Method and apparatus for recovery of encoded data using central value

Method and device for recovering coded data using central value

<p>100. . . Signal</p><p>1.........N. . . Envelop</p><p>120. . . decoder</p><p>170. . . Processing system</p><p>175. . . processor</p><p>185. . . Output</p><p>300. . . Encapsulation structure</p><p>110. . . Encoder</p><p>135. . . Transmission media</p><p>130. . . Signal</p><p>190. . . Memory</p><p>180. . . enter</p><p>195. . . Display device</p>

The purpose, characteristics and advantages of the present invention will be easily understood by those who are familiar with the art with reference to the following detailed description, among which:

Fig. 1A shows an embodiment of the processing of signal encoding, transmission, and subsequent decoding.

1B and 1C are embodiments of the present invention implemented as software executed by a processor.

Figures 1D and 1E show a valuable embodiment of the present invention implemented as hardware logic.

Figure 2 shows an embodiment of the sealing structure.

3A and 3B show the difference between the actual and restored Q code levels when the dynamic range (DR) is over-evaluated according to an embodiment.

Background of the invention

1. Field of the Invention The present invention relates to data encoding processing to provide robust error recovery from data loss caused by signal transmission or signal storage.

2. Background of the Invention Nowadays, several technologies exist to reconstruct lost/damaged data caused by random errors during signal transmission or storage. However, these technologies cannot deal with the loss of continuously enclosing data. The enclosing data of consecutive losses is described as a series of errors in this technique. The burst error may cause the reconstructed signal with degraded quality. The degraded quality is obvious to the end user. In addition, the compression technology used to complete high-speed communication is combined with the signal degradation caused by the burst error, so it is added to the degradation of the reconstructed signal. . Examples of the loss of burst errors that affect the transmission and/or storage of signals can be seen in high-definition television (HDTV) signals, mobile telecommunications applications, and video storage technologies. Such video storage technologies include video discs (such as DVD), micro Disc and video recorder (VCR).

For example, the advent of HDTV has resulted in television signals with a higher resolution than the current standard proposed by the National Television System Committee (NTSC). The HDTV signal mentioned is digital. Therefore, when a color television signal is converted for digital use, often the luminance and chrominance signals can be digitized using eight bits. The digital transmission of NTSC color television signals requires a nominal bit rate of approximately 260 million bits per second. For higher HDTV transmission rates, the rate may be rated at approximately 1200 million bits per second. These high transmission rates may have exceeded the bandwidth supported by current wireless standards. Therefore, an effective compression method is needed.

In mobile telecommunications applications, compression methods are also an important role. Typically, the packetized data is transmitted between remote terminals in mobile telephony applications. The limited number of transmission channels in mobile communications requires an effective compression method before the transmission is packetized. Several compression techniques can be used to achieve high-speed transmission rates.

Adaptive Dynamic Range Coding (ADRC) and Discrete Cosine Transform (DCT) coding provide image compression techniques known in this art. These two techniques utilize local co-correlation in an image to achieve a high compression ratio. However, an effective compression deduction method may cause synthetic error transmission due to errors in a more important encoded signal during subsequent encoding. This false amplification may cause a degraded video image, which can be quickly seen by the user.

Summary of the invention

The present invention advances a method to include data by determining the median value greater than the minimum value of the data range and less than the maximum value of the data range. In one embodiment, the central value is selected to be a value that actually reduces the decoding error when the range value is evaluated later. In one embodiment, the central value reduces the expected mean square error when there is an error during reconstruction. In one embodiment, the maximum value and the minimum value represent the intensity of the pixel data of an image. In another embodiment, the compression process is adapted to dynamic range coding, and the central value is a value within the dynamic range of the data in addition to the maximum and minimum values.

Schematic description

The purpose, characteristics and advantages of the present invention will be easily understood by those who are familiar with the art with reference to the following detailed description, among which:

Fig. 1A shows an embodiment of the processing of signal encoding, transmission, and subsequent decoding.

1B and 1C are embodiments of the present invention implemented as software executed by a processor.

