JP2000036760A - Device and method for decoding data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain speedy recovery of decoding abnormality by returning the position of a bit to be decoded through a decoding means for just a continuous zero count value, when an abnormality detecting means detects the decoding abnormality, and attaining resynchronization by searching a unique word from that bit position. SOLUTION: When it is judged that error has been detected, a resynchronizing part 4 detects whether the bit under decoding at present is included in one part of a start code or not. This detection is judged based on the continuous zero count value. When the continuous zero count value is more than '1' at the time of error detection, it is judged that the possibility exists for the existence of invading one part of the start code is and the resynchronizing part 4 executes processing for a returning the decoding bit position of a decoder 1 by just for the bits of the continuous zero count value. Then, the start code is searched again, and processing for attaining resynchronization is executed. The continuous zero count value returned is reset.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data decoding device and a data decoding method for enhancing resistance to bit errors occurring in a bit stream.

[0002]

2. Description of the Related Art FIG. 14 is a data structure diagram showing a structure of an MPEG-1 bit stream to which a conventional data decoding method is applied, and FIG. 15 is an explanatory diagram showing a relationship between data constituting a bit stream. .

Next, the operation will be described. The largest chunk of the bitstream is called a "sequence". A sequence is a unit in which a plurality of GOPs (groups of pictures) are put together. The sequence data is
A sequence header indicating the beginning of the sequence data is added to the beginning. In order to indicate the start of the sequence header, a 32-bit unique word (a code word that can only be uniquely interpreted in a bit stream) called a “sequence header code” is multiplexed.

Similarly, "GOP" is a unit in which a plurality of pictures are put together. A GOP header indicating the beginning of the GOP data is added to the head of the GOP data. GOP
To indicate the start of the header, a 32-bit unique word called “GOP start code” is multiplexed.

[0005] A "picture" is a unit obtained by combining a plurality of slices, and generally corresponds to a "frame" which is a sampling unit in the time direction of a television signal. Sequence and GO
Similarly to P, a picture header indicating the beginning of the picture data is added to the head of the picture data. The head of the picture header is a 32-bit unique word called "picture start code".

[0006] A "slice" is a unit in which a plurality of macroblocks are put together. At the beginning of the slice data, 32
A slice header starting from a “slice start code” that is a unique word of bits is multiplexed. A “macro block” is image block data composed of a luminance component of 16 pixels × 16 lines and a chrominance component of 8 pixels × 8 lines, and there is no start code in this unit. Also, image data of 8 pixels × 8 lines in a macroblock is called a “block”. There is no start code in this unit.

Here, the macro block and the block are:
In order to encode image data using variable-length coding frequently,
It should be noted that the bit length generally differs on a bit stream. It should be noted that all start codes are equally 32 bits long and include 23 bits of 0 and 1 bit of 1 at the beginning, so that 24 bits are common to all start codes (FIG. 16). The bit pattern of the start code is HEX display. Each of them has 23-bit 0 and 1-bit 1 at the head, and the type is distinguished by 8-bit data thereafter.

In the above-described structure of the MPEG-1 video bit stream, the layers from the sequence to the slice include a start code for identifying each layer. Not included. Each of these start codes
In addition to indicating the start of the hierarchy, when a bit error occurs in the bit stream and a decoding error occurs, it can be used to return to normal decoding.

For example, as shown in FIG. 17, a decoding error is detected while decoding the data of the third macroblock (MBn, 2) belonging to the n-th slice, and the preceding bit stream is decoded therefrom. Consider the case where decoding cannot be performed. In this case, by detecting the start code of the next (n + 1) -th slice (arrow indicated by a dotted line in the figure), decoding is correctly resumed from the subsequent bit stream data (= macro block belonging to the (n + 1) -th slice). It is possible to do. As described above, the start code is used for the resynchronization process in which the process is out of synchronization due to a decoding error and returns to the normal decoding operation.

