JP2000333180A - Skip macro block inhibit control method, skip macro block inhibit controller and medium recording skip macro block inhibit control program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a method which prevents the first and last macro blocks of each slice consisting of one macro block or over from being changed into skip macro blocks by re-quantization processing in trans-coding that conducts re-quantization processing and to obtain its device and a program medium. SOLUTION: The method includes a quantization step (S1) where image information is quantized for each macro block, skip macro block detection steps (S2, S3) that detect a macro block which has been changed into a skip macro block having no motion compensation and needing no coding through the quantization processing, and macro block type revision steps (S4, S5) that revise the macro block type of the skip macro block detected by the skip macro block detection processing so as to avoid production of the skip macro block.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
For a slice layer composed of one or more macroblocks, a method, an apparatus, and a program for preventing the first and last macroblocks of each slice in question from being changed to skip macroblocks by the requantization process are recorded. It is about media.

[0002]

2. Description of the Related Art In recent years, multimedia that expresses different information such as characters, figures, sounds, and images by digital data, and integrates these media to handle them in a unified manner has been receiving attention in recent years. One of the key technologies to realize this multimedia more effectively is information compression technology.
Information compression technology is a technology that focuses on information redundancy and reduces the amount of information by reducing the information in redundant parts, thereby efficiently processing, storing, and transmitting a large amount of information. It becomes possible.

[0003] There is a large difference in the amount of information of various media. In particular, since a moving image has a huge amount of information, a large amount of information compression is required. There are various methods for compressing information, and by combining these methods,
Significant compression has been achieved. Generally, these compression functions are provided as an LSI (Large scale integrated circuit).

"Basics of digital image compression: Hiroshi Yasuda,
As described in "Watanabe Hiroshi", source coding of speech and images (source coding: coding for performing data compression of information sources using the statistical properties of signals) is an irreversible code. (Lossy coding) and lossless coding (lossless coding)
It consists of two parts. Normal image compression is lossy encoding. Compression is achieved by omitting a portion of the image information that is inconspicuous in visual characteristics.

[0005] Irreversible coding is mainly processing of signal prediction, conversion and quantization. Prediction is a technique for obtaining a predicted value of a current signal using a signal at a previous time, and encoding a difference between the two. The quantization is a technique for coding coefficients, and the quantization is a technique for coding a number of a representative value by replacing continuous signals with representative values at appropriate intervals.

Image data is compressed for the first time by performing quantization, irrespective of whether prediction or conversion is performed. As shown in FIG. 6, the number of states that a symbol can take is reduced by quantization, and the average code length is compressed. However, directly quantizing an image is equivalent to performing AD conversion with a small number of bits, and noise such as false contours appears. Therefore, it is necessary to perform processing so that an error due to quantization is concentrated on a visually inconspicuous portion. In predictive coding, quantization errors can be collected in a steep part of the luminance gradient. In the transform coding, an effect close to that of a low-pass filter can be obtained by roughening the quantization of the high frequency band.

[0007] The MPEG coding algorithm is based on a hybrid coding system using motion compensated inter-frame prediction and DCT. Although inter-frame prediction is effective for compressing moving image information, since the information of the previous frame is always used, the random access function cannot be performed. Therefore, a layer including a plurality of image frames called a GOP (group of pictures) is provided.

[0008] The GOP has three types of coded frames, that is, intra-coded picturs.
e: I-picture), inter-frame coded frame (predi
ctive-coded picture: P-picture, frame interpolation coding frame (bidirectionally predictive-coded p
picture: B-picture). FIG. 7 shows an example of the GOP structure. I, B, and P indicate picture types. → indicates the direction of motion compensation prediction. The numbers for each frame in the figure indicate the order of input to the encoder.
As shown, the first frame encoded in each GOP must be an I-picture. Since the I-picture is intra-frame encoding using DCT, it does not need information of a previous frame, so random access is possible by using address information of the start point of encoded data.

[0009] The coded data structure consists of six layers, to which signal processing and functions are assigned. They are shown in FIG. The GOP is located on the second layer. The upper layer is a sequence, from the start to the end of the video. The GOP layer is further divided into a picture layer, a slice layer, a macroblock layer, and a block layer.

One picture is composed of several slices. A slice is a collection of any number of macroblocks starting from the top left of the image and continuing to the bottom right in raster scan order. Since a synchronization code is assigned to the head of slice data, there is an advantage that synchronization can be restored in the next slice even if there is a data read error during decoding. In addition, since data of a slice can be decoded independently for that slice, parallel processing can be performed on a slice basis to speed up decoding.