Figures 1D and 1E show a valuable embodiment of the present invention implemented as hardware logic.

Figure 2 shows an embodiment of the sealing structure.

3A and 3B show the difference between the actual and restored Q code levels when the dynamic range (DR) is over-evaluated according to an embodiment.

Symbol description of main components

100. . . Signal

1.........N. . . Envelop

120. . . decoder

170. . . Processing system

175. . . processor

185. . . Output

300. . . Encapsulation structure

110. . . Encoder

135. . . Transmission media

130. . . Signal

190. . . Memory

180. . . enter

195. . . Display device

Detailed description

Not invented to provide a method to encode and arrange a signal stream to provide robust error recovery, and a method to perform error recovery. In the following description, for the purpose of explanation, various details are described to provide a complete understanding of the present invention. However, for those who are familiar with the art, these details are not necessary in implementing the present invention. In other examples, known electrical structures and circuits are shown in block diagrams to prevent unnecessarily limiting the present invention.

The following description is in the article of adaptive dynamic range coding (ADRC) encoding video images, more specifically, about the restoration of a loss or damage (loss/damage) such as dynamic range (DR) compression parameters. However, it is conceivable that the present invention is not limited to video, and is not limited to ADRC encoding and the specific compression parameters generated; instead, it is obvious that the present invention is applicable to different compression technologies and different types of co-related data, including, but not It is not limited to two-dimensional static images, holographic images, three-dimensional static images, video, two-dimensional moving images, three-dimensional moving images, monophonic sound, and N channel sound. The present invention can also be applied to different compression parameters including, but not limited to, the central value (CEN), which can be used in ADRC processing. In addition, the present invention can also be applied to different types of ARDC processing including edge matching and non-edge matching ADRC. For further information on ARDC, please refer to the 4th International HDTV Conference held in Tunis, Italy from September 4 to 6, 1991. "Adaptive Dynamic Range Coding Design for Future HDTV Digital VTR" published by Fujimori and Nakaya. "middle.

Signal encoding, transmission and subsequent decoding processing are shown in Figure 1A. The signal 100 is the data stream input to the encoder 110. The encoder 110 follows the adaptive dynamic range coding (ADRC) compression deduction method and generates packet 1,..., N for transmission along the transmission medium 135. The decoder 120 receives the packet 1,..., N from the transmission medium 135, and generates a signal 130. The signal 130 is a reconstruction of the signal 100.

The encoder 110 and the decoder 120 can be implemented in various ways to perform the functions described herein. In one embodiment, the encoder 110 and/or the decoder 120 may be implemented as software stored on a medium and executed by a computer or data processing system of general purpose or special specifications. The computer system typically includes a central processing unit. The unit, memory, and one or more input/output devices and co-processors are shown in Figures 1B and 1C. Alternatively, the encoder 110 and/or the decoder 120 may be implemented as logic to perform the functions described herein, as shown in FIGS. 1D and 1E. In addition, the encoder 110 and/or the decoder 120 can be implemented as a combination of hardware, software, or firmware.

Examples of circuits for encoding and restoring loss/damage compression parameters are shown in Figures 1B and 1C. The method described here can be executed on a special specification or general purpose processor system 170. The instructions are stored in the memory 190 and received by the processor 175 to execute many steps described herein. The input 180 receives the input stream and forwards the data to the processor 175. Output 185 output data. In Figure 1B, the output may contain encoded data. In FIG. 1C, once the compression parameters are restored, the output may include decoded data, such as decoded image data, sufficient to drive an external device such as the display 195.

In another embodiment, the output 185 outputs the restored compression parameters. The recovered compression parameters are then input to other circuits to generate decoded data.

Figures 1D and 1E show an embodiment of a circuit for encoding compression parameters and recovering lost/damaged compression parameters. The methods described here can be implemented in special configuration logic, such as the application of special integrated circuit (ASIC), large size integrated circuit (LSI) logic, programmable array, or one or more processors.