However, for example, as shown in FIG. 18, due to the effect of bit errors, the last macroblock in a certain slice (in this figure, the case where the nth slice contains eleven macroblocks is described. ), The decoding may be continued from the start code of the next (n + 1) th slice. This is because the start code itself is not detected as a unique word due to a bit error, and the variable length coding code used in the block data of the macroblock before the start code is changed to another codeword due to the bit error. In some cases, it is determined that the bits that should normally end decoding have not yet reached the end of the macroblock.

The case where the start code itself is committed by a bit error is considered to be avoided to some extent by lowering the error rate at the location of the start code by contrivance such as transmission line coding. It is not likely that much. However, the latter case relates to general block data, and it is difficult to devise a method particularly in transmission path coding, and it is considered that the occurrence frequency is higher than that of the former case. In the latter case, despite the presence of a correct start code in the bit stream, decoding is continued without being recognized as a start code. Therefore, when a decoding error is detected, the start code of the (n + 2) th slice is detected. To re-sync. As a result, the data of the (n + 1) th slice which may have been correct is completely discarded.

This becomes a more serious problem when considering a higher-order start code than a slice. For example, considering the end of a picture, the picture start code that appears next is decoded as a part of the data of the last macroblock of the current picture. Data is discarded. This causes a mismatch in the reference image of the motion compensation prediction when decoding the image after resynchronization, which has an adverse effect such as a significant deterioration in quality.

[0013]

Since the conventional data decoding method is configured as described above, for example, a decoding error occurs at the time of decoding block data (variable-length codeword), and a start code (unique word) is generated. ) Is decoded as block data, the start code is not recognized as a start code, so resynchronization cannot be achieved until the next start code is detected, and a decoding error is quickly recovered. There were issues such as being unable to do so.

[0014] The present invention has been made to solve the above-described problems. Even if a decoding error occurs during decoding of block data and a part of the start code is decoded as block data, the present invention is made. It is an object of the present invention to provide a data decoding device and a data decoding method capable of quickly recovering from a decoding error.

[0015]

A data decoding apparatus according to the present invention returns a bit position to be decoded by the decoding unit by a zero-run count value when the abnormality detecting unit detects a decoding abnormality, and sets a unique bit position from the bit position. A resynchronization means for resynchronizing by searching for a word is provided.

In the data decoding apparatus according to the present invention, when the resynchronization means detects a unique word, the zero count value is reset to zero.

In the data decoding apparatus according to the present invention, the count value of the zero run is incremented only while the decoding means is decoding the variable length codeword.

The data decoding apparatus according to the present invention monitors the leading byte position of digital data and resets the zero count value to zero when the leading byte position is detected.

In the data decoding method according to the present invention, when a decoding error of digital data is detected, the bit position for decoding the digital data is returned by the count of zero consecutive counts, and a unique word is searched from the bit position to resynchronize. It is intended to be.

In the data decoding method according to the present invention, when a unique word is detected, the zero run count value is reset to zero.

In the data decoding method according to the present invention, the count value of the zero run is incremented only during the period of decoding the variable length codeword.

In the data decoding method according to the present invention, the leading byte position of the digital data is monitored, and when the leading byte position is detected, the zero count value is reset to zero.

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a data decoding apparatus according to Embodiment 1 of the present invention. In FIG. 1, reference numeral 1 denotes digital data in which a start code (unique word) and macroblock data (variable length codeword) are multiplexed. A decoder (decoding means) 2 for decoding, when the bit data constituting the digital data is zero bit,
If the bit data is a non-zero bit while incrementing the zero-run count value, a zero-run counter (counting means) 3 that resets the zero-run count value to zero causes the decoding abnormality to occur based on the decoding result of the decoder 1. When the abnormality detection unit 3 outputs the abnormality detection signal, the abnormality detection unit (abnormality detection unit) 4 that detects and outputs an abnormality detection signal returns the bit position decoded by the decoder 1 by the zero consecutive count value. A resynchronization unit (resynchronization means) that searches for a start code from a bit position and resynchronizes
It is.

Next, the operation will be described. In the first embodiment, an error occurs in a video data area immediately before a start code in a bit stream generated by MPEG-1 video compression encoding, and a part of the start code is erroneously decoded as video data. In such a case, a description will be given of a method of detecting the presence of a start code and quickly performing resynchronization.

