KR20020073961A - coding method based on optimal search pattern of Error Resilient Entropy Coding and Apparatus thereof - Google Patents

Abstract

PURPOSE: A coding method based on the optimum search of error resilient entropy coding and a device thereof are provided to prevent the rapid error propagation in case of decoding by using search patterns of minimum deterioration of screen quality of estimated recovery images. CONSTITUTION: A coding method based on the optimum search of error resilient entropy coding includes the steps of coding an input signal(motion picture coding part;22), storing coded a bit list by error resilient entropy coding unit(temporary bit list storing part:24), determining a bit number generated in each macro block in the error resilient entropy coding units, determining an optimum searching pattern from the determined bit number(searching pattern determining part:26), and reconstructing the stored bit list according to the optimum searching pattern(error resilient entropy coding part:28).

Description

{Coding method based on optimal search pattern of Error Resilient Entropy Coding and Apparatus}

The present invention relates to an error resilient video coding method, and more particularly, to an efficient image coding method and apparatus based on an error processing technique known as error resilient entropy (EREC). .

Error resilient entropy encoding is a method in which the decoder knows in advance the starting positions of important bits necessary for decoding by rearranging the coded bit stream into a data structure called a slot. For example, when the error resilient entropy encoding algorithm is applied to each macroblock (MB) that is a basic unit for motion vector estimation, a transmission error is detected because a new macroblock is always started at the beginning of every slot. In this case, new decoding can be performed in the next slot (or macroblock), thereby enabling fast resynchronization.

This concept was published by David W. Redmill and Nick G. kingsbury in April 1996, by IEEE Trans. It was introduced in a paper entitled "an error resilient technique for coding variable-length blocks of data" published in Image Processing. In the case of entropy encoding, when an error occurs, the error continues to propagate. In order to improve this problem, in order to improve the error, by rearranging a variable length block into a slot of a constant length, an effective isolation of the channel error is realized. Propagation was prevented.

In addition, R. Chandramouli et al., Aug. 1998, IEEE Trans. In the paper entitled "adaptive quantization and fast error-resilient entropy coding for image transmission," published in Circuits and System for Video Technology, The inherent characteristics are used to reduce the complexity of the error recovery entropy coding system, and to obtain better reconstructed picture quality.

This approach is also described in the technical manual on error resilient video coding algorithms proposed by NTTDoCoMo at the MPEG-4 conference held in November 1995 in Dallas, by Toshio Mickey et al. It has been included as an important error recovery technique. Since then, in April 2000 by Kim In-cheol et al. Hierarchical synchronization technique based on the error resilient error coding based on error resilient entropy coding for robust transmission of H.263 bit streams presented in Circuits and System for Video Technology Has been extended.

Figure 1 illustrates a bitstream relocation algorithm of error resilient entropy coding. Referring to FIG. 1, the first step of the algorithm places as many bits as possible in the slot corresponding to each block. Then, as shown, there will be blocks having more bits left to be stored and slots having empty space. After the second step of the algorithm, the remaining bits are placed in the empty space according to a specific search pattern. FIG. 1 illustrates disposing bits remaining after the first step according to a unidirectional search pattern among several search patterns.

When this error resilient entropy coding technique is applied, a very fast resynchronization can be obtained for a transmission error, but at the expense of later information existing in other blocks in which an error does not occur. For example, in the case of decoding the slots in which the rearrangement of the bit string is completed in step 6 of FIG. 1, if an error exists in the slot 2, the data of the slot including the transmission error during decoding is no longer in the block to be decoded yet. Since it cannot be referenced, the first macroblock loses information placed in slots 3 and 4 by error resilient entropy coding.

As described above, when a unidirectional, bidirectional, or fast search mode is used in a system that obtains fast resynchronization by using error resilient entropy coding, a transmission error that has occurred is rapidly propagated as the decoding step proceeds. .

In general, the latter data is known as less important information than the header or motion information such as the discrete cosine transform (DCT) coefficient in the high frequency domain, but the error propagation in the error resilient entropy coding is a decoding algorithm. While performing the step, each step becomes deeper, resulting in deterioration of the quality of the reconstructed image.

