KR20050089796A - A method and a device for processing bit symbols generated by a data source, a computer readable medium, a computer program element - Google Patents

Abstract

A method for processing bit symbols generated by a data source, in particular a video, still image or audio source, comprising the following steps of constructing a plurality of bit- planes from the data source, each bit-plane comprising a plurality of bit- plane symbols; scanning the bit-plane symbols of each bit-plane to generate a binary string of bit-plane symbols; and encoding the binary string of the bit-plane symbols using a statistical model, wherein the statistical model is based on statistical properties of a Laplacian probability distribution function which characterizes the data source.

Description

Method and apparatus for processing bit symbols generated by a data source, a computer readable medium, a computer program element

Embedded coding has generated considerable interest in video, image and audio processing. This is because, due to the embedded encoding, the encoding process can be terminated at any point in time so that the encoder meets a predetermined target bit rate. Moreover, the decoder can truncate the bit-stream at any point in time and still obtain appropriate decoded video, image or audio. That is, an ideal embedded coding system is capable of providing truncated bit-streams optimized for rate-distortion, and for this reason, the fine granularity scalability (FGS) of the embedded coding system is fine. It is an ideal coding tool for forming systems with scalability.

A general method of implementing an embedded coding system is due to sequential bit-plane coding (BPC) due to its simplicity. In a BPC, input data vectors from a data source are represented in bit-planes, the bit-planes starting at the most significant bit-plane representing the most significant bits (MSB) of the input data vectors, and thus the input data. It is sequentially coded down to the least significant bit-plane representing the least significant bits (LSB) of the vectors. In addition to its structural simplicity, this coding sequence from the MSB to the LSB of the input data vectors satisfies the principles of the embedded coding process as disclosed in [1], which is most relevant to the quality of the video / image / audio data. Affecting bits must be coded first.

In general, implementing bit-plane encoding that provides an optimized value of the rate-distortion curve is extremely complex and requires significant computational resources. This is because for general data sources there are statistical dependencies between the data samples as well as between the bit-planes. In order to capture these dependencies, the entropy coder must use a frequency table with multiple entries, which not only increases the complexity of the entropy coder, but also incurs a lot of modeling costs [2] that degrade coding performance. Can cause. Therefore, most practical bit-plane encoding implementations usually adopt a compromised approach to reduce computational complexity, which unfortunately results in performance degradation.

Thus, it is desirable to have a bit-plane encoding process that provides an optimized value of the bit rate-distortion curve, with low computational complexity and without causing substantial degradation in performance.

1 is a diagram illustrating a general structure of a video / image / audio encoding system.

2 is a diagram illustrating a general structure of a bit-plane encoding system.

3 is a diagram illustrating a modified structure of bit-plane encoding according to an embodiment of the present invention.

It is an object of the present invention to provide an embedded coding scheme having low computational complexity but comparable performance to any of the systems described above.

This object is achieved by the features of the independent claims. Additional features are derived from the dependent claims.

The present invention relates to a method for processing bit symbols generated by a data source, in particular a video, still image or audio source, wherein each bit-plane is generated using the bit symbols generated by the data source. Constructing a plurality of bit-planes comprising a plurality of bit-plane symbols, scanning the bit-plane symbols of each bit-plane to produce a binary string of bit-plane symbols; And encoding a binary string of bit-plane symbols using a statistical model, wherein the statistical model is based on statistical properties of a Laplace probability distribution function characterizing the data source.

Bit symbols generated by a data source comprising a plurality of input data vectors are first arranged in such a way that a plurality of bit-planes are formed. Each bit-plane corresponds to a respective bit symbol of the data source and includes a plurality of bit-plane symbols.

The data source can refer to any type of data signal that can be captured by the capture device for further processing. In particular, the data source herein refers to a video, still image or audio source that can be captured by a video recorder, camera and microphone, respectively, for further processing.