FIG. 2 is a flowchart showing a data decoding method according to the first embodiment. In the figure, "start" means that the decoder 1 starts analyzing the bit stream. The data in the first embodiment is M
In order to use the decoder for decoding the PEG-1 video bit stream, “data decoding” in step ST1
MPEG-1 defined by O / IEC 111722-2
A process of analyzing a bit stream based on a video decoding method and decoding individual image data will be described.

The "zero run count" in step ST2 is as follows.
This is performed every time each data element according to the MPEG-1 video bitstream syntax is decoded. For example, when a certain data element is decoded as the code word “0010000010”, the process of counting the zero run in the zero run counter 2 is performed as follows. (1) Since the first two bits are zero run, the zero run length is “2” here. (2) Next, since there is “1”, the zero run length is reset to “zero” here. (3) Since zero continues for four bits, the zero string is "4".
Becomes (4) Next, since there is “1”, the zero run length is reset to “zero” here. (5) Next, since there is one bit of zero, the zero run becomes “1”.

At the point when this data element is decoded through the above process, the zero run count becomes "1". This count value becomes an initial value at the time of the zero-run count of the next data element. That is, assuming that the code word “000000010000” is decoded following the above code word, the zero run count value starts from 1 and the first 6
Since the bits are consecutive zeros, it becomes "7" at this point. Since the next bit is "1", it is reset to zero here. As described above, every time one data element is decoded, the count value of zero run is obtained. The above steps (1) to (5) conceptually describe the zero-sequence counting process. In software, a bit operation is performed. In hardware, a counting process is performed on a bit buffer stored in a register or the like. Is carried out.

Step ST3 is a step of detecting an error in the abnormality detecting section 3. Figure 3 shows the detailed processing flow.
Shown in Here, it is assumed that bit errors and error detection are performed as shown in FIG. This causes a bit error in the macroblock data at the end of the t-th picture (* in the figure).
), The effect is that the subsequent codewords will continue to be erroneously decoded and will not be detected as an error.
This shows a case where a part of the start code of the + 1st picture is decoded as data of the tth picture. It is assumed that the start code of the (t + 1) th picture is not committed to a bit error. Also, t + 1
An error is detected in a situation where a non-defined codeword is detected while trying to continue decoding a part of the start code of the t-th picture as data of the t-th picture.
(The position indicated by).

FIG. 5 illustrates this situation in more detail. In the figure, when a bit error occurs at the position of *, as shown in the figure, the codeword to be decoded is different from the codeword that should be decoded correctly, and decoding is performed up to the start code of the t + 1th picture. Would.
Such a situation can always occur in the following decoding states. (1) In the case of a coded MB (macroblock in which DCT coefficient data is coded), decoding of DCT coefficient data is being performed. (2) In the case of not a coded MB, a macro including attribute information (motion vector, mode information, etc.) of the macroblock. During decoding of block header A flow for detecting the possibility of both (step ST11)
To ST13) is FIG. Therefore, when errors are detected during these decoding states, it is necessary to consider the possibility of the situation shown in FIG.

When it is determined in step ST3 that an error has been detected, in step ST4, the resynchronization section 4
Detects whether the bit currently being decoded is included in a part of the start code. This detection is determined based on the zero consecutive count value. That is, in the example of FIG. 5, the zero run count value is "3" at the time when the codeword "01000" is decoded. In other words, if zero continues for 20 bits or more after this, the possibility of a start code is high. As described above, when an error is detected in step ST3, when the zero-run count value is equal to or greater than 1 (Y in step ST4).
In the case of es), it is considered that there is a possibility that a part of the start code is consumed, and the resynchronization unit 4 executes the process of returning the decoding bit position of the decoder 1 by the number of bits of the zero consecutive count value (step ST6). ). Then, a process for searching for a start code again and performing resynchronization is executed (step ST5). The returned zero count value is reset here.