In addition, when a unit for performing error resilient entropy encoding is not properly selected, error resilient entropy encoding error propagation due to transmission error is intensified, and sometimes a large amount of information corresponding to a large portion of an image needs to be discarded. there was.

The technical problem to be achieved by the present invention is rapid error propagation that appears during decoding by using a search pattern that minimizes error propagation of error resilient entropy coding and image quality deterioration of the estimated reconstructed image in the process of relocating the coded bit stream. This is to prevent the error and to obtain an optimal reconstructed image.

In addition, when designing a still and moving picture code / decoding system based on an error resilient entropy coding scheme, the present invention improves error propagation characteristics of error resilient entropy coding, thereby obtaining a more robust and superior reconstructed image for transmission error. The present invention provides an encoding / decoding method and apparatus.

In addition, the present invention is to apply a reconstruction image quality deterioration estimation technique of the present invention even in the step of selecting a bit string relocation unit of the error resilient entropy encoding coding method for fast resynchronization to obtain a further improved image quality It is to.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 illustrates a bit string relocation algorithm of error resilient entropy coding.

2 is a block diagram illustrating an encoder and a decoder according to the present invention.

3 is a block diagram illustrating a bit string grammar structure of an encoder according to the present invention.

4 is a diagram illustrating decoding of an I-th macroblock in a k-th search step in an embodiment according to the present invention.

5 illustrates a stepwise rearrangement process of error resilient entropy encoding.

6 is a diagram illustrating a convergence process of an i-th macroblock bit string of error resilient entropy encoding.

<Explanation of symbols for the main parts of the drawings>

22: coder 24,33: bit string storage unit

26: search pattern determination unit 28: entropy coding unit

31: bitstream 32: initial information decoder

34: block state storage unit 35: buffer

37: variable length decoder

In the coder according to the present invention for solving the above technical problem, in addition to the functional blocks of the basic video encoder, a temporary for separately storing the bit stream of each encoded macroblock by an error resilient entropy coding unit A bit string storage unit, a search pattern determiner for determining an optimal search pattern by receiving the number of bits of each macroblock generated in the error resilient entropy coding unit, and bits generated according to the defined optimal search pattern Error recovery entropy code to reconstruct the column.

In addition, the decoder according to the present invention, in addition to the functional blocks of the basic video decoder, the initial information decoder for detecting the basic information for decoding the error resilient entropy encoding and decoding the first error resilient entropy encoding, Temporary bit string storage unit for temporarily storing the transmitted bit strings individually by the size of the slot, Decoded state storage unit for storing the decoded state of the DCT coefficients of each block in the macro block, Storing the number of bits used in each slot And a slot state storage unit for performing the variable length decoding, which can perform repetitive variable length decoding efficiently.

Further, in an encoding method based on error resilient entropy encoding according to the present invention, the method includes: encoding an input signal, storing the encoded bit string as an error resilient entropy coding unit, and recovering the error. Determining the number of bits of occurrence of each macroblock in the sex entropy coding unit, and determining an optimal search pattern from the determined number of bits.

2 shows an encoder 20 and a decoder 30 according to the present invention. The encoder 20 performs error resilient entropy encoding on a video encoder 22 having basic functional blocks for encoding an input original image, and a bit string of each macroblock of an image encoded by the video encoder 22. A temporary bit string storage unit 24 for storing individually by units, for example, in a single picture or GOB unit, and is encoded in the video coder 22 and stored in the temporary bit string storage unit 24. A search pattern determiner 26 for determining an optimal search pattern from bit number information of each macroblock, and an error of reconstructing the bit stream stored in the temporary bit string storage unit 24 according to the determined optimal search pattern. A resilient entropy encoder 28 is included.

In addition to the functional blocks of the basic video decoder, the decoder 30 detects basic information for decoding from the error resilient entropy coded bit stream 31 and performs initial decoding of the first error resilient entropy code. (32), a temporary bit string storage section 33 for temporarily storing the transmitted bit stream individually by the size of the slot, a block state storage section 34 of the DCT coefficients of each block in the macro block, and each slot A slot state storage unit 36 for storing the remaining number of bits used in the and a variable length decoder 37 capable of efficiently performing repeated variable length decoding.

