JP3380841B2 - Error correction system and decoding device using the error correction system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an error correction system in a decoding device for decoding image data that is entropy-coded and transmitted as a code stream, and particularly corrects an error in the code stream generated on a transmission path. The present invention relates to an error correction method for an entropy-coded code stream.

[0002]

2. Description of the Related Art ISDN (Integrated Services Digita)
l Network: Comprehensive digital communication network), and image communication services have been realized as a new communication service against the backdrop of the spread of high-speed digital communication networks and advances in image processing technology. Further, with the development of wireless transmission networks represented by PHS, demands for further sophistication, diversification and portability of services are increasing.

Generally, when transmitting image information, the information amount of the image is enormous, but in view of the line speed and cost of the line used for the transmission, the information amount of the image of the transmission is compression-coded to obtain the information amount. It becomes necessary to reduce the transmission. As a coding method for compressing image information, JPEG (Joint Photog
raphic Coding Experts Group), H.264 as a moving picture coding method. 261. MPEG as a moving picture coding method for storage
(Moving Picture Coding ExpertGroup) 1, MPEG
2 has already been internationally standardized.

As a compression method of these image information, hybrid encoding is performed by combining the following three ideas. (1) In natural images and the like, the values between adjacent pixels are close to each other, so information compression is performed using the intra-screen (spatial) correlation (strictly speaking, pixel difference values between screens are also compressed. Do the calculation). (2) Information on the previous screen is stored and represented by a difference value from the current screen and the previous screen, thereby performing information compression using the inter-screen (temporal) correlation. (3) When encoding by the methods of (1) and (2) above, information compression is performed using the bias of the appearance probability of the code called entropy coding.

The above (3) will be described with reference to the diagram of FIG. 6 showing an example of entropy coding. As shown in FIG. 6, this is a coding method that shortens the average code length by having a code table in which the shortest code is assigned to the data with the highest appearance rate and the longest code is assigned to the code with the lowest appearance rate. .

[0006] Next, as an example of the structure of image data, H.264. Two
A schematic diagram of the structure of the image data 61 is shown in FIG. The structure is largely a picture layer, GOB (Group of Blocks)
It forms a hierarchical structure consisting of four layers: a layer, a macroblock (MB) layer, and a block layer. A picture is screen data for one frame, and a picture header 702
And GOB data 703 following it. The picture header 702 is data indicating attributes common to one frame such as a PSC (Picture Start Code) 701 which is a sync word of a picture start signal.

The GOB 703 is screen data obtained by dividing one screen into 12 and includes a GOB header 705 and MB data 706 following the GOB header 705. GOB header 705
GBSC (Gob Star
t Code) 704 etc. is data indicating information common to GOBs.

The MB data 706 is the MB header 707.
And block data 708 following it. M
B706 is a pixel block obtained by dividing GOB into 33 pieces, and an MB header 707 is composed of data indicating information common to MBs. The block 708 is composed of four luminance blocks and two chromaticity blocks, and is composed of a DCT transform coefficient TCOEF (Transform Coefficients) 709 and an end code EOB (End of Block) 710. Here, the MB layer and the block layer are code streams coded by entropy coding using the bias of the code appearance rate.

[0009]

Due to transmission errors such as data loss and errors that occur when the above code stream is transmitted to the decoding device, when the transmitted encoded data is decoded and displayed, it is regarded as a large error. It may be displayed. In particular, since the encoding is performed by entropy encoding, if an error occurs during transmission, there is a problem that the erroneous decoding screen is displayed until the break of the code stream is unknown and a synchronization word such as GBSC is obtained. .

In order to solve this problem, an intra-coded (intra-coded) code stream is transmitted at a constant cycle to refresh the screen, or when decoding is performed as shown in the flow chart of FIG. The entropy-encoded code stream is extracted in sync word units such as GOB units (step S801, the following step is abbreviated and is simply expressed as S), and GOB
When the entropy code of is decoded, code words that are not in the code table are decoded, or when the code stream is decoded, the number of transform coefficients in the block layer becomes inconsistent. Etc., and C
When RC (Cyclic Redundancy code) or the like is added, if a contradiction such as an error is detected (S803), source decoding is not performed on the code stream after that. By displaying the data of the previous screen (S806), the decoding screen is displayed by a decoding process such as making the error less noticeable until refreshed.