On the other hand, if the zero consecutive count value is "0", the process of step ST6 does not need to be performed for the following reason, and the process of resynchronization is executed as it is (step ST5). Reason 1: Even if there is a start code from the next bit, since the decoding bit position does not go into the start code, it is not overlooked in the resynchronization processing. Reason 2: If there is no start code after the next bit,
Because resynchronization takes place at the closest start code.

The above processing will be described with reference to the example of FIG. 5. In this example, after decoding the code “01000”,
The decoding is performed by considering the long zero run as DCT coefficient data.
Since the first embodiment is the case of MPEG-1,
When 12 bits of zero bits continue, DCT coefficient data is regarded as an unspecified codeword and an error is detected (step ST3). At this time, the zero run count value is
Codeword "01000" decoded without being regarded as an error
Of the start code is “3”, it is determined that a part of the start code has been decoded (step S
T4) Return the decoding position by 3 bits (step ST
6). Then, resynchronization is performed from this position (step ST5). Thus, the start code of the t-th picture is detected by resynchronization, and it is possible to return to normal decoding from the data of the t-th picture.

While the bit stream up to the resynchronization position is skipped during the resynchronization processing, it is not necessary to detect an error in the skipped bit stream, so the zero run count is stopped during the resynchronization processing. May be. Since re-synchronization is performed at the start code of the start code, the zero count is repeated from here (step ST2).
Resume.

As described above, according to the first embodiment, when an error is detected at the time of decoding the macroblock data of the MPEG-1 video bit stream, the error detection part is a part of the next correct start code. Even in the case where the signal has reached the maximum, the decoding can be correctly restored to the normal decoding at the nearest start code position, and quick error recovery can be performed. Also, the first embodiment
Then, the example of the picture start code was given.
Similar processing is possible for the sequence header code, GOP start code, and slice start code that can occur next to the macroblock data among the start codes in FIG. 16 defined by G-1.

Embodiment 2 In the first embodiment, the case where the zero-run counter 2 always performs the zero-run count has been described. However, as shown in FIG. 6, the zero-run count may be performed only while macroblock data is being decoded. (Step ST7). The problem to be solved by the present invention (the case as shown in FIG. 5 described in the first embodiment) occurs only when the variable length code and the start code are continuous. In the MPEG-1 video bit stream, such a case has only a connection point between the macroblock data and the start code. While decoding other than the macroblock data, the zero run count value is always set to zero. Thus, the processing after step ST3 can be shared with the first embodiment. Therefore, in the second embodiment, the number of locations where zero-counting is performed is reduced to reduce the processing amount. Such local processing is particularly effective when the data decoding device is configured by software.

Embodiment 3 In the MPEG-1 video bit stream, all start codes are specified to be byte-aligned (start codes start from the beginning of bytes). According to this provision,
As shown in FIG. 7, when the resynchronization process in step ST10 is configured to detect only the byte-aligned start code, the re-synchronization process may be configured to reset the count of zeros at the beginning of a byte. (Steps ST8 and ST9). As a result, the zero consecutive count value is always the count value from the head of the byte.
The resynchronization process at T10 can securely detect the byte-aligned start code.

Embodiment 4 FIG. In the fourth embodiment, M
In a bit stream generated by PEG-2 video compression encoding, if an error occurs in a video data area immediately before a start code and a part of the start code is erroneously decoded as video data, Describes how to detect presence and quickly resynchronize.

The video data structure of MPEG-2 is shown in FIG.
5 is exactly the same as the MPEG-1 video data structure shown in FIG. However, a picture can be selected from a frame picture and a field picture. Since the type of picture does not directly affect the second embodiment, a frame picture is described as a picture in the following description. FIG. 8 is a flowchart showing a data decoding method according to the fourth embodiment. As is clear from the figure, the basic processing flow is the same as that of the MPEG-1 video decoder according to the first embodiment. In the figure, "start" means that the decoder starts to analyze the bit stream. Since the decoder according to the fourth embodiment is a decoder that decodes an MPEG-2 video bit stream, “data decoding” in step ST21 is performed according to ISO / IEC 138.
A process of analyzing a bit stream based on the MPEG-2 video decoding method specified in 18-2 and decoding individual image data will be described. The “zero run count” in step ST2 is performed every time each data element is decoded according to the MPEG-2 video bit stream syntax. The processing of the zero-run count follows the first embodiment.