FIG. 2 shows an embodiment of a data structure or packet structure 300 used for data transmission between peer-to-peer connections and networks. The encapsulation structure 300 is generated by the encoding 110 and transmitted between the transmission media 135. For one embodiment, the packet structure 300 includes five-byte header information, eight DR bits, eight CEN bits, a moving flag bit, a five-bit critical index, and a 354-bit Q code. The encapsulation structure described here is exemplified and can be implemented for transmission on Asynchronous Transfer Mode (ATM) networks. However, the present invention is not limited to the encapsulation structure described here and the encapsulation structure used in each network can be used.

In one embodiment, the data structure 300 can be stored in a computer readable memory, so the data structure 300 can be accessed by a program running on a data processing system. The data structure 300 stored in the memory includes a dynamic range object (DR) and a central value data object (CEN) related to the dynamic range data object. The median data object has a value that is greater than the minimum value of the dynamic range data object and less than the maximum value of the dynamic range data object. When the dynamic range data target is evaluated, the median data target substantially reduces a decoding error. Each data structure 300 may also be a encapsulated structure.

As mentioned above, the above example systems and devices can be used to encode images, such as video or moving images using ADRC. ADRC has established the same achievable real-time technology to encode and compress images in preparation for fixed bit rate transmission.

The discrete data points that make up a digital image are known as pixels. Each pixel can be independently represented by 8 bits, but other representations can also be used for compression or analysis purposes. Many representations start by dividing the original data into separate data groups. For conventional reasons, these groups, which can be composed of one or more pieces of data or pixels, are regarded as blocks, even if they do not have the conventional block shape. These data can then be characterized by compression parameters. In an embodiment, these compression parameters include block parameters and bitstream parameters.

A piece of parameter includes data describing how the image is. Therefore, block parameters can also be used to define one or more attributes of the block. For example, in ADRC, the block width information may include the minimum pixel value (MIN), the maximum pixel value (MAX), the central value (CEN), the dynamic range of the pixel value (DR), or a combination of these values.

The bitstream parameters can also include data on how the image is encoded. In one embodiment, the bitstream parameter also indicates the number of bits used to encode data. For example, in ADRC, the bitstream parameters can also include Q bits and movement flag (MF) values. Therefore, in this embodiment, the bitstream parameter can indicate how the data is encoded, which indicates that a pixel value is within the range specified by the global information.

In the case of using ADRC encoding, the block data includes MIN, DR, and Q bit number (defined below), and the pixel data includes Q code. DR can be defined as MAX-MIN or MAX-MIN+1. In the embodiment, as described below, CEN can also be defined as a value between MIN and MAX. For example, CEN can be equal to MIN+DR2.

The Q code is in the range [0, 2 ^Q-2 An integer in ], which indicates one of the values in the group {MIN, MIN+1,..., CEN,..., MAX}. Due to the Q bit, Q, is usually small and the DR value can be quite large, it is generally impossible to correctly present all the pixel values. Therefore, when the pixel value is reduced to the Q code value, some quantization errors will be introduced. For example, if the Q bit number is 3, it can show 2 from the group {MIN, MIN+1,..., CEN,..., MAX} ³ = 8 value without error. Pixels with other values surround one of these 8 values. These ring heats can introduce errors.

Temporary compression is feasible for image sequences that are expanded more than once in time. An image frame is defined as a 2-dimensional set of pixels that rise in a given time period. It is known that data from related locations that are temporarily close to the image frame tend to contain similar values. When this is true, compression can be improved by encoding these similar values only once.

In the second example, by adding a moving flag (MF) to the block information of the first example, multiple image frames can be encoded. This MF indicates whether the data from each frame uses separate Q code encoding. If there is no indication to move, use the same Q code to represent each frame of data. If movement is indicated, separate Q codes are used to encode each frame.

Two methods of ADRC encoding can be used: non-edge-matched ADRC and edge-matched ADRC. The difference between the two methods is the exact formula used to generate the Q code value. On the other hand, these two methods have a lot in common. The two methods start by segmenting the image into blocks, and then determine the maximum (MAX) and minimum (MIN) pixel values of each block. In 2DADRC, a quantization code (Q code) value is determined for each pixel. In 3DADRC, a move flag (MF) value (if the move is 1, otherwise 0) is determined to be used for each block. When the moving flag is 1, a unique Q code can be determined to be used for each block. When the moving flag is 0, the relevant pixel values can be averaged for each block, the block parameters are updated accordingly, and a single Q code that will present the relevant pixels from each frame can be determined.