Starting from a bit-plane, preferably a bit-plane comprising the MSB of the input data vectors, all bit-plane symbols are generated in a specific manner to produce a binary string of bit-plane symbols. Scanned to select. The binary string of bit-plane symbols generated by the scanning process is then encoded using a statistical model. The statistical model is generated based on the statistical properties of the Laplace probability distribution function (pdf) of the data source, in particular the video / image / audio source.

The advantage of using a statistical model based on the statistical properties of Laplace pdf to encode a binary string of bit-plane symbols is that the computational complexity of the encoding process based on this type of statistical model is very low. If the statistical model is based on the statistical properties of a typical pdf, an extremely large probability table needs to be maintained at the encoder, which is not suitable for applications with limited computational resources and storage capacity. To overcome this problem, most BPC schemes of the state of the art entropy encode only a limited subset of bit-plane symbols with a very asymmetric distribution, which results in a significant loss of coding efficiency.

By using the statistical properties of the Laplace pdf of the data source according to the invention, the need for such a large probability table is eliminated, significantly reducing the computational complexity without any substantial loss of quality.

The encoding method according to the present invention uses an entropy encoding process which is a form of a data compression method based on statistical models. Preferably, an operational encoder is used as the entropy encoder for encoding the binary string of bit-plane symbols generated by the scanning process.

Operational coding, i.e., the entropy coding process, is preferred because it provides a good compression rate.

Laplace pdf can be defined using the following functions:

Where σ is the standard deviation or distribution parameter of the Laplace pdf.

According to one embodiment of the invention, the equation of Laplace pdf is used to determine the probability allocation for each of the bit-plane symbols. The determined probability allocation is subsequently used to determine a statistical model for encoding the binary string of the bit-plane symbols.

In particular, the probability allocation for each of the bit-plane symbols is determined using the following equation:

Where P _j is the probability allocation for the bit-plane symbol and j represents the bit-plane.

The probability allocation equation is obtained from the Laplace pdf and used to determine the probability of each bit-plane symbol. This probability or statistical information of the data source is subsequently used by an encoder, in particular an operational encoder, for encoding a binary string of bit-plane symbols.

Due to the statistical properties of the Laplace pdf, the complexity for determining the probability distribution of each bit-plane is significantly reduced.

In another embodiment in which the standard deviation σ is not known, the probability allocation for each bit-plane symbol is determined based on knowledge from encoding previous bit-plane symbols.

This adaptive process is useful in practical applications where knowledge about the statistical properties of the data source is unknown or when the data source is non-fixed. In such cases, the statistical properties of the data source are determined based on information obtained from previously coded bit-plane symbols.

In particular, the probability allocation for each bit-plane symbol in this embodiment is provided by the following equation:

Where P _j is the probability allocation for the bit-plane symbol, N _a is the number of bit-plane symbols encoded until the termination of the previous bit-plane, and N is the current bit-plane symbol The number of bit-plane symbols to be encoded, Is an estimate of P _j after the observation of N _a bit-plane symbols, Is the maximum likelihood estimate of P _j for the current bit-plane and is defined by the following equation.

_Wherein b _{i, j} is the bit-plane symbol.

Preferably, an estimate of P _j from previous coded bit planes ( ) Is estimated by updating from the previous bit-plane using the following equation:

In the above formula, Is an estimate of P _j from the previous bit-plane.

In another embodiment of the present invention, a method for processing bit symbols generated by a data source comprises determining an optimal bit-plane (referred to as a lazy plane) from an input data vector to be encoded, and the lazy Determining a probability allocation for each bit-plane based on each bit-plane relationship to the plane, wherein the probability allocation for the bit-plane encodes a binary string of bit-plane symbols It is used as a statistical model for

In this embodiment, the computational complexity of the encoding process is further reduced since the probability allocation for each bit-plane is explicitly determined by the relationship with the lazy plane.