Step ST22 is an error detection step. As in the first embodiment, assume a bit error and error detection situation as shown in FIG. In other words, a bit error occurs in the macroblock data at the end of the t-th picture (the position indicated by * in the figure), and due to this effect, the subsequent codewords continue to be erroneously decoded. Shows a case where part of the start code of the picture No. is decoded as data of the t-th picture.

It is assumed that the start code of the (t + 1) th picture is not committed by a bit error. Also,
While trying to continue decoding a part of the start code of the (t + 1) th picture as the data of the tth picture, an error is detected (indicated by X in the figure) in a situation where an unspecified codeword is detected. Shall be. In such a situation, even when a part of the start code is decoded as the macroblock data in exactly the same manner as described in the first embodiment, re-synchronization is correctly started from the start code. Can be.

Note that the resynchronization processing in step ST23 is as follows.
The process proceeds by appropriately setting the start code to be resynchronized according to the start code of MPEG-2 and the order in which it is generated. In the cases of FIGS. 4 and 5, an error is detected at the end of the picture. However, when the error is detected, it is not possible to know that this is the last macroblock of the picture. Detects a high level start code. However, since the extended start code and the user data start code are always defined to appear after the sequence, GOP, and picture headers, they are excluded from the resynchronization target.

As described above, according to the fourth embodiment, when an error is detected at the time of decoding the macroblock data of the MPEG-2 video bit stream, the error detection part is a part of the next correct start code. Even in the case where the signal has reached the maximum, the decoding can be correctly restored to the normal decoding at the nearest start code position, and quick error recovery can be performed.

Since MPEG-2 has a mode for switching whether or not to encode macroblock data in units of macroblocks, similarly to MPEG-1, the DCT coefficient data is encoded using the processing flow of FIG. In a macroblock in which is not present, equivalent error detection is performed in a variable length code of a macroblock header (such as a motion vector and mode information), and quick error recovery can be performed by the method described in the fourth embodiment.

As described in the second embodiment,
Also in the case of MPEG-2, the above error detection situation can occur only at the point where the macroblock data and the start code are connected. Therefore, it is possible to configure so that the zero run count is performed only for the macroblock data (see FIG. 6). Such local processing is particularly effective when the data decoding device is configured by software. Also, in MPEG-2, the start code is defined as being byte-aligned. Therefore, as described in the third embodiment, if the zero-count is reset at the head of the byte, the byte-alignment can be performed. A similar effect can be obtained by using resynchronization processing for resynchronizing the set start code (see FIG. 7).

Further, in the fourth embodiment, an example of a picture start code has been given. However, among the start codes of FIG. 9 specified by MPEG-2, a sequence header code, GOP Similar processing is possible for a start code and a slice start code. The bit pattern of the start code is HEX
It is a display. In each case, 23 bits of 0 and 1 bit of 1
At the beginning, and the type is distinguished by the subsequent 8-bit data.

Embodiment 5 FIG. In the fifth embodiment, M
When an error occurs in a video data area immediately before a start code in a bit stream generated by PEG-4 video compression encoding and a part of the start code is erroneously decoded as video data, Describes how to detect presence and quickly resynchronize.

FIG. 10 shows the video data structure of MPEG-4.
Shown in In MPEG-4, a layer corresponding to an MPEG-1 / 2 picture is a VOP (Video Object).
tPlane), not only a rectangular image area such as MPEG-1 / 2, but also an image area having an arbitrary shape can be used as a unit. The VOP of FIG. 10 shows an example in which the person area is set as the VOP. GOV (Group of VOPs) is a unit obtained by grouping several VOPs.
Call. Also, the unit in which GOVs are collected is VOL (Vid
eo Object Layer). When an arbitrary-shaped area is set as a VOP, the size of the VOP at each time can change every moment according to the movement of the subject, the position on the screen, and the like, as shown in FIG.