In FIG. 2, the configuration of the macroblocks used in the search pattern determination unit 26 and the error resilient entropy encoder and decoder in the configuration for error resilient entropy encoding and decoding follows the detailed description of the present invention. Is performed according to the prior art.

3 shows a bit string grammar structure of an encoder used in the present invention. The syntactic meaning of the fields used in FIG. 3 is described below.

The data for each picture is composed of a picture header and a group of macroblocks (hereinafter referred to as "GOBs") following the picture header. The picture header is a PSC, NOG x shown in FIG. Fields such as 3, TR, PTYPE, PQUANT, and the like. The meaning of these fields is as follows, and the omitted contents are the same as in H.263.

The PSC is a picture start code, which is comprised of 26 bit words in this embodiment, and is given a value of 1111 1001 1100 1111 1100 1110 01.

NOG × 3 indicates the number of GOBs included in the picture layer. NOG × 3 is a 9-bit word and consists of a NOG consisting of three 3-bit words having the same value.

The TR indicates a time reference and consists of 5 bit words, meaning each bit word is the same as in the picture head of H.263.

PTYPE indicates a picture type, which is composed of 11 bit words, and the meaning of each bit word is the same as in the picture head of H.263.

PQUANT represents quantizer information and consists of 5 bits, each bit word being the same as in the picture head of H.263.

The GFL indicates the group of block frame length and indicates the slot length of each GOB. GOBs are arranged in order from the shortest slot to the longest slot. The GFL for the last GOB represents the binary value of the length of the GOB slot itself, and the GFLs for the other GOBs represent the difference between the slot length of that GOB and the slot length of the next GOB. The code word for the last GFL is 8 bits and the code word for the remaining GFLs is 6 bits. The GFL field is repeated the number of times of NOG. In addition, if the last GFL value exceeds 256 or the remaining GFL values exceed 64, 0 is used as the ESCAPE value. If the GFL field is ESCAPE, there is an LGFL field for the GOB as shown in FIG.

COD is a coded macroblock indication, which is a flag indicating whether the macroblock is encoded, that is, a flag indicating whether the macroblock is encoded or skipped. If the NOG is 1, 2, or 3, the code word length of this field is 2 bits, and if the NOG is more than 3 bits. This field is repeated according to the number of macroblocks (e.g. 99 for QCIF format) and the order is from the top left macroblock (macroblock number = 0) to the bottom right macroblock (macroblock number = 98). )to be. This sequence is the same as in H.263. If the macroblock is not encoded, the COD field is set to 0 (00 for a 2-bit code word, 000 for a 3-bit code word). If the macroblock is encoded, the bits of this codeword indicate the number of the GOB that contains the macroblock. For example, if the COD is 1, this indicates that the macroblock is included in the first GOB.

SP indicates a search pattern, and this field is variable length.

This code has a structure in which the search pattern of ceil (log2 (M)) bits is located when the 1-bit flag is followed by 1, and the loop is repeated by the number of GOBs defined in the NOG field. And M denote the number of macro blocks belonging to a specific GOB determined from the COD field.

There may be a maximum of M-1 search patterns for one GOB, and if the first bit is 0 and there are no SP terms, the error resilient entropy encoding decoding of the corresponding GOB is performed according to a bidirectional search. In addition, even when the number of SP terms for a particular GOB is less than M-1, it is preferable to decode the bidirectional search after the determined last search pattern. For example, if one frame of an image is composed of three GOBs, and each GOB has five, four, and three macro blocks, it is assumed that the bit string of the transmitted SP is 10111001011100. According to the number of macroblocks belonging to each GOB, the first GOB is represented by a 3 bit search pattern following the flag, and the next two GOBs are represented by 2 bits.

Thus, the first 101110010 of the bit string is the search pattern for the first GOB, and then 1110 and 0 are the search patterns for the next two GOBs, respectively. From the bit string, the first two search patterns of the first GOB are 3 and 1, respectively, the first search pattern of the second GOB is 3, and the last GOB performs a bidirectional search. As a result, the search pattern of each GOB is defined as 3, 1, 2, 5, 4 and 3, 4, 2, 1 and 1, 3, 2, respectively.