First, the lazy plane is selected from a plurality of bit-planes. The lazy plane is represented by an integer (L) that satisfies the following inequality:

In the above formula, φ is Defined by

θ is Is defined as

The decision rule actually divides the support of the distribution parameter [sigma] into disaggregated regions, and the lazy plane corresponding to each divided region is specified such that it satisfies the inequality.

After the lazy plane is determined in accordance with the present invention, a probability allocation for each bit-plane is determined. The probability allocation for each bit-plane is based on the relationship of each bit-plane to the optimal bit-plane, as given by the following equation:

In the above formula, Is the probability allocation for the jth bit-plane.

Alternatively, if the length and absolute sum of the input data vectors of the data source are known, the lazy plane can be determined using the following equation:

Wherein N is the length of the input data vector and A is the absolute sum of the input data vector.

The determination of the optimal bit-plane can be implemented by slightly modifying the algorithm as disclosed in [3] to extend the range of orders L to negative integers.

In another variant embodiment, the probability allocation for each bit-plane based on the relationship of each bit-plane to the lazy plane can be determined using the following equation:

In this embodiment, the encoder can be implemented using the asymmetric encoder disclosed in [4].

As mentioned above, the two variants described above have the advantage of further reducing the computational complexity of encoding a binary string of bit-plane symbols.

Moreover, a method is provided for processing an encoded binary string of bit-plane symbols to produce output data representing the data source, wherein the method comprises a plurality of bit-planes comprising the bit-plane symbols. Decoding the encoded binary string of bit-plane symbols to produce an additional binary string of bit-plane symbols to be reconstructed. Since the plurality of bit-planes are reconstructed with a probability assigned by an additional statistical model, output data representing the input data vectors can be reconstructed. The statistical model is based on a Laplace probability distribution function characterized by the bit-plane symbols.

The statistical model generated from the decoding process of the binary string of bit-plane symbols is the same as the statistical model used in the encoding process. That is, the probability allocation P _j or used to form the statistical model in the encoding process ) Is regenerated in the decryption process.

Thus, the plurality of bit-planes are reconstructed using the same statistical model used in the encoding process, which results in the reconstructed until the encoded binary string of bit-plane symbols is terminated by the decoder. Make sure the output data is exactly the same as the original data source.

Moreover, an optimal mean squared error (MSE) reconstruction of the source vectors is generated with a probability assigned by the statistical model. In particular, the probability allocation P _j is used to form the statistical model in the encoding process, and the data source is reconstructed using the following equation:

In the above formula,

Is the reconstructed data source,

s _i is Is the sign symbol for

T is the bit-plane at which the encoded binary string of bit-plane symbols terminates.

Similarly, the probability allocation ( Is used to form the statistical model in the encoding process, the data source is reconstructed using the following equation:

As can be seen above, the second addition ( ) Is used to improve the quality of the reconstructed data source, which can be stopped once the desired quality is achieved.

The described embodiments of the invention apply not only to the above method but also to the apparatus, the computer readable medium and the computer program.

1 is a diagram illustrating a general structure of a video / image / audio encoding system 100. A data source, in particular a video, still image or audio source, is received by the capture device 101. The capture device 101 can be a video recorder, camera or microphone for capturing different types of data sources. The captured data is first converted into a digital signal by an analog-to-digital (A / D) converter 102 for further processing.

Bit symbols of the data source generated in the A / D converter are received by a bit-plane encoding system 103 (to be described in detail later) that includes an encoder unit 104 and a decoder unit 105. The encoder unit 104 encodes the bit symbols and transmits the encoded symbols to the decoder unit 105 over a channel.

The decoder unit 105 decodes the encoded symbols and transmits the decoded symbols to an output device 107, for example a digital television or a digital camera, to be displayed. If the output device 107 is an analog device (e.g. an audio speaker), the digital-to-analog (D / A) converter 106 before the decoded symbols are output to the output device 107, the decoded symbols. Can be used to convert them into analog signals.