Non-edge matching ADRC can define the DR value such as DR=MAX-MIN+1 (1)

And a quantization code such as

<maths><img file="TW477151B_D0001.tif" /></maths>

Where Q is the number of quantization bits, and X _i Is the original pixel value (or average pixel value, in the case of non-moving blocks in 3DADRC). The pixel value can be reconstructed or restored according to the following formula:

<maths><img file="TW477151B_D0002.tif" /></maths>

Where MAX represents the maximum level of the block, MIN represents the minimum level of the block, Q represents the number of quantization bits, q _i Represents the quantization code (encoded data), X'i represents the decoding level of each sample, and its expected value is X' _i<img file="TW477151B_D0003.tif" /> X _i 。

Edge matching ADRC can define the DR value such as DR=MAX-NIN (4)

And a quantization code such as

<maths><img file="TW477151B_D0004.tif" /></maths>

Where Q is the number of quantization bits, and X _i Is the original pixel value (or average pixel value, in the case of non-moving blocks in 3DADRC). The pixel value can be reconstructed or restored according to the following formula:

<maths><img file="TW477151B_D0005.tif" /></maths>

Where MAX represents the maximum level of the block, MIN represents the minimum level of the block, Q represents the number of quantization bits, q _i Represents the quantization code (encoded data), X'represents the decoding level of each sample, and its expected value is X' _i<img file="TW477151B_D0006.tif" /> X _i 。

Although the above example quantization code and reconstruction formula for ADRC use the MIN value, any value greater than or equal to MIN, and less than or equal to MAX can also be used with DR to encode and decode pixel values. For edge-matched and non-edge-matched ADRC, the DR value will be lost during transmission. If the DR is lost, the pixel value can use an evaluation weight for DR.

When the DR is over-evaluated (or under-evaluated), the maximum decoding error is related to the values used for encoding and decoding pixel values, such as block parameters. Figures 3A and 3B show the difference between the actual and restored Q code levels when the DR is over-evaluated by 20%.

For example, Figure 3A shows the maximum decoding error when the DR is over-evaluated by 20%, and the MIN value is used for encoding and decoding. Figure 3B shows the maximum decoding error when the DR is over-evaluated by 20%, and the CEN value is used. The maximum decoding error in Figure 3B, which uses CEN, is smaller than the maximum decoding error when MIN is used.

The axis 210 on the left side of FIG. 3A is the proper restoration of the Q code in a 2-bit ADRC block using non-edge-matched ADRC. The right axis 220 is the Q code restored if the DR is over-evaluated by 20%. As shown ^， The maximum decoding error occurs at the maximum Q code value. (The same result occurs when the DR is under-assessed)

The performance shown in Figure 3A can be compared with the one shown in Figure 3B using a median value instead of MIN. Assuming that the same DR evaluation error is reached, when CEN is used, the maximum recovery error becomes half. Furthermore, the expected mean square error is reduced, thereby providing a corresponding increase in the signal-to-noise ratio (SNR) of the recovered signal. Therefore, by using CEN, the recovery of the Q code used for encoding, transmission, and decoding of image data can be enhanced, and when the DR evaluation error occurs, the two-mean-square decoding error and the maximum decoding error can be substantially reduced, and even smaller .

The central value can be selected as a value to substantially reduce the DR evaluation, or even reduce the expected mean square error for reconstruction, and have a fixed DR evaluation error. This value can be determined by the following processing.

The general type of ADRC decoding equation without truncation error is:

<maths><img file="TW477151B_D0007.tif" /></maths>

Where z _i The values of, M, and K are provided in Table 1. The general form of formula (7) also simplifies the ADRC symbol and allows the non-edge matching and edge matching formulas to be derived at the same time.