Once the search pattern is defined, decoding of each GOB is performed in that order, and variable length decoding is repeated until the decoding of all macro blocks of one GOB is completed or until all search patterns are used.

Hereinafter, with reference to FIGS. 1 and 2, a role and a function of the search pattern determination unit of the present invention will be described.

First, consider a binary symmetric channel with a transition probability ε to analyze the error propagation characteristics of error resilient entropy coding. Hereinafter, the length of the j-th macroblock bit string is denoted by b _j , and the length of the i-th slot is denoted by s _i . Decoding of the i-th macroblock is called DMB _i , and e ^k is denoted as an event in which an error occurring within the k-th search occurs.

Therefore, in the k-th step of the i-th macroblock, an event in which a decoding error occurs in error-resilient entropy encoding decoding can be represented by e ^k (DMB _i ), and the analysis of error-resilient entropy encoding error propagation characteristics is searched. It can be achieved by calculating the probability Pr {e ^k (DMB _i ) as a function of the pattern p (k).

In the above case, since p (k) means a search pattern applied during decoding, in order to rearrange the bit string of the i-th macroblock, the i + p (k) -th slot is searched in the k-th search process.

For example, suppose that the i-th macroblock is searching for the (i + p (k) mod N) th slot because the decoding is not completed until the k th decoding step of the error resilient entropy encoding. Here, mod denotes a modulo operation, and N denotes the number of macro blocks belonging to the GOB currently being decoded.

In the process of rearranging the bit string of each macro block during encoding, if the bit string of the i-th macro block is partially stored in the (i + p (k) mod N) th slot through the k th search, the stored i th macro block The bit string of is _denoted by b _{i + p (k)} and the index of the macroblock to which the bit strings stored in the (i + p (k) mod N) th slot originally belongs is _denoted by I _{i + p (k)} .

In the case of decoding the rearranged bit strings, the i-th macroblock has its own bit string b _{i +} recorded in encoding according to the k-th error resilient entropy encoding in the (i + p (k) mod N) th slot. Decoding is performed by referring to _{p (k)} .

Hereinafter, after describing the encoding process step by step, it will be considered a decoding error that can occur in the decoder.

First, in the first step where k = 0, the bit string of each macroblock that is generated is filled as much as can be filled in the slot corresponding to the same index. In this case, if b _i <s _i , an empty space remains inside the slot as a result of block error resilience entropy encoding. If b _i > s _i , an extra bit string not filled in the slot is included in the macroblock. You will gather in turn later.

Therefore, the probability that an error occurs in the first stage of decoding can be considered as follows.

[Equation 1]

If k = 1, when the i-th macroblock is b _i <s _i , decoding of this macroblock is performed regardless of whether an error is in another slot, so Pr {e ¹ (DMB _i ) = Pr {e ⁰ Will be (DMB _i ).

However, when b _i ≥ s _i , information of the (i + p (1) mod N) slot is needed to decode the i-th macroblock, which is the (i + p (1) mod N) slot. It is not limited to errors in b _{i + p (1)} stored in.

We can think of this as two cases: b _{i + p (1)} ≠ That is, considering that the bit sequence of the i-th macroblock is stored in the i + p (1) slot, b _{i + p (1)} must be free of errors in order to ensure that there is no error in decoding of the i-th macroblock. s _{i + p (1)} must have no errors in the bit string written in step 0. This, the s _{i + p,} if ₍₁₎ the bit stream transmitted error recorded in the 0th stage in has occurred, s _{i + p} in ₍₁₎ the 0-th stage decoding of the error is generated b _{i + p (1)} This is because wrong information is passed to the i-th macroblock regardless of an error of.