2 shows the general structure of a bit-plane encoding system 103 comprising an encoder unit 104 and a decoder unit 105. The encoder unit 104 further includes a bit-plane configuration and scanning unit 110, a first statistical model unit 111, and an entropy encoder 112. The decoder unit 105 further includes an entropy decoder 122, a second statistical model unit 121, and a bit-plane reconstruction unit 120.

At the start of the encoding process, bit symbols 130 are received by the bit-plane configuration and scanning unit 110. The bit symbols 130 include a plurality of input data vectors that can be expressed as Equation 1 for a k-dimensional input data vector.

Wherein x _i is any alphabet Are extracted from independent and identically distributed (iid) random sources.

x _i can also be expressed in binary form:

In the above formula, s _i is a sign symbol represented by Equation 3,

b _{i, j} is the magnitude symbol to be. The binary representation of x _i is also normalized as an integer M that satisfies the following inequality:

When each input data vector of the bit symbols 130 is received by the bit-plane configuration and scanning unit 110, the input data vector has its sign symbol _si and magnitude symbols b _{i, j} ). The sign and magnitude symbols of the input data vectors are arranged to form a plurality of bit-planes, each bit-plane being a sign symbol (s _i ) or a magnitude symbol (b _{i, j} ) from each input data vector. It includes. In general, magnitude symbols b _{i, j} corresponding to the most significant bit MSB of the input data vectors are arranged in a first bit-plane, and magnitude symbols b _{i, j} of the second MSB are Arranged on the second bit-plane, other size symbols are likewise arranged in a manner. In addition, the code symbols of input data vector _(i s) is another separate bit-plane are arranged on. All sign and magnitude symbols of the bit-planes are referred to as bit-plane symbols.

Once the bit-planes are configured, all the bit-plane symbols included in the bit-planes are scanned starting at the bit-plane containing the MSB of the input data vectors. The scanning process is to select the bit-plane symbols to form a binary string 131 of bit-plane symbols. One possible scanning process is summarized in the following steps:

1. Start scanning at bit-plane j = M-1 including the MSB of the input data vectors.

2. Select the magnitude symbols b _{i, j} with the corresponding magnitude symbols of all previous bit-planes that are zero: b _{i, M-1} = b _{i, M-2} = ... = b _{i , j + 1} = 0.

3. If the magnitude symbol b _{i, j} is " 1 ", the sign symbol _si is also selected. Steps 2) and 3) are known as critical passes.

4. Select unselected size symbols b _{i, j} in the critical path. This step is known as a refinement pass.

5. Proceed to next bit-plane j-1.

The above steps are repeated up to some termination criterion, for example when a predefined bit rate is met or a predefined bit rate-distortion limit is reached.

Once a binary string of bit-plane symbols 131 is generated by the scanning process, the binary string is additionally coded or compressed at entropy encoder 112. Statistical properties 132 of the bit-plane symbols of the data source 130 to provide a probability allocation 133 that is used by the entropy encoder 112 to encode a binary string of bit-plane symbols 131. It is used in the statistical model 111.

Encoded data 134 from the entropy encoder 112 is transmitted over the channel and then received and decoded by the entropy decoder 122. The channel may be an internet network, a wide area network (WAN), or a wireless communication network.

The entropy decoder 122 receives the encoded data 134 and decodes it into a binary string of bit-plane symbols 135. In theory, the binary string of the bit-plane symbols 135 generated by the entropy decoder 122 is the same as the binary string of the bit-plane symbols 131.

The statistics of the bit-planes 137 are used by the statistical model 121 to generate the same probability allocation 136 as the reference number 133 so that the bit-plane symbols can be correctly decoded. The bit-plane symbols 135 are then used to reconstruct the bit-planes by the bit-plane reconstruction unit 120 to produce output data 138 representing the bit symbols 103 of the data source. do.

If optimal MSE reconstruction is desired, probability allocation 136 is also used to regenerate output data 138 by reference numeral 120.