Each VOP can be divided in units called video packets. Video packets are MPEG-
MPEG-1 is a structure corresponding to a half slice.
/ 2 is a unit in which a plurality of macroblocks are combined. When the VOP has an arbitrary shape, data for expressing a shape called an alpha block is separately added to the macro block. The alpha block is 8-bit gradation data in which each pixel takes a value from 0 to 255, and has the same size (16 pixels × 16 lines) as the luminance component of the macro block. On the bit stream, the data of the alpha block is multiplexed before the block data.

FIG. 11 is a flowchart showing a data decoding method according to the fifth embodiment. As is apparent from the figure, the basic processing flow is the same as that of the first embodiment.
EG-1 video decoder, MPEG according to the fourth embodiment
-2 Same as video decoder. In the figure, "start" means that the decoder starts to analyze the bit stream. The decoder according to the fifth embodiment has M
In order to use the decoder for decoding the PEG-4 video bit stream, “data decoding” in step ST
SO / IEC JTC1 / SC29 / WG11 / N22
02 (MPEG-4 Visual Final Co)
MPEG-specified by “Mittee Draft”
4 shows a process of analyzing a bit stream based on a 4-video decoding method and decoding individual image data.

Note that ISO / IEC JTC1 / SC2
9 / WG11 / N2202 is an ISO / IEC 14496-2 (MPEG- standard) standardized in December 1998.
4Visual International Sta
ndard), the technical contents will be almost the same, and there is no change in the syntax where the macroblock data and the start code are adjacent to each other. The “zero run count” in step ST2 is performed every time each data element is decoded according to the MPEG-4 video bit stream syntax. The processing of the zero-run count follows the first embodiment.

Step ST25 is an error detection step. Here, a bit error and an error detection situation as shown in FIG. 12 are assumed. That is, a bit error occurs in the macroblock data at the end of the t-th VOP (* in the figure).
), The effect is that the subsequent codewords will continue to be erroneously decoded and will not be detected as an error.
A part of the start code of the + 1st VOP is changed to the tth VOP.
A case where the data is decoded as OP data will be described. t +
It is assumed that the start code of the first VOP is not committed by a bit error. Also, the (t + 1) th VOP
It is assumed that an error is detected (the position indicated by X in the figure) in a situation where an unspecified codeword is detected while trying to continue decoding a part of the start code as the t-th VOP data.

Such a situation can always occur in the following decoding states. (1) In the case of a coded MB (macroblock in which DCT coefficient data is encoded), decoding of DCT coefficient data is being performed. (2) In the case of not a coded MB, attribute information of the macroblock (alpha block data, motion vector, mode information, etc.) During the decoding of the macroblock header including ()), the processing flow of FIG. 3 can be used as a flow for detecting the possibility of both, although the detailed criterion is different. Therefore, if an error is detected during these decoding states, it is necessary to consider the possibility of the situation shown in FIG.
In such a situation, even when a part of the start code is decoded as the macroblock data in exactly the same manner as described in the first embodiment or the fourth embodiment, the start code is correctly read. You can resync from.

The re-synchronization processing in step ST26 is as follows.
The start code for resynchronization is appropriately set in accordance with the start code of MPEG-4 and the generation order thereof, and the process proceeds. In the case of FIG. 12, an error is detected at the end of the VOP. However, at the time when the error is detected, it is not possible to know that this is the last macroblock of the VOP.
Video packet start code (resync_ma)
rker), or a start code with a higher hierarchical level than that is detected.

FIG. 13 shows a start code in MPEG-4. The bit pattern is HEX display. Each of them has 23-bit 0 and 1-bit 1 at the head, and the type is distinguished by 8-bit data thereafter. this house,
The start codes related to the video data itself are a video object layer start code, a GOV start code, a VOP start code, and a user data start code. On the other hand, for the start code of the video packet not described in FIG. 13, the pattern of the unique word is different, and 15 + f_code (f_co
de is a value that defines the range of motion vectors in the VOP and is multiplexed in the VOP header).
Bit 1

As described above, according to the fifth embodiment, when an error is detected at the time of decoding macroblock data of an MPEG-4 video bit stream, the error detection part is a part of the next correct start code. Even in the case where the signal has reached the maximum, the decoding can be correctly restored to the normal decoding at the nearest start code position, and quick error recovery can be performed.

