KR20060032944A - Method and apparatus for coding point sequences on laser binary representation - Google Patents

Abstract

The present invention proposes a hybrid encoding method for a point sequence in order to maximize the bit efficiency of the LASeR binary representation.

The point sequence encoding method according to the present invention includes selecting Exponential Golomb (EG) encoding or Fixed Length Coding (FLC) encoding for each point sequence, and if FLC encoding is selected, Encoding a point sequence according to an encoding method, calculating a parameter k that can most effectively perform EG encoding when EG encoding is selected, and encoding each point sequence according to the EG encoding method using the calculated k . Steps.

Lightweight Application Scene Representation (LASeR), Point Sequence, Fixed Length Coding (FLC), and Exponential Golomb (EG) Coding

Description

Method and apparatus for coding point sequences on LASeR binary representation

1 (a) and 1 (b) show graphs of an occurrence distribution function and a probability density function, respectively, for each of the five test files of the map group.

2 (a) and 2 (b) show graphs of occurrence distribution functions and probability density functions, respectively, for each of the six test files of the comic group.

3 (a) and 3 (b) show graphs of the occurrence distribution function and the probability density function, respectively, for each of the three test files of the vector graphics group.

4 shows a flowchart of a process for encoding a point sequence in accordance with the present invention.

The present invention relates to a method and apparatus for efficiently encoding a point sequence, and more particularly, to a method and apparatus for efficiently encoding a point sequence of a Lightweight Application Scene Representation (LASeR) binary representation.

Lightweight Application Scene Representation (LASeR) is a multimedia content format defined for providing a simple multimedia service in a terminal lacking resources such as a mobile phone. The motivation of LASeR is to apply MPEG-4 technology to mobile environments. Consider applying LASeR to images such as maps, animations, and two-dimensional vector graphics. Many of these image data are characterized by consisting of point data. Therefore, a method of efficiently encoding point data needs to be devised. Two major elements of LASeR to consider for this are efficient binary representation and small memory on the decoder.

The LASeR Text of ISO / IEC 14496-20 CD, published in July 2004, proposes a fixed length coding (FLC) method for point data encoding of LASeR binary representation. The FLC method encodes the actual point data when the number of points (nbPoint) is less than 3, and examines the entire point sequence to determine the dynamic range of the point sequence when the number of points is more than 3, and uses the resultant fixed length. By encoding. This method can be implemented very simply, but there are not only 10 bits of overhead to specify the length field for each point sequence, but there are also a number of bits unnecessarily allocated to subsequent data fields.

Entropy coding can be considered for efficient compression of data in images. In general, entropy encoding is a method of encoding common values using a short number of bits in consideration of various values that data can take. Entropy coding consists of modeling that estimates the probability of each symbol and encoding that yields a bit sequence using this probability. Accurate probability estimation is necessary to achieve good compression ratios.

Although there are various entropy encoding methods, it can be broadly classified into two groups: an encoding method using an encoding table and an encoding method not using an encoding table. The Huffman coding method represents a group using an encoding table. This method can yield a near optimal compression ratio, but the overhead of having to transmit the encoding table and accessing the memory location each time the point data is decoded is increased on the decoder side (terminal). have. Since LASeR requires a lightweight memory and minimal complexity, the Huffman coding scheme using the coding table is not suitable for encoding point data.

As another group of entropy coding schemes (that is, coding schemes not using coding tables), arithmetic coding and EG coding can be given. Although arithmetic coding is a fairly efficient coding scheme, it is also difficult to be used for LASeR due to the lack of error resilience.

On the other hand, EG (Exponential Golomb) coding has some characteristics suitable for LASeR environment. This method can select a parameter k that is suitable for a particular distribution, with little overhead on the encoder side. It can also be easily converted to Reversible Variable Length Coding (VLC) for the addition of error recovery (ISO / IEC JTC 1 / SC 29 / WG 1 (ITU-T SG8), "Reversible VLC for added error resilience "). Another advantage is the low overhead on the decoder side. Since the decryption process can be done using only addition and bit shifting, this method can be implemented with no large overhead on small devices such as mobile phones.