In transcoding for performing requantization, the code of the input bit stream is used as it is as the code of the target bit stream up to the slice layer. A requantization value is determined according to the ratio. Assuming that the quantization parameter value at the time of the first quantization is Q1, the quantization parameter value at the time of requantization is Q2, and the quantization coefficient is level, the quantization coefficient modified_level after requantization is obtained by the following equation. However, modified_level is an integer value, and expression (1)
If the result of the operation is a real number, the decimal part is rounded off.

(Equation 1) As an example, an input bit stream of 8 [Mbps] is set to 4
FIG. 9 shows a change in the quantization coefficient value when transcoding to a target bit stream of [Mbps]. According to equation (1), when the target bit rate is much smaller than the input bit rate or when the original quantization coefficient value is small, all the requantization coefficients in the macroblock (hereinafter, MB) become zero.

The MB type of the P picture is (1) with motion compensation / encoding required, (2) without motion compensation / encoding required,
(3) with motion compensation / no coding, (4) intra,
There are seven types: (5) with motion compensation / coding required / quantization parameter change, (6) without motion compensation / coding required / quantization parameter change, and (7) intra / quantization parameter change. Of these, as shown in FIG.
If the quantized coefficients in the MB of "encoding required" or "no motion compensation / encoding required / quantization parameter change" become all 0 by the requantization processing, the MB of "no motion compensation / no encoding required" That is, a skip macro block occurs.

[0013]

However, the first and last MBs of each slice must not be skipped macroblocks (macroblocks having no coded data; decoding is not performed). Each slice has at least one MB, and there is no gap between slices. The position of the slice may be different for each image. The first slice must start with the first MB in the image, and the last slice must end with the last MB in the image.

Therefore, in the present invention, in each of the six blocks included in 1 MB of the P-picture, when a re-quantization process results in a skipped macroblock, M
Change the B type to avoid the occurrence of skipped macroblocks.

[0015]

According to the first aspect of the present invention,
In order to solve the above problem, a quantization step of quantizing image information for each macroblock, and detecting a macroblock in which the macroblock has become a skipped macroblock without motion compensation and without encoding by the quantization process. It is characterized by comprising a skipped macroblock detection step and a macroblock type change step of changing the macroblock type of the skipped macroblock detected by the skipped macroblock detection processing.

According to a second aspect of the present invention, there is provided a skip macroblock inhibition control method according to the first aspect, wherein a skip-inhibited macroblock detecting step for detecting the first and last macroblocks of each slice layer. Wherein the skipped macroblock detection step detects skipped macroblocks only for the first and last detected macroblocks.

According to a third aspect of the present invention, there is provided a skip macroblock prohibition control method according to the first aspect, wherein the macroblock type changing step includes the step of changing the skipped macroblock without motion compensation and coding. Is changed to the macro block type.

According to a fourth aspect of the present invention, there is provided the skip macroblock prohibition control method according to the first aspect, wherein the macroblock type changing step includes the motion compensation of the skipped macroblock and no encoding required. Is changed to the macro block type.

According to a fifth aspect of the present invention, in order to solve the above problem, in the skip macroblock inhibition control method according to the third aspect, the macroblock type changing step includes performing requantization during the skipped macroblock. The macroblock type is characterized by generating a code that minimizes the accompanying error and the shortest combination of CBP, quantization coefficient, and EOB code, and changing the macroblock type by coding.

According to a sixth aspect of the present invention, there is provided the skip macroblock prohibition control method according to the third aspect, wherein the macroblock type changing step includes performing requantization during the skipped macroblock. It is characterized in that the macroblock type is changed by preparing in advance a code that minimizes the accompanying error and the shortest combination of CBP, quantization coefficient, and EOB code and adding it.

According to a seventh aspect of the present invention, in order to solve the above-mentioned problems, a quantization means for quantizing image information for each macroblock, and the macroblock is provided with no motion compensation and no coding required by the quantization means. A skip macro block detecting unit that detects a macro block that is a skipped macro block; and a macro block type changing unit that changes a macro block type of the skipped macro block detected by the skip macro block detecting unit. It is a feature.

According to an eighth aspect of the present invention, there is provided a program for controlling the generation of skip macroblocks without motion compensation and without encoding when image information is quantized for each macroblock. In the recorded medium, a skip macroblock detecting step of detecting a macroblock in which a macroblock has become a skip macroblock by quantization processing, and changing a macroblock type of the skipped macroblock detected by the skip macroblock detection processing And a macro block type changing step.

[0023]

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

FIG. 1 shows a flowchart of the skip macroblock generation avoidance processing by the encoding generation of the present invention.

First, a re-quantization process is performed (s1), and then it is determined whether or not a macroblock (hereinafter, MB) has a quantization coefficient (s2). If there is a quantization coefficient in the MB, the skip MB occurrence avoidance processing ends, and if not, it is determined whether or not a motion vector exists (s3). If there is a motion vector, the process is terminated; otherwise, skip M
Since it is B, the following processing is performed.