However, in the above decoding process, as shown in FIG. 9, the position at which a contradiction is detected in the decoding of the entropy code is located after the normally decodable portion 902, so that the normally decodable portion 903 is not present. However, since one part performs source decoding and displays the decoding screen, there is a problem that the one part displays an incorrect decoding screen as it is. The present invention has been made in view of the above problems, and provides an error correction method and a decoding device using the error correction method, which solves the above problems.

[0012]

The invention according to claim 1 of the present application is as follows.
Code stream entropy coded image data
In the error correction system, decodes the entropy code
When a codeword that is not in the codebook is decoded or converted
If there is a discrepancy in the decoding of the entropy code that causes the number of coefficients to become incorrect , or if a decoding error occurs
Includes inconsistent entropy code, if detected
Each macro block starts from the beginning of the code stream of a predetermined unit.
And the decoded values of the macroblocks around it as a constant reference value
Determine whether the decoded value is correct by comparing
Step (S301) and the step (S30)
From the result of 1), the code position that is highly likely to have an error is estimated.
Setting step (S302) and the step (S30)
For the code position where the error estimated in 2) is likely to occur,
Assume an error pattern in the corresponding code stream
Using the step (S303) and the assumed error pattern
The code stream corrected by
Step of determining whether or not tropy decoding is possible (S30
4), and if there is no contradiction and decoding is possible to the end, an error occurs
Motion vector, depending on the location of the estimated code stream.
Toll value or the difference value between the conversion coefficient and the reference screen is the decoded value
The step of obtaining the evaluation value of the certainty (S30
5) and changing the error pattern,
Performs the same processing as (S304 to S305) without any contradiction
Error that can be decoded up to the end of the code stream of the predetermined unit
If there is a pattern, the motion vector
The value is the closest to the average value obtained from the surrounding motion vectors.
Is the one with the smallest difference value between the conversion coefficient and the reference screen
Is the highest evaluation value, and the evaluation value is the highest
Selecting a pattern (S309),
Adopt the error pattern selected in step (S309)
Perform source decoding on the corrected code stream
Characterized by a step (S310).

[0013]

[0014]

[0015] the invention of claim 2 is a decoding device, e Bei processing unit employing the error correction method according to the <br/> claim 1
It is characterized by that .

In the decoding device having the above-mentioned structure, first, when decoding the entropy-coded code stream,
The code stream is taken out in sync word units and the entropy code is decoded. At that time, if there is a contradiction in the decoding of the entropy code, the decoded image is used to check the validity of the decoded value from the beginning of the extracted code stream, and to estimate the location where an error is likely to occur. .

In a code stream corresponding to a place where an error is likely to occur, a pattern of an error that is a bit error after the specified code position or a deviation of the start position of the code is assumed, and the assumed error is corrected. Entropy decode the stream again to determine if entropy decoding is possible without contradiction.

Then, the assumed error pattern is changed and the same processing is performed. If there is a pattern that can be decoded to the end without contradiction, the error candidate with the highest evaluation value is adopted and source decoding is performed to correct the error that occurred during transmission.

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. First, the environment of the present embodiment will be described with reference to FIG. FIG. 1 is a configuration block diagram of an encoding device, a decoding device according to the present invention, and peripheral devices thereof. An image is input from an image input device 101 such as a camera, an input image is encoded by a transmission side encoding device 102, and the reception side decoding device 1 is transmitted via a transmission path 103 such as a radio.
The encoded data is transmitted to 04. Receiving side decoding device 104
Then, when the encoded data is received, the decoding is started and the decoded image is displayed on the display device 105 such as a CRT.

Next, the operation of the decoding device during transmission of the entropy-encoded code stream, which is a feature of the present invention, will be described with reference to the flowchart of FIG. First,
From the received entropy-encoded code stream, the GOB synchronization word GBSC is detected to extract one GOB code stream (S201).

Next, the GOB entropy code is decoded (S202). Then, when decoding the entropy code, a code word that is not in the code table that is already stored is decoded, or when the code stream is decoded, the number of transform coefficients in the block layer becomes inconsistent (when entropy decoding is performed up to the end, Bits are left over) etc. Also, C
When RC is added, it is determined whether or not there is a contradiction in the decoding of the entropy code such as an error detected (S203).