Accordingly, an object of the present invention is to provide a hybrid encoding method for a point sequence in order to improve the compression efficiency of the LASeR binary representation.

Another object of the present invention is to provide an encoding method that improves the compression efficiency of the LASeR binary representation and does not impose a large overhead on the decoder by using the EG encoding method for encoding the point data.

According to an aspect of the present invention, a point sequence encoding method includes selecting Exponential Golomb (EG) encoding or Fixed Length Coding (FLC) encoding for each point sequence based on a selected criterion. And if the FLC encoding is selected, encoding the point sequence according to the FLC encoding scheme, and if the EG encoding is selected, obtaining a parameter k that can perform the EG encoding most effectively, and using each parameter k , points Encoding the sequence according to the EG encoding scheme.

An apparatus for encoding a point sequence according to another aspect of the present invention, for each point sequence, Exponential Golomb (EG) encoding or Fixed Length Coding (FLC) encoding based on a predetermined criterion Means for selecting, means for encoding a point sequence according to the FLC encoding scheme if FLC encoding is selected, and parameter k for performing EG encoding most effectively, if EG encoding is selected, and the parameter k Means for encoding each point sequence in accordance with the EG encoding scheme.

Hereinafter, with reference to the embodiments shown in the accompanying drawings, the present invention will be described in detail by way of example. However, the following detailed description is provided for illustrative purposes only and should not be construed as limiting the inventive concept to any particular physical configuration.

First, FIGS. 1 to 3 show graphs of occurrence distribution functions and probability density functions of test files to which the encoding method according to the present invention is applied.

Table 1 below lists the application files available on sites known in the ISO / IEC “LaSeR Core Experimentation on LASeR Binary Representation” document published July 2004 and the number of points included in each of them.                     

Among the above-mentioned files, several test files including a large number of point numbers were selected for the performance test of the coding scheme according to the present invention. For convenience, the test files are grouped as "map", "cartoon" and "vector graphics" group. Classified as The "Map" group includes Africa, Central America, South America, Europe, and USA map files, and Figures 1 (a) and 1 (b) show the occurrence distribution function for each of the five test files of the map group. And probability density function graphs, respectively. The "cartoon" group includes six test files of bonhom, cc001, kangaroo, marche, karate and rhino, and Figures 2 (a) and 2 (b) show the occurrence distribution function for each of the six test files of the comic group. And probability density function graphs, respectively. Finally, the "vector graphics" group contains three files, desk, face_frame1 and tiger, while Figures 3 (a) and 3 (b) show the occurrence distribution function and probability density function for each of the three test files of the comic group. Each graph is shown.

From the graphs shown in Figures 1 to 3, it can be seen that the shape of the generation distribution function of the test file generally follows the geometric distribution, and the probability density function follows the geometric density function (GDF). In other words, the lower the code number, the more it occurs, and as the code number increases, the number of occurrences decreases.

In general, EG coding is known to yield very high compression rates when the probability density function follows the geometric density function. In other words, the ideal probability density function for EG coding has a very high probability at code number 0 and decreases rapidly as the code number increases.

Although most PDFs of the test files generally follow the ideal PDF format, the probability at code number 0 is not very high and their distribution is much wider than the ideal PDF distribution for EG. In this case, a better compression ratio may be calculated by adjusting the parameter k in accordance with the present invention.

It will be described below that the compression rate can be improved by applying the coding scheme according to the present invention to the three groups of test files described so far.

4 shows a flowchart of a process for encoding a point sequence in accordance with the present invention. As shown, in step 410, it is first determined whether to use EG encoding or FLC encoding for each point sequence. If FLC encoding is selected, the point sequence is encoded in the manner proposed in the Commission Draft of LASeR (step 440). The FLC encoding method may be performed according to the following procedure.