It should be noted here that the entropy encoder 112 needs to encode the bit-plane symbols to obtain an optimal compression of the binary string of the bit-plane symbols 131 of the data source with a general probability distribution function. The number of bits to be given is given by -log ₂ Pr (s _i , b _{i, M-1} , ...), in which case the probability (Pr (s _i , b _{i, M-1} , ...) )) May be expressed as in Equation 5.

In the above formula, _Denotes a conditional probability of b _{i, Mj} for previously coded bit-planes.

Indeed, implementing such an entropy encoder that encodes all bit-plane symbols of a data source generally requires a frequency / probability table with multiple entries. For encoding at high bit rates, the number of entries to be kept in this frequency table is very large and therefore not practical in systems with particularly limited computation and storage capabilities. In addition, this can introduce significant modeling costs [2] for adaptive setup for data sources with unknown distributions. Thus, a simplified approach is adopted for most practical systems. In most practical systems, as mentioned in [5] and [6], only bit-plane symbols (symbols scanned in the critical pass) with very asymmetric distributions are encoded by the entropy encoder. do.

According to the present invention, the properties of the Laplace probability distribution function (pdf) inherent in most data sources, in particular video, still image and audio sources, for bit-plane encoding are determined by the entropy encoder 112. Used for encoding of data sources.

In particular, the statistical model 111 uses the statistical properties of Laplace pdf of the data source to generate the probability allocation 133 for encoding the binary string of bit-plane symbols 132. Laplace pdf of the data source can be expressed using Equation 6.

In the above, σ is the standard deviation or distribution parameter of Laplace pdf.

It can be easily verified from Equation 6 that the bit-plane symbols of the Laplace source have the following independent attributes.

The probability allocation of the entropy encoder for each bit-plane j is given by equations (7) through (9).

From Equation 6, P _j can be calculated as Equation 10.

If the distribution parameter (σ or standard deviation) of Laplace pdf is known, P _j can be determined directly using Equation 10.

If P _j is determined, the probability of each bit-plane symbols can be determined using equations (7) through (9) and this statistical information of the data source is determined by the entropy encoder 112 and the bit-plane symbols 131. Used to encode a binary string of.

As can be seen above, by using the statistical properties of the Laplace pdf of the data source, the maintenance of a large frequency table according to the prior art is not necessary, thus greatly simplifying the encoding process of the binary string 131 by the entropy encoder 112. It is.

In another embodiment of the present invention, the probability allocation P _j for each bit-plane symbol determined in Equation 10 is used to regenerate the binary bit-plane symbols 135. Binary bit-plane symbols 135 are received by bit-plane reconstruction unit 120 to generate output data 138 representing bit symbols of data source 130.

In particular, when optimal MSE reconstruction is required, when decoding up to the bit-plane T of the encoded data 134 through the entropy decoder 122, optimal reproduction of the output data 138 according to the present invention Is given by equation (11).

First addition ( Is the reconstruction of the bit-plane symbols, and the second addition ) Is the interpolation of the corresponding bit-plane symbols for Laplace pdf.

The second addition can be terminated if a predefined criterion is met, for example if the quality of the desired data source has been obtained.

In a variant embodiment of the invention, the probability allocation P _{j for} bit-plane symbols is adaptively determined based on knowledge from the coding of previous bit-plane symbols. This adaptive bit-plane coding (ABPC) process is useful when the distribution parameter σ of Laplace pdf is not known as in most practical situations.

Given a string of k individual symbols, starting from Lidston's law of success, if the i th symbol occurs n _i times in the past n instances, then the probability estimate of the i th symbol occurrence is given by Is estimated.

Wherein lambda is a positive parameter. Rewriting Equation 12 by substituting Equation 14 into Equation 13 indicates that Equation 12 is an interpolation of the maximum likelihood estimate (n _i / n) and uniform prior (1 / k).

Applying equation (13) to the embodiment of the present invention, equation (15) is given.