If there is no contradiction, G extracted as usual
If the OB is the last GOB, source decoding is performed (S205), and if it is not the last, the next 1 GOB code stream is extracted (S204). If a contradiction occurs during the decoding of the entropy code, the processing shown in the error correction flowchart of FIG. 3 is performed (S206). First, in order to select a code in which an error has occurred, the validity of the entropy-decoded value is checked for each MB from the beginning of the taken GOB (S301).

Here, the checking method will be described with reference to the flowchart of FIG. First, the value of the motion vector is checked from the first MB (S401).
If there is a code word (value outside the motion vector search range) that is not in the code table, it is determined that there is an error in the code stream representing the value of the motion vector, and the check ends (S40).
2).

If the values are normal, then the sum of the conversion coefficients of the YUV blocks (the sum of the difference values from the reference screen) is calculated,
Compare with a certain standard value. The constant reference value is a value close to 0 (S403). When it is less than a certain reference value, that is, when the motion prediction is considerably correct, it is difficult to consider that the code stream of the block has an error, and therefore each block of the MB is assumed to have no error.

In the case of a certain standard or more, that is, in the case where the motion prediction is considerably deviated, both the case where there is an error in the code stream of the block and the case where the motion is vigorous are considered. M around
For B as well, the sum of the conversion coefficients (the difference value from the YUV value of the reference screen) of each YUV block is calculated and compared with a constant reference value (S404). If the YUV blocks of the surrounding MBs are also above a certain standard, it is considered that the MB to be checked is a scene with a lot of movement, and if it is below a certain standard, only the MB to be checked is predicted. Since it is hard to think that the value is significantly deviated, it is determined that there is an error in the code stream representing the transform coefficient of the block of the MB, and the check is ended (S405). If no error is found (not found), if it is not the last MB, the same processing as above is performed for the next MB (S
406).

Next, it is estimated which code position in the code stream the error of the MB or block determined by checking the validity of the above decoding value is (S30).
2). Based on the estimated code position, an error pattern such as a subsequent bit error or deviation of the code head position is assumed, and error pattern candidates are given (S303).

Here, a method of selecting a hypothesized error pattern candidate will be described with reference to FIG. The location of the error estimated in the above code stream is assumed to be the broken line in the figure. The code stream after the code position is compared with the code table, and candidates are listed in order from those that are considered to be error patterns that are considered to be 1-bit errors (code 50).
1). Next, the candidates are raised in the order of 2 bits. Although it is preferable to raise all possible candidates for error correction, it is terminated to some extent according to the processing amount of the decoding device.

Further, by assuming an error due to the deviation of the start position of the code and comparing it with the code table, an error pattern candidate is given (indicated by reference numeral 502). Furthermore, the average of the motion vector value, the luminance value, and the chromaticity value of the MB around the MB that is determined to have an error is averaged to obtain the predicted value of each value.

Then, a code candidate is given by comparing the code table with the obtained predicted value (code 503).
). It is also possible to reduce the processing amount of the decoding device by narrowing down the error pattern candidates by adopting the error pattern candidates having the same code as the code candidate. Then, the error pattern candidates assumed by the error pattern candidate selection method are adopted one by one, and the entropy decoding is performed again.

Similarly to the above (S203), if a code word that is not in the code table that is already stored is decoded or the code stream is decoded, the number of transform coefficients in the block layer becomes inconsistent (entropy decoding was performed until the end. It is determined whether the entropy code can be decoded without contradiction (S304). If entropy decoding can be performed without contradiction, an evaluation value of certainty is obtained based on the decoded screen (S305).

Here, how to obtain the evaluation value of the certainty will be described. The location of the error code stream is M
If it is a place that represents the motion vector of B, the one that adopts the error pattern which is the motion vector having the value closest to the average value obtained from the motion vectors of the MBs around the errored MB is adopted as the highest evaluation value. To do. Further, if the location of the code stream in which the error has occurred is the location representing the transform coefficient of the block, the one with the smallest sum of the transform coefficients of the blocks (the difference value with the YUV value of the reference screen) is the highest evaluation value. And In other words, it means that the motion estimation is the best, and the error pattern is most likely to be adopted.