Suppose a point sequence consists of n + 1 points of (x0, y0) (x1, y1) ... (xn, yn),

1) If there are two or less points, record the value of (x0, y0) (x1, y1) itself.

2) If not,

i) Read x0 and y0 to get the number of bits needed for writing

ii) Record x0 and y0.

iii) From (x1, y1) to (xn, yn), read the dynamic range of x and y.

iv) Record dx and dy in order for n points after (x0, y0).

On the other hand, if EG encoding is selected, a parameter k that can most effectively perform EG encoding is obtained (step 420), and then each point sequence is encoded according to the EG encoding method using the obtained k (step 430).

EG encoding is one of entropy encoding schemes, and has a regular configuration scheme that favors small numbers by assigning shorter codes to smaller numbers. Another Huffman coding scheme of the entropy coding scheme requires a code table that specifies a relationship between a symbol and a code number, and the code table must be stored in a user terminal. In contrast, EG encoding is more suitable for LASeR in that EG coding does not require a code table because it uses a regular configuration. EG encoding does not assign codewords according to the exact statistics of the symbols. Instead, EG encoding can be matched to various variances of the geometric distribution by adjusting the parameter k . In EG encoding, each code may be configured as follows.

[M zeros [1] [INFO]]

Where M is the number of leading zeros, and INFO is the surface offset value of the (m + k) bits that carry the information. Leading zeros and the next "1" serve as a prefix code that separates each code. The code number CodeNum is determined as follows.

Since the INFO value does not affect the length of the leading zero, M can be obtained according to Equation 2 by ignoring the INFO term in the above equation.

Next, INFO can be obtained according to the following equation (3) derived from equation (1) above.

In one embodiment, the EG encoding scheme may be performed according to the following procedure. Suppose a point sequence consists of n + 1 points of (x0, y0) (x1, y1) ... (xn, yn),

1) If there are two or less points, record the value of (x0, y0) (x1, y1) itself.

2) If not,

i) Read x0 and y0 to get the number of bits needed for writing

ii) Record x0 and y0.

iii) Record k value in 4-bit space.

iv) for each dx and dy of n points after (x0, y0)

① Obtain M and INFO according to k value.

② Code is generated by recording M number of “0”, “1” and INFO in order.

Table 2 below exemplarily shows 11 EG coding codes configured when parameters k = 0, k = 1, k = 2, and k = 3.

From the table it can be seen that the codes are in logical order.

Since EG codes only have unsigned code numbers, signed EG codes must be mapped to unsigned EG codes. Signed EG codes are parsed by retrieving an unsigned EG code number ( CodeNum ) from the bit stream and mapping it to the signed using the following rules.

If ( CodeNum is 0) signed code = 0;

else if ( CodeNum is even) signed code = -CodeNum / 2;

else if ( CodeNum is odd) signed code = ( CodeNum +1) / 2;

An example of an unsigned-signed mapping is shown in the table below.

Unsigned EG code number 0 One 2 3 4 5 6 7 8 9 … Signed EG code number 0 One -One 2 -2 3 -3 4 -4 5 …

The encoding process of the point sequence according to the present invention may be expressed by the following syntax and semantics.

Syntax:

uint GetCodeNum (int diff) {

if (diff> 0) return diff * 2-1;

else return diff * (-2)

}

uint GetMvalue (uint CodeNum, uint (4) kvalue) {

int i = | CodeNum + 2 ^kvaule |

    int j = 0;

    while (i> 0) {

j ++;

i >> = 1;

    }

    return j-kvalue;

}

uint GetINFO (uint CodeNum, uint (4) kvalue, uint Mvalue) {

return CodeNum + 2 ^kvaule -2 ^{(kvalue + Mvalue)}

}

pointSequence {

  uint (lenBits) nbPoints;

  uint (1) flag;

  if (flag == 0) {

  if (nbPoints <3) {

      uint (5) bits;

      for (int i = 0; i <nbPoints; i ++) {

        uint (bits) x [i];

        uint (bits) y [i];