Note that MPEG-4 has a mode for switching whether or not to encode DCT coefficient data in units of macroblocks. Therefore, using the processing flow of FIG. The same error detection is performed in the variable length code of the block header (alpha block data, motion vector, mode information, etc.), and quick error recovery is possible by the method described in the fifth embodiment.

As described in the second embodiment,
Also in the case of MPEG-4, the above error detection situation can occur only at the point where the macroblock data and the start code are connected. Therefore, it is also possible to configure so that the zero run count is performed only on the macroblock data (see FIG. 4). Such local processing is particularly effective when the data decoding device is configured by software.

Further, in MPEG-4, the start code is defined as being byte-aligned. Therefore, as described in the third embodiment, the start code may be reset at the head of the byte. The same effect can be obtained by using resynchronization processing for resynchronizing the byte-aligned start code.

Embodiment 6 FIG. Although the error recovery methods described in the first to fifth embodiments have been described with reference to an MPEG-1 / 2/4 bit stream which is an international standard video encoding method, other error recovery methods having the following properties are described. Applicable to any digital data. (1) Digital data (bit stream) which is hierarchically structured and includes a unique word which can be uniquely decoded and designates the type of layer at the head of each layer (bit stream). Certain digital data (bit stream) (3) Digital data (bit stream) including variable length code in data immediately before unique word

As an example of a bit stream having such a property, other examples of video coding standards for video telephony and video conferencing, such as ITU-T Recommendation H.264. H.261, ITU-T Recommendation H.264 assuming a low bit rate line such as a public analog line network. H.263 (H.263 +) and MPEG having the above-mentioned properties (1) and (2) as in MPEG-4 video
-4 visual object bit stream (still image object, mesh object, face object, etc.). The present invention is applicable to all of these bit streams, and enables quick error recovery.

[0061]

As described above, according to the present invention, when the abnormality detecting means detects a decoding abnormality, the bit position to be decoded by the decoding means is returned by the number of zero consecutive counts, and the unique word is converted from the bit position. Since resynchronization means for searching and resynchronizing is provided, a decoding error occurs when decoding a variable length codeword, and a situation occurs in which a part of a unique word is decoded as a variable length codeword. Also,
Decoding can be restarted correctly from the unique word, and as a result, decoding errors can be quickly recovered.

According to this invention, when the resynchronization means detects the unique word, the zero repetition count value is reset to zero.
There is an effect that the nearest unique word can be searched.

According to the present invention, since the zero-run count value is incremented only during the period when the decoding means is decoding the variable-length codeword, the period when the decoding means is decoding the unique word is set. During this process, the process of incrementing the zero run count value becomes unnecessary, and as a result, the CP
There is an effect that the amount of processing such as U can be reduced.

According to the present invention, the head position of the byte of digital data is monitored, and when the head position of the byte is detected,
Since the zero run count value is configured to be reset to zero, the zero run count value is always incremented from the byte start position, and as a result, the byte-aligned unique word can be detected securely. There is.

According to the present invention, when the decoding error of the digital data is detected, the bit position for decoding the digital data is returned by the count value of the zero run count, and a unique word is searched from the bit position to achieve resynchronization. Therefore, even if a decoding error occurs during decoding of a variable-length code word and a part of the unique word is decoded as a variable-length code word, decoding can be restarted correctly from the unique word. As a result, decoding errors can be quickly recovered.

According to the present invention, when a unique word is detected, the zero-run count value is reset to zero. Therefore, even if a decoding error occurs again, the nearest unique word can be searched. There are effects that can be done.

According to the present invention, the zero-run count value is incremented only during the period in which the variable-length code word is being decoded. Is unnecessary, and as a result, the processing amount of the CPU or the like can be reduced.