The MB type is set to “no motion compensation / encoding required”
(S4), and a code is forcibly generated (s5).

As described above, when a skip MB occurs due to the requantization process, a CBP (coded block pattern), a quantization coefficient, and an EOB code (a flag indicating the end of block data: end of b
Forcibly generate and encode a code that minimizes the combination of locks. As a result, an effect equivalent to forcibly generating a quantization coefficient when a skip MB is generated as shown in FIG. 2 can be obtained, and the MB type is changed to “no motion compensation / encoding required”. Further, by preparing a combination having the shortest code length in advance, the scanning process becomes unnecessary as compared with the case where the quantization coefficient is forcibly generated.

Next, the forcibly generated code will be described.

As shown in FIG. 3, the CBP assigns a number to each of the MBs of the luminance signal and the chrominance signal, and performs numbering so that the invalid block can be identified. 1 for valid / invalid for block
Defined by a variable Pn (n = 1,..., 6) having a value of / 0,
The pattern number CBP is

CBP = 32 · P1 + 16 · P2 + 8 · P
3 + 4 · P4 + 2 · P5 + P6. This value indicates which block is invalid.

When the value of the CBP is 32, 16, 8, and 4, the code length becomes 4 bits. In each of these cases, only one block generates a quantization coefficient in the MB. C
When BP = 60, the code length is as short as 3 bits (111), but this is equivalent to the case where the quantization coefficient occurs in four blocks, so even if one coefficient occurs in each of the four blocks, the total Thus, four coefficients are generated, and the overall code amount increases. Therefore,
It is appropriate to use any one of 32, 16, 8, and 4 for the CBP value, and the generated code amount is 4 bits. The codewords at this time are 1010, 1011, 110
0 and 1101.

In the case of generating a coefficient at a DC coefficient position, the variable length code is the shortest code 1s (2 bits). At this time, run = 0 and level = 1. The run indicates the number of consecutive 0 coefficients (coefficients that have been quantized to 0), and the level is the quantization level immediately after that. Further, s is 0 when the level is 0 or more, and is 1 when the level is less than 0. On the other hand, when it is generated as another coefficient (AC coefficient position), reference numeral 011s
(4 bits) is the shortest code, in this case, run,
Both levels are 1. Although it is possible to use either case, when a new quantization coefficient at the DC coefficient position is generated, an error in the subsequent processing increases.

The EOB code is 10 (2 bits).

As described above, the following examples are given as examples of combinations of codes that are forcibly generated.

(1) When CBP = 8, DC coefficient position (absolute value of level <0):

1100 1110

(2) When CBP = 4, AC coefficient position (absolute value of level ≧ 0):

1101 0110 10

Next, FIG. 4 shows a flowchart in the case where a motion vector is set and skip MB occurrence avoidance processing is performed when a skip MB occurs.

As in the skip MB generation avoidance processing by encoding generation, first, re-quantization processing is performed (s11), and then it is determined whether or not a quantization coefficient exists in the MB (s12). If there is a quantization coefficient in the MB, the skip MB occurrence avoidance processing is terminated. If not, it is determined whether or not a motion vector exists (s1).
3). If there is a motion vector, the processing is terminated. If not, the following processing is performed because a skip MB exists.

Motion vector information in which both the horizontal and vertical components are 0 is generated (s14), and the MB type is changed to "with motion compensation / no coding required" (s15).

By the above processing, when a skip MB is generated, as shown in FIG. 5, the MB type is "with motion compensation / no coding required", the motion vector becomes an MB of 0 in both the horizontal and vertical directions. Can be avoided.

The change of the MB type of the skip MB may be performed only at the beginning and end of each slice layer. However, detection of the first and last MB of each slice layer requires a large delay. Therefore, all the MBs are targeted, but the detection of the MB position may be selected according to the situation.

[0044]

According to the present invention, in transcoding for performing requantization processing, when the macroblock type becomes a skipped macrobook in which there is no coded data due to requantization processing and decoding processing is not performed. Since the macroblock type is forcibly changed, the first and last macroblocks of each prohibited slice layer can be prevented from being skipped macroblocks.

Further, it is possible to select whether to change the macroblock type of the skipped macroblock only at the beginning and end of each slice layer or to target all macroblocks according to the situation.

Further, by generating and encoding a code in which the combination of the CBP, the quantized coefficient and the EOB code is the shortest, or adding it in advance, it is equivalent to forcibly generating the quantized coefficient. The effect can be obtained, and the scanning process becomes unnecessary.

Further, by setting a motion vector of 0 in the horizontal and vertical directions and changing the macroblock type, it is possible to avoid the occurrence of a skipped macroblock without affecting the image after quantization. Can be.