Similarly, b _{i + p (1)} = That is, when the bit string of the i-th macroblock is not stored in the i + p (1) slot, the data to be transferred for decoding the i-th macroblock does not exist in the i + p (1) -th slot, but 0 If an error occurs in the bit string of another macroblock recorded in the i + p (1) th slot in the first decoding, an incorrect bit string may be transferred to the i-th macroblock decoding. That is, when an error occurs in the bit string of another macroblock recorded in the i + p (1) th slot to be searched, the error may be propagated to the decoding of the i th macroblock. This propagation of errors can be even more severe if the search phase is more than three times (ie k> 1).

Hereinafter, the decoding of the i-th macroblock in the k-th search step will be described with reference to the bit string stored in the i + p (k) -th slot shown in FIG. 4.

Referring to FIG. 4, in the k-th search step, considering decoding of the i-th macro block, it can be seen that data of the j-th macroblock is previously placed between blocks 1 and 3 in the k-1 th search step. have. As in the error propagation analysis in the case of k = 1, the bit string of the i + p (1) -th macroblock present in the i + p (k) -th slot to be searched (shadowed in FIG. 4), The bit strings b _{i + p (k-1)} and b _{i + p (k)} of the j-th macroblock affect the decoding of the i-th macroblock.

Here, the more critical role for the error of the i-th macroblock decoding is the data b _{j + P (k-1)} of the j-th macroblock, for the following reason. From the point of view of the j-th macroblock, the bit string b _{j + p (k-1)} is the last bit information rearranged through k-1 searches (if not the last bit information, b _{j + p (k-1). )} there will be more bits of the j-th macroblock in the back). That is, the decoding of the j-th macroblock starts at s _j and starts with s _{j + p (1),.} . . s _{j + p (k-1)} would have been performed. Therefore, when an error occurs in data included in these slots searched for j-th macroblock decoding, b _{j + p (k-1)} data is not correctly transmitted to decoding of the j-th macroblock, An error also occurs in decoding the i-th macroblock.

In this embodiment, the following assumptions are made to analyze the worst case error resilient entropy coding (hereinafter referred to as "EREC") error propagation characteristics.

1. A macro block in which a decoding error occurs once remains in error even if EREC decoding continues.

2. A slot that has transmitted an incorrect bit string due to a decoding error continues to provide the wrong bit string as the EREC decoding proceeds.

3. If a macro block that has been decoded without a decoding error refers to an error slot and decodes, an error occurs.

4. If a slot that provided a bit string without error is referenced by a block that contains an error, it will provide the wrong bit string and become a faulty slot.

Based on the above assumptions, the propagation characteristics of the error resilient entropy coded decoding error can be expressed by the following equation.

[Equation 2]

Here, e ^{~ k-1} (DMB _i ) represents an event of e ^k-1 (DMB _i ), which means that the i th macroblock is decoded without error until the k-1 th error resilient entropy encoding decoding. And e ^k (h _{i + p (k)} ) indicates an event in which an error occurs in the information provided from (i + p (k)) mod Nth slot in the k th decoding of the i th macroblock. 3].

[Equation 3]

Here, even if (i + p (k)) modN-th slot does not include the bit of the i-th macroblock (that is, b _{i + p (k)} = Note that h _{i + p (k)} is not a null event (an event with probability 0 defined). That is, b _{i + p (k)} = Even though the information h _{i + p (k)} is not null, it is defined as a null event only for the decoded macroblock. For example, whenever b _i ≤ s _i

[Equation 4]

to be.

Hereinafter, a method for estimating deteriorated image quality degradation obtained by a decoding end using the derived error recovery entropy coding error propagation probability of Equation 2 will be described in detail.

In general, when an error occurs in decoding of a specific macro block in one frame, the macro block is restored by an error concealment method. The i-th macro block is _denoted by MB _i , and an image block of the same size to be used for error concealment of this macro block is denoted by EMB _i . Subtract the pixel value of the corresponding EMB _i from each pixel of MB _i , square each of them (or take the absolute value), and add them together to define the error activity of MB _i and call it a _i Let's express it. Then, when a specific search pattern p (k) is used, image quality deterioration that appears in the reconstructed image of the decoding end is expressed as follows.