According to the present invention, the head position of the digital data is monitored, and when the head position of the digital data is detected,
Since the zero run count value is configured to be reset to zero, the zero run count value is always incremented from the byte start position, and as a result, a byte-aligned unique word can be detected securely. There is.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a data decoding device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a data decoding method according to the first embodiment.

FIG. 3 is a detailed processing flow of step ST3 in FIG. 2;

FIG. 4 is an explanatory diagram showing an error occurrence case.

FIG. 5 is an explanatory diagram illustrating a detailed situation of FIG. 4;

FIG. 6 is a flowchart showing a data decoding method according to the second embodiment.

FIG. 7 is a flowchart showing a data decoding method according to the third embodiment.

FIG. 8 is a flowchart showing a data decoding method according to the fourth embodiment.

FIG. 9 is an explanatory diagram showing types of start codes used in an MPEG-2 video bit stream.

FIG. 10 is an explanatory diagram illustrating an MPEG-4 video data structure.

FIG. 11 is a flowchart showing a data decoding method according to the fifth embodiment.

FIG. 12 is an explanatory diagram showing an error occurrence case.

FIG. 13 is an explanatory diagram showing types of start codes used in a bit stream of an MPEG-4 video & visual object.

FIG. 14 shows MPEG applied to a conventional data decoding method.
FIG. 3 is a data structure diagram showing a structure of a -1 bit stream.

FIG. 15 is an explanatory diagram showing a relationship between data items forming a bit stream.

FIG. 16 is an explanatory diagram showing types of start codes used in an MPEG-1 video bit stream.

FIG. 17 is an explanatory diagram illustrating a state of error detection and resynchronization in an MPEG-1 video bit stream.

FIG. 18 is an explanatory diagram illustrating a case where a part of a start code is erroneously decoded as image data due to a bit error in a macroblock data area of an MPEG-1 video bit stream.

[Explanation of symbols]

1. Decoder (decoding means), 2 zero counter (counting means), 3 abnormality detecting section (abnormality detecting means), 4 resynchronizing section (resynchronizing means).

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Kotaro Asai 2-3-2 Marunouchi, Chiyoda-ku, Tokyo F-term (reference) 5C025 BA25 BA30 DA01 5C059 ME01 RC02 RF01 RF28 UA05 5J065 AA01 AB02 AC02 AE08 AH07

Claims (8)

[Claims] 1. A decoding means for decoding digital data in which a unique word and a variable-length codeword are multiplexed, and, when bit data constituting the digital data is zero bits, incrementing a zero-run count value, If the bit data is a non-zero bit,
Counting means for resetting the zero-sequence count value to zero,
Abnormality detecting means for detecting a decoding error based on a decoding result of the decoding means, and when the error detecting means detects a decoding error, the bit position to be decoded by the decoding means is returned by a zero count value, and And a resynchronizing means for resynchronizing by searching for a unique word from a position. 2. The data decoding apparatus according to claim 1, wherein the counting means resets the zero count value to zero when the resynchronization means detects a unique word. 3. The data decoding apparatus according to claim 1, wherein the counting means increments the zero-run count value only while the decoding means is decoding the variable-length codeword. . 4. The counting means monitors a byte head position of the digital data, and upon detecting the byte head position,
4. The data decoding apparatus according to claim 1, wherein the zero run count value is reset to zero. 5. Decoding digital data in which a unique word and a variable-length codeword are multiplexed, and detecting a decoding error based on the decoding result. When the bit data constituting the digital data is zero bits, , Increment the zero-run count value, and if the bit data is a non-zero bit, reset the zero-run count value to zero and detect an abnormal decoding of digital data.
A data decoding method in which a bit position for decoding digital data is returned by a count value of zero consecutive counts, and a unique word is searched from the bit position for resynchronization. 6. The system according to claim 5, wherein when the unique word is detected, the count value of the zero run is reset to zero.
The data decoding method described in the above. 7. The data decoding method according to claim 5, wherein the count value of the zero run is incremented only during a period in which the variable length codeword is being decoded. 8. The digital camera according to claim 5, wherein a byte head position of the digital data is monitored, and when the byte head position is detected, the zero count value is reset to zero. Data decryption method.