Then, after the likelihood evaluation value is obtained by the above method, the next error pattern candidate is adopted, and the same processing is performed up to the final assumed error pattern candidate (S306, S307). When there is a pattern that can be entropy decoded without contradiction by adopting the error pattern, the error pattern having the highest evaluation value among the evaluation values obtained in (S305) is adopted in the code stream (S3
09), and source decoding is performed (S310). Although an error pattern is adopted, if there is no pattern that can be entropy-decoded without contradiction (S308), error processing similar to the conventional one is performed (S311), and source decoding is performed.

[0033]

According to the present invention, when an error occurs during transmission in a code stream coded by a coding device for coding an image, the above-mentioned processing is performed so that the processing of the coding device is conventionally performed. remain the same as, Rukoto to correct revised the error
That Do and is possible. Further, it is possible to obtain the effect of conventional error handling and the party even impossible to correct Ri completely erroneous. Further, by changing the method of selecting error pattern candidates, it is possible to change the error correction efficiency depending on the processing amount of the decoding device.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a general encoding device, a decoding device according to the present invention, and a peripheral device.

FIG. 2 is a flowchart showing a processing operation of an error correction system according to the present invention and a decoding device using the error correction system.

FIG. 3 is a flowchart showing an error correction operation during transmission of a code stream of a decoding method using the correction method and the error correction method according to the present invention.

FIG. 4 is a flowchart showing a processing operation for checking the validity of entropy-decoded values of a decoding system using the correction system and the error correction system according to the present invention.

FIG. 5 is a diagram illustrating a correction method according to the present invention and a method of selecting an error pattern candidate assumed by a decoding apparatus using the error correction method.

FIG. 6 is a table showing an example of entropy coding.

7: H. FIG. 26 is a schematic diagram of the configuration of image data of H.261.

FIG. 8 is a flowchart showing a processing operation of a conventional decoding device.

FIG. 9 is a diagram of a codestream when an error occurs during transmission in the entropy-coded codestream.

[Explanation of symbols]

101 ... Image input device, 102 ... Transmission side coding device, 103 ... Transmission path, 104 ... Reception side decoding device, 105 ... Display device, 501 ... From estimated code position , An error pattern candidate assuming a subsequent bit error, 502 ... An error pattern candidate assuming an error due to a deviation of the start position of the code, 503 ... A value of a motion vector of surrounding MBs, a luminance value, Predicted value obtained from chromaticity value, 701 ... PSC, 702 ... Picture header, 703 ... GOB, 704 ... GBSC, 70
5 ... GOB header, 706 ... Macro block (MB), 707 ... MB header, 708 ... Block, 709 ... TCOEF, 710 ... EOB,
901 ... Start code, 902 ... Part that can be normally decoded, 903 ... Part that cannot be normally decoded

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H04N ^7/ 24-7/68

Claims (2)

(57) [Claims] 1. A entropy-coded error correction scheme of the code stream of the image data, when decoding the entropy code, the code word is not in the code table
Is decoded or the number of transform coefficients is incorrect
Such, when a conflict in the decoding of the entropy code occurs, or decoding when an error is detected, the inconsistency raw
Each macroblock and its surrounding macroblocks are recorded from the beginning of the codestream of the specified unit including the same entropy code.
The decoded value by comparing the decoded value of
For determining whether or not there is an error (S301)
Then, it is possible that an error occurred from the result of the step (S301).
Of the highly accurate code position (S302) and the possibility that the error estimated in the step (S302) has occurred
Error in the code stream corresponding to a highly reliable code position
And a code stream corrected using the assumed error pattern (S303)
Whether or not entropy decoding can be performed without any contradiction
And the step of determining (S304), it is estimated that an error has occurred if the decoding can be completed up to the end without contradiction.
The value of the motion vector, depending on the location of the code stream
Or, the difference value between the conversion coefficient and the reference screen can be confirmed from the decoded value.
And a step (S305) of obtaining the evaluation value of the resilience, and the error pattern is changed, and each step (S30) is performed.
4 to S305) and perform the above-mentioned predetermined without conflict.
Error pattern that can be decoded up to the end of the unit code stream
If there is a motion vector, the value of the motion vector
Which is the closest to the average value obtained from the motion vector of
The smallest difference value from the conversion screen reference screen is the highest
An error pattern whose evaluation value is the highest as an evaluation value
Of the error pattern selected in the step (S309) and the step (S309).
Source decoding of the code stream corrected using
It is characterized by a step (S310) of performing
Error correction scheme that. 2. The error correction method according to claim 1 is adopted.
A decoding device comprising a processing unit used for the decoding device .