[Brief description of the drawings]

FIG. 1 is a flowchart of skip MB generation avoidance processing by encoding generation according to the present invention.

FIG. 2 is a diagram illustrating avoidance of skip MB generation by code insertion.

FIG. 3 is a diagram showing MBs of numbered luminance and color difference signals for CBP (encoded block pattern).

FIG. 4 is a flowchart of a skip MB occurrence avoiding process by setting a motion vector value according to the present invention.

FIG. 5 is a diagram illustrating how to avoid skip MB generation by setting a motion vector value.

FIG. 6 is a diagram illustrating quantization in lossy encoding.

FIG. 7 is a diagram showing a GOP structure.

FIG. 8 is a diagram showing layers of an encoded data structure.

FIG. 9 is a diagram illustrating a change in a quantization coefficient value due to a requantization process.

FIG. 10 is a diagram illustrating a change in MB type by a requantization process.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Kazuyuki Takaya 1-3-10 Nishi-Waseda, Shinjuku-ku, Tokyo Waseda University International Information and Telecommunication Research Center (72) Inventor Hiroyuki Kasai 1-chome Nishi-Waseda, Shinjuku-ku, Tokyo No. 3-10, Waseda University International Information and Telecommunication Research Center (72) Inventor Satoshi Nishimura 1-3-10, Nishi-Waseda, Shinjuku-ku, Tokyo Inside Waseda University International Information and Telecommunication Research Center (72) Inventor Wataru Kameyama Tokyo 1-3-10 Nishi-Waseda, Shinjuku-ku, Waseda University International Information and Communication Research Center (72) Inventor Hideyoshi Tominaga 1-3-1-10 Nishi-Waseda, Shinjuku-ku, Tokyo Waseda University International Information and Communication Research Center F-term ( Reference) 5C059 KK02 KK23 LA02 LC10 MA00 MA05 MA14 MA23 MC11 PP05 PP06 PP07 RB03 RB12 RB13 SS20 SS26 TA00 TA11 TA17 TA21 TA45 TA57 TB06 TB07 TC00 TC04 TC06 TC12 TC32 TC37 TD01 TD02 TD18 UA05 UA39

Claims (8)

[Claims] 1. A quantization step for quantizing image information for each macroblock, and a skip macro for detecting a macroblock in which said macroblock has become a motion-free and no-encoding skip macroblock by said quantization processing. A skip macroblock inhibition control method, comprising: a block detection step; and a macroblock type change step of changing a macroblock type of a skipped macroblock detected by the skipped macroblock detection process. 2. The skip macroblock prohibition control method according to claim 1, further comprising a skip prohibition macroblock detection step of detecting first and last macroblocks of each slice layer, wherein the skip macroblock detection step includes the detection step. A skip macroblock prohibition control method, wherein a skipped macroblock is detected only for the first and last macroblocks. 3. The skip macroblock prohibition control method according to claim 1, wherein said macroblock type changing step changes said skipped macroblock to a macroblock type requiring no motion compensation and coding. A skip macroblock prohibition control method. 4. The skip macroblock prohibition control method according to claim 1, wherein said macroblock type changing step changes said skipped macroblock to a macroblock type with motion compensation and without coding. A skip macroblock prohibition control method. 5. The skip macroblock prohibition control method according to claim 3, wherein the macroblock type changing step is such that an error due to requantization is minimized in the skipped macroblock, and CBP, a quantization coefficient, A skip macroblock prohibition control method, characterized in that a code that minimizes the combination of EOB codes is generated and encoded to change the macroblock type. 6. The skip macroblock prohibition control method according to claim 3, wherein the macroblock type changing step is such that an error due to requantization is minimized in the skipped macroblock, and CBP, quantization coefficient, A skip macroblock prohibition control method, characterized in that a code that minimizes the combination of EOB codes is prepared in advance and the macroblock type is changed by adding the code. 7. Quantizing means for quantizing image information for each macroblock, and a skip macro for detecting a macroblock in which the macroblock has been determined to be a motion-compensation-free and coding-free skip macroblock. A skip macroblock prohibition control device comprising: a block detection unit; and a macroblock type change unit that changes a macroblock type of a skipped macroblock detected by the skip macroblock detection unit. 8. A medium in which a program for prohibiting the generation of a skipped macroblock without motion compensation and encoding is not recorded when quantizing image information for each macroblock, the macroblock is skipped by a quantization process. A skipping macroblock detecting step of detecting a macroblock that has become a macroblock; and a macroblock type changing step of changing a macroblock type of the skipped macroblock detected by the skipping macroblock detecting process. On which a skip macroblock prohibition control program is recorded.