Where N _a is the number of bit-plane symbols encoded until the end of the previous bit-plane, N is the number of bit-plane symbols encoded in the current bit-plane symbol, and P _j ^Na is N _a is an estimate of P _j after the observation of the bit-plane symbols, P _j ^ML is the maximum likelihood estimate of P _j for the current bit-plane, and μ provides interpolation between the two probability estimates. Preferably, μ is given by equation (16).

The maximum likelihood estimate of P _j for the N symbols (b _{i, j} ) in the j th bit-plane is given by:

Since P _j ^ML can be defined by Equation 18,

Preferably, from equation (10), P _j ^Na is represented by the previous bit-plane ( ) Can be updated.

The embodiments mentioned so far relate to finding probability allocations for all bit-plane symbols of a data source.

In another embodiment of the present invention, if the probability allocation for all the bit-planes used by the entropy encoder to encode the binary string of bit-plane symbols is determined from the statistics of the data vector to be encoded, the " 2-pass ( a two-pass "approach is adopted.

In this embodiment, the optimal bit-plane is selected from a plurality of individual bit-planes, which is referred to as a lazy plane. Information about the selected lazy plane is transmitted by the encoder unit 104 to the decoder unit 105, so that the encoded data 134 can be correctly decoded.

3 illustrates a modified general structure of a bit-plane encoding system according to an embodiment of the present invention.

Information about the selected lazy plane included in the encoded data 134 is received by the statistical model unit 121. Statistical model unit 121 generates a probability allocation 136 received by entropy decoder 122, so that bit-plane symbols of encoded data 134 can be correctly decoded. The decoded bit-plane symbols 135 are then received by bit-plane reconstruction unit 120 to generate output data 138 representing bit symbols 130 of the data source. Rebuild the planes.

Consider the code family provided by (21).

Where G ^L represents the bit-plane symbols of the data source and L is an integer representing the lazy plane.

The probability allocation according to the embodiment of the present invention is given by Equation 22.

In the above, Q _j ^L follows the probability update rule defined by equation (19) for bit-planes (i≥L) and coding for probability allocation of (½) for bit-planes (i <L). Is achieved by outputting the input symbols directly to the encoded binary string), which is the probability allocation for the jth bit-plane entering the "Lazy mode". This code family C may be referred to as a Bit-Plane Golomb Code (BPGC).

The lazy plane L can be obtained by finding the integer value for L that best satisfies the following inequality.

In the above formula, L is an integer representing the optimal bit-plane, φ is Is defined by Equation (24).

If sufficient statistics are known, such as the absolute sum and length of the input data vectors, the decision rule of equation (23) can be further simplified to

Wherein N is the length of the input data vector and A is the absolute sum of the input data vectors.

The selection process as mentioned in this embodiment can be implemented using the algorithm mentioned by [3]. When the algorithm of [3] is used to determine the value of L, only a positive integer range of L can be determined. In order to extend the range of orders L to negative integers, the algorithm mentioned by [3] is modified.

In particular, the modified algorithm of [3] is given as follows.

if (N <= A)

for (L = 1; (N << (L + 1)) <A; L ++)

else

for (L = -1; (N >> (-L))> = A; L--)

When the lazy plane L is determined, the probability allocation for the bit-plane used for encoding the binary string of bit-plane symbols may be determined by the entropy encoder.

In another variant of the invention, the probability allocation for each bit-plane based on the relationship of each bit-plane to the optimal bit-plane is determined using equation (26).

Where L is an integer representing the optimal bit-plane that can be determined by equation (23).

In this embodiment, the probability assignment given by Equation 26 enables the use of a skew coder as mentioned in [4] as a very low complexity implementation of the entropy coder instead of the normal operational coder. The skew encoder of [4] entropy with only a few bit-shift and addition operations by limiting the probability interval width corresponding to the least probable symbol (LPS) to a power of two. The coding process can be simplified. In addition, the skew encoder of [4] retains its unique simplicity in implementing an acceleration technique [6] that encodes the most probable symbols (MPS) typical for encoding bit-planes with high probability skew. do.