[Equation 5]

Optimal search pattern determination is performed by selecting one of all possible search patterns through the measure calculation specified in Equation (5). However, if it is determined that the computational complexity caused by the direct calculation of Equation 5 is too high, the following approximation method can be used.

<Calculation method of restoration quality deterioration measure>

(input):

(Output): J (p (k))

Calculate

Calculate

Calculate

In the measure calculation method Is the i-th macroblock of the previous frame decoded without error, and L (h _{i + p (k)} ) is the length of the associated bit string where no transmission error should occur in order for h _{i + p (k)} to be passed without error. , L (b _{i + p (k)} ) denotes the length of b _{i + p (k)} and n _i is defined as a linked bit-list. Also, in calculation 3, >> stands for right shift operation.

As can be deduced from the error propagation characteristics of the above-mentioned error resilient entropy encoding, if the size of each macroblock bit string is significantly different, these macroblocks are gathered into one unit and error resilient entropy encoding is performed. Error propagates rapidly.

Therefore, in configuring the macroblocks of one frame as a criterion for performing error resilience entropy encoding, macroblocks coded in the Intra mode may separately configure GOB to perform bit string relocation, thereby increasing the size of the macroblock bit string. By making it uniform, the problem of error propagating rapidly toward the second half of decoding when the magnitudes are different is eliminated.

In addition, the early part of the bitstream relocation operation can have a much more significant effect than the latter part, so instead of obtaining the optimal pattern among all possible search patterns, the combination of only the first 2-3 search patterns is possible and the rest is bidirectional. A search pattern in which the reconstructed picture quality deterioration measure is minimum can be calculated and used among the search patterns for searching.

5 illustrates a bit rearrangement process for each EREC encoding step. Hereinafter, the EREC error propagation characteristics according to the search pattern will be described in detail with reference to FIG. 5.

First, while performing EREC decoding based on the above analysis, let's calculate the restoration error occurring in each macro block. Consider the following image, which consists of four macro blocks. The transition probability ε of the additive white gaussian noise (AWGN) channel is 0.01, and for the sake of simplicity, the bit stream of each generated macroblock is divided into four slots according to a specific search pattern without considering the internal block. Only look at the process of rearrangement.

When 13, 6, 10, and 3 bits are generated in each macro block as shown in FIG. 5, the slot length of which the total number of bits is 32 will be 8.

Through the one-way search, each generated bit string is rearranged as shown in FIG. 5. In the 0 th step, the bit string generated in each macro block is filled as much as possible to fill the slot corresponding to each macro block, and the remaining bits will be located behind the slot as shown in FIG. At this time, considering the 0-step decoding of each macroblock from the decoding stage, it is possible to calculate the probability that an error occurs in decoding by Equation 1.

That is, for the 0th macroblock, Pr {e ⁰ (DMB ₀ )} = 1- (1-ε) ⁸ = 0.0773, and for the 1st macroblock, Pr {e ⁰ (DMB ₁ )} = 1- ( 1-ε) ⁶ = 0.0585.

Now let's look at the error propagation of the first search phase. In the case of the 0th macroblock, two bits of data 00 are stored in the first slot in the first search. Therefore, in the decoding process through the first search of the 0th macroblock, e ¹ (h _{0 + p (1)} ) causes an error in the data 110010 of the 1st MB recorded in the 1st slot or b _{0 + p (1)} = 00 occurs when an error occurs. In other words, the error does not occur in the 0th decoding for decoding the 0th MB, and the probability that an error occurs in the information received from the 0 + p (1) th slot in the first search may be calculated as follows.

Similarly, the decoding error resulting from the first search of the second MB can also be calculated. However, in the case of the 1st and 3rd MB, the decoding error is maintained since there is no more information to refer to if the error has not occurred in the 0th decoding.

Hereinafter, the case of the second search will be described. First, in decoding through the second search of the 0 th MB, the 0 th MB will search the second slot to see if there is a bit string to be passed. Suppose there is an error in the second slot you are searching for now.

This is not always the case, but the second MB of data may fail and the decoding of the second MB may be ended by only some of the data filled in the second slot.