      }

    }

    else {

     uint (5) bits;

     uint (bits) x [0];

     uint (bits) y [0];                     

     uint (5) bitsx;

     uint (5) bitsy;

     for (int i = 1; i <nbPoints; i ++) {

       uint (bitsx) dx;

       uint (bitsy) dy;

        x [i] = dx + x [i-1];

        y [i] = dy + y [i-1];

      }

    }

  }

else {

  uint (5) bits;

  uint (bits) x [0];

  uint (bits) y [0];

  uint (4) kvalue;

  uint Mvalue = 0;

  uint CodeNum = 0;

  for (i = 1; i <nbPoints; i ++) {

    CodeNum = GetCodeNum (dx [i]);

    Mvalue = GetMvalue (CodeNum, kvalue);                     

    uint (Mvalue) 0;

 uint (1) 1;

 uint (Mvalue + kvalue) INFO_dx [i] = GetINFO (CodeNum, kvalue, Mvalue)

 CodeNum = GetCodeNum (dy [i]);

 Mvalue = GetMvalue (CodeNum, kvalue);

     uint (Mvalue) 0;

  uint (1) 1;

  uint (Mvalue + kvalue) INFO_dy [i] = GetINFO (CodeNum, kvalue, Mvalue)

}

  }

}

Semantics:

flag-the flag that determines FLC encoding (flag = 0) and EG encoding (flag = 1)

kvalue-The kind value of the geometric distribution. For example, as the value of kvalue increases, the geometric distribution becomes smoother.

Mvalue-the number of leading zeros

CodeNum-code number

INFO_dx-a value containing information about the difference between points in the X coordinate

INFO_dy-a value with information about the difference in points y coordinates

When the encoding method according to the present invention is applied to the fourteen test files described above, the following results are calculated. In the table below, the original file size represents the binary file size for each test file (.lsr), and the new file size represents the binary file size obtained according to the encoding method of the present invention.

As shown in the table above, data compression gains of 4 to 28% are obtained for all 13 selected test files except face_frame1. Because the purpose of these files is for testing, the size is considerably smaller than the files that can actually be served. For example, karate runs for about 20 seconds. Considering serviceable content that runs for about 10 minutes, a 3.71% data compression would result in savings of about 520,800 bits.

Another element of LASeR is the complexity of the decoder. When a point is encoded in a fixed length manner, it is assumed to be of complexity zero. As mentioned above, when the EG coding scheme according to the present invention is added, the decoder needs more computation. The equation used in the decoder is:

Inferring the complexity of the decoder from the above equation, the complexity of the decoder is as follows.

Decoder Complexity = (M + 2k Shift + 1 Add + 1 Subtract) * Number of Points

As described above, in the case of encoding the point sequence according to the encoding method of the present invention, a data compression gain of 4 to 28% is obtained, and the complexity of the decoder may be covered by shift, add and subtract operations. Thus, if the coding scheme according to the present invention is used in applications such as maps, vector graphics and cartoons containing a large number of point sequences, good compression efficiency will be achieved with low decoder complexity.

Claims (2)

In the method of encoding a point sequence, Selecting, for each point sequence, Exponential Golomb (EG) coding or Fixed Length Coding (FLC) coding based on a selected criterion; If FLC encoding is selected, encoding a point sequence according to the FLC encoding scheme; If EG encoding is selected, obtaining a parameter k that can perform EG encoding most effectively, and encoding each point sequence according to the EG encoding method using the parameter k . How to include. An apparatus for encoding a point sequence, Means for selecting Exponential Golomb (EG) coding or Fixed Length Coding (FLC) coding for each point sequence based on the selected criteria; Means for encoding a point sequence according to the FLC encoding scheme, if FLC encoding is selected, If EG encoding is selected, a means for obtaining a parameta k that can perform EG encoding most effectively, and using the parameter k to encode each point sequence according to the EG encoding scheme Point sequence encoding apparatus comprising a.

Images