each one or a plurality of macroblocks. 11. The high compression coding apparatus according to claim 1, wherein the high compression coding apparatus performs processing based on a cost function for each one or a plurality of frames.