It should be noted here that in all the embodiments of the invention mentioned above except for the embodiment generating the BPSC, the operational coder should preferably be used as the entropy coder.

In an additional embodiment of the embodiments of the present invention mentioned above, the probability allocation Q _j ^L for each bit-plane symbol determined in Eq. 22 or Eq. 26 determines the bit symbols 130 of the data source. The output data 138 representing is used to generate via the bit-plane reconstruction unit 120.

In particular, when decoding up to the bit-plane T of the data 134 transmitted through the entropy decoder 122, the optimal reproducibility of the output data 138 according to the present invention is given by equation (27).

Like Equation 11, the first addition ( Is the reconstruction of the bit-plane symbols, and the second addition ) Is the interpolation of the corresponding bit-plane symbol of the output data 138 of Laplace pdf.

The second addition can be terminated if a predefined criterion is met, for example if the quality of the desired data source is obtained.

The above-mentioned embodiments of the present invention apply not only to a method but also to an apparatus, a computer readable medium, and a computer program.

While embodiments of the present invention have been described so far, embodiments of the present invention merely illustrate the principles of the present invention. Other embodiments and configurations may be devised without departing from the spirit of the invention and the appended claims.

The following documents are cited herein.

[1] J. Li and S. Lie, "An embedded still image coder with rate-distortion optimization", IEEE Trans. Image processing, vol. 9, pp. 1158-1170, July 2000.

[2] J. Rissanen, Stochastic Complexity in Statistical Inquiry, London, UK: World Scientific, 1989.

[3] M.J. Weinberger et al., "The LOCO-I lossless image compression algorithm: principles and standardization into JPEG-LS", IEEE Trans. Image processing, vol. 9, pp. 1309-1324, August 2000.

[4] G.G. Langdon and J. Rissanen, "A simple general binary source code", IEEE Trans. Information Theory, vol. 28, pp. 800-803, 1982.

[5] D. Taubman and A. Zakhor, "Multirate 3-D subband coding of video", IEEE Trans. Image processing, vol. 3, pp. 572-588, September 1994.

[6] E. Ordentlich et al., “A low-complexity modeling approach for embedded coding of wavelet coefficients”, HP Labs Tech. Report, HPL-97-150, 1997.

Claims (20)