The characteristic of the error propagation considered in the present embodiment is to consider the worst case, and assume that the wrong bit string (or information) is always transmitted to the decoding of the 0th macroblock when an error occurs in the second slot. Therefore, the probability that an error appears in decoding of the 0th MB by receiving wrong information in the second search may be calculated as follows.

In the case of the second search of the second MB, no error occurs until the first search, and if decoded, there is no information to be passed in the second search, so no further error propagation occurs. Finally, let's calculate the error propagation that occurs in the third search. In the case of the third search, since all other MBs have already been decoded except the 0th MB, no further error propagation occurs.

Hereinafter, the third search of the 0th MB will be described. In the third slot to be searched, the bit string of the third block and the bit string of the second block are already stored, followed by the 0th MB data. Then consider the following event.

Here, the case in square brackets represents all transmission errors included in the second slot and the third slot in the third step of FIG. Denotes an event in which all the data involved in the 0th MB decoding until the 2nd search, that is, no error occurs in the bit strings of the 0th slot, the 1st slot, and the 2nd slot in FIG. 5. Therefore, the probability does not occur in all bits of the 0th, 1st, and 2nd slots, but is calculated as follows as a probability that an error occurs in the data of the 3rd slot.

Looking at the EREC error propagation characteristic described above, the biggest factor influencing the decoding error is performing several searches until the encoded i-th MB bit string converges, and before starting the search in the searched slots. It can be seen whether the MB bit strings are stored in advance.

As described above, the present invention prevents rapid error propagation during decoding by using a search pattern that minimizes error propagation of error resilient entropy encoding and image quality deterioration of the estimated reconstructed image in the process of relocating the coded bit stream. It is effective to obtain an optimal reconstructed image. In addition, even in the step of selecting a bit string relocation unit of the error resilient entropy encoding code scheme for fast resynchronization, a reconstructed image having a further improved image quality may be obtained by applying the reconstructed image quality degradation estimation technique of the present invention. .

Claims (11)

In the coding method based on error resilient entropy coding, Encoding the input signal; Storing the encoded bit string in error resilient entropy coding units; Determining the number of bits of occurrence of each macroblock in the error resilient entropy coding unit; And determining an optimal search pattern from the determined number of bits. The encoding method of claim 1, further comprising reconstructing the stored bit string according to the determined optimal search pattern. The encoding method according to claim 1 or 2, further comprising inserting the determined optimal search pattern information into a bit string. The method of claim 1, wherein the error resilient entropy coding unit is a picture or a group of macroblocks. The encoding method of claim 1, wherein only intra-encoded macroblocks are used as separate error resilient entropy coding units. The method of claim 1, wherein the determining of the optimal search pattern is performed by the following equations (1) to (4). Where ε is the transition probability, b _j is the length of the bit string of the j-th macroblock, s _i is the length of the i-th slot, and e ^k (DMB _i ) is the error recovery entropy decoding of the k-th step of the i-th macroblock. Decoding error occurs at e ^{~ k-1} (DMB _i ) means that the i-th macroblock is decoded without error until the k-1 th error recovery entropy decoding, e ^k (h _{i + p (k )} ) Is the event that an error occurs in the information provided from (i + p (k)) modN-th slot in the k-th decoding of the i-th macroblock, a _i is the error activity of the macroblock, J (p (k) ) Is an image quality deterioration that appears in a reconstructed image of a decoding end when a specific search pattern p (k) is used. In the encoding apparatus based on error resilient entropy encoding, Encoding means for encoding the input signal, Storage means for storing the signal output from the encoding means in an error resilient entropy coding unit; Means for determining the number of bits of each macroblock in the error resilient entropy coding unit stored in the storage means; And means for determining an optimal search pattern from the determined number of bits. 8. The apparatus of claim 7, further comprising means for reconstructing the stored bit strings according to the determined optimal search pattern. 9. The encoding apparatus according to claim 7 or 8, further comprising means for inserting the determined optimal search pattern information into a bit string. The encoding method of claim 7, wherein the error resilient entropy coding unit is a picture or a macroblock group. The encoding method of claim 7, wherein only intra-encoded macroblocks are used as separate error resilient entropy coding units.