A method of processing bit symbols generated by a data source, in particular a video, still image or audio source, Using the bit symbols generated by the data source to construct a plurality of bit-planes, each bit-plane including a plurality of bit-plane symbols; Scanning the bit-plane symbols of each bit-plane to produce a binary string of bit-plane symbols; And Encoding a binary string of the bit-plane symbols using a statistical model, And wherein said statistical model is based on statistical properties of a Laplace probability distribution function characterizing said data source. The method of claim 1, Encoding of the binary string of bit-plane symbols is performed by an entropy encoder. The method of claim 2, And an arithmetic encoder is used as the entropy encoder. The method according to any one of claims 1 to 3, The Laplace probability distribution function is And defined by Is a standard deviation of the Laplace probability distribution function. The method of claim 4, wherein The probability allocation for each bit-plane symbol is determined based on the Laplace probability distribution function and used to determine a statistical model for encoding a binary string of bit-plane symbols. . The method of claim 5, Probability allocation for the bit-plane symbol, Determined by the above formula, Is a probability allocation for the bit-plane symbol, Is a bit-plane. The method of claim 4, wherein The probability allocation for each bit-plane symbol is determined based on previously encoded bit-plane symbols. The method of claim 7, wherein Probability allocation for the bit-plane symbol, Determined by the above formula, Is the probability allocation for the current bit-plane symbol, Is a bit-plane, Is the number of bit-plane symbols that are coded up to the end of the previous bit-plane, Is the number of bit-plane symbols encoded up to the current bit-plane symbol, Is After observation of bit-plane symbols Is an estimate of, For the current bit-plane Maximum likelihood estimate of Defined by the above formula, Is a bit-plane symbol. The method of claim 8, remind After observation of bit-plane symbols Estimate of ), Updated by, From the previous bit-plane It is an estimate of the method of bit symbols characterized in that. The method of claim 4, wherein The method of processing the bit symbols, Determining an optimal bit-plane from the plurality of configured bit-planes; And Determining a probability allocation for each bit-plane based on the relationship of each bit-plane to the optimal bit-plane, The probability allocation to the bit-plane is used as a statistical model for encoding a binary string of bit-plane symbols. The method of claim 10, The optimal bit-plane, Is determined by determining an integer that best satisfies Is an integer representing the optimal bit-plane, Is Defined by Is, A method of processing bit symbols, characterized in that it is defined as. The method of claim 11, Probability allocation for the bit-plane, Determined by the above formula, silver And a probability allocation of the first bit-plane. The method of claim 11, Probability allocation for the bit-plane, Determined by the above formula, silver And a probability allocation of the first bit-plane. The method according to claim 6 or 8, The method of processing the bit symbols, Decoding the encoded binary string of bit-plane symbols using an additional statistical model to produce an additional binary string of bit-plane symbols; And Reconstructing a plurality of bit-planes containing the bit-plane symbols using the additional binary string of the bit-plane symbols, And wherein said additional statistical model is based on statistical properties of a Laplace probability distribution function characterized by bit-plane symbols of said reconstructed bit-planes. The method of claim 14, The data source, Reconstructed from the bit-planes by Is the reconstructed data source, Is Is the sign symbol for Is a bit-plane symbol, Is a coded binary string of bit-plane symbols is a terminated bit-plane. The method according to claim 12 or 13, The method of processing the bit symbols, Decoding the encoded binary string of bit-plane symbols using an additional statistical model to produce an additional binary string of bit-plane symbols; And Reconstructing a plurality of bit-planes containing the bit-plane symbols using the additional binary string of the bit-plane symbols, And wherein said additional statistical model is based on statistical properties of a Laplace probability distribution function characterized by bit-plane symbols of said reconstructed bit-planes. The method of claim 16, The data source, Reconstructed from the bit-planes by Is the reconstructed data source, Is Is the sign symbol for Is a bit-plane symbol, Is a coded binary string of bit-plane symbols is a terminated bit-plane. Apparatus for processing bit symbols generated by a data source, in particular a video, still image or audio source, A bit-plane construction unit, each bit-plane comprising a plurality of bit-planes comprising a plurality of bit-plane symbols, from the data source, the bit-plane of each bit-plane symbol to generate a binary string of bit-plane symbols. A bit-plane configuration unit for scanning bit-plane symbols; A statistical model unit for providing statistical information based on statistical attributes of the Laplace probability distribution function characterizing the data source; And And a coding unit for encoding a binary string of bit-plane symbols based on the statistical information provided by the statistical model unit. A computer readable medium having a program recorded thereon, wherein the program causes a computer to perform a procedure for processing bit symbols with a data source. The procedure for processing bit symbols with the data source, Using the bit symbols generated by the data source to construct a plurality of bit-planes, each bit-plane including a plurality of bit-plane symbols; Scanning the bit-plane symbols of each bit-plane to produce a binary string of bit-plane symbols; And Encoding a binary string of the bit-plane symbols using a statistical model, And the statistical model is based on statistical properties of a Laplace probability distribution function characterizing the data source. A computer program element for causing a computer to perform a procedure for processing bit symbols generated by a data source, The procedure for processing the bit symbols generated by the data source, Using the bit symbols generated by the data source to construct a plurality of bit-planes, each bit-plane including a plurality of bit-plane symbols; Scanning the bit-plane symbols of each bit-plane to produce a binary string of bit-plane symbols; And Encoding a binary string of the bit-plane symbols using a statistical model, The statistical model is based on statistical properties of the Laplace probability distribution function characterizing the data source, the data source in the form of a Laplace probability distribution function.