<tables><img file="TW477151B_D0008.tif" /></tables>

If the MIN value is not transmitted, but other values are transmitted and the DR value is absolutely positive, the other values can be expressed as: VAL=MIN+αDR (8)

Where α is a constant. Therefore, the ADRC decoding equation is:

<maths><img file="TW477151B_D0009.tif" /></maths>

Make DR _e Represents the error assessment of the dynamic range. Then the error decoding can be expressed as:

<maths><img file="TW477151B_D0010.tif" /></maths>

Where X' _error(i) Indicates an error in decoding, and therefore decoding error error _i =X' _I -X' _error(i) Can be written as:

<maths><img file="TW477151B_D0011.tif" /></maths>

Therefore, the mean square error (MSE) can be expressed as a function of α:

<maths><img file="TW477151B_D0012.tif" /></maths>

<chemistry general="n"><img file="TW477151B_D0013.tif" /></chemistry>

<maths><img file="TW477151B_D0014.tif" /></maths>

Where error i represents a decoding error, N represents the number of incorrectly decoded pixels, and α is a non-negative real number.

The expected mean square error can be expressed as a function of α to optimize α:

<maths><img file="TW477151B_D0015.tif" /></maths>

<maths><img file="TW477151B_D0016.tif" /></maths>

Where MSE(α) represents the mean square error expressed as a function of α, and E represents the expected mean square error.

By calculating the first and second deviations, the conditions for reduction can be checked:

<maths><img file="TW477151B_D0017.tif" /></maths>

<maths><img file="TW477151B_D0018.tif" /></maths>

It can be seen from equation (18) that when DR _e When DR, the second deviation

It is absolutely positive; therefore, the point on E'(MSE(α))=0 is all the smallest. This can be achieved if:

<maths><img file="TW477151B_D0019.tif" /></maths>

Therefore, equation (19) is:

<maths><img file="TW477151B_D0020.tif" /></maths>

Assuming that one of the Q code values is uniformly distributed, the expected value can be:

<maths><img file="TW477151B_D0021.tif" /></maths>

Therefore, in the case of non-edge matching ADRC, equation (20)

<maths><img file="TW477151B_D0022.tif" /></maths>

Similarly, in the case of edge matching ADRC, equation (20) becomes:

<maths><img file="TW477151B_D0023.tif" /></maths>

Substituting α=1/2 into equation (8), the best value for transmission is:

<maths><img file="TW477151B_D0024.tif" /></maths>

For non-edge matching ADRC or edge matching ADRC.

Although this deviation assumes that one of the Q code values is uniformly distributed, the non-uniform distribution of Q code values near the middle of the area can also support the use of CEN values.

Using equation (16) and substituting α=0, therefore, VAL=MIN, and a=12, therefore, VAL=CEN, which can quantify the transmission advantages of CEN.

Assuming that one of the Q code values is uniformly distributed, E(q _i² ) Can be calculated as follows:

<maths><img file="TW477151B_D0025.tif" /></maths>

The ratio of the mean square error used for CEN value decoding to the mean square error used for MIN value decoding is tabulated under various Q bit values Q as shown in Table 2:

<maths><img file="TW477151B_D0026.tif" /></maths>

<tables><img file="TW477151B_D0027.tif" /></tables>

Therefore, hypothesis one can be quantified. The reduction of mean square error of DR recovery under these common types of ADRC coding. Therefore, the CEN value is the best corresponding part of the mean square of the DR in the DR loss of ADRC transmission. This type of code is regarded as the central value ADRC.

In the central value ADRC, the central value (CEN) can be transmitted instead of the MIN value. In one embodiment, as described above, the CEN value can be defined as

<maths><img file="TW477151B_D0028.tif" /></maths>

In this embodiment, it is used to reconstruct X _i The formula of 'can be obtained by substituting MIN=CEN-DR/2 into equations (3) and (6). That is, for non-edge matching ADRC:

<maths><img file="TW477151B_D0029.tif" /></maths>

And in the example of edge matching ADRC:

<maths><img file="TW477151B_D0030.tif" /></maths>

In the error-free case, the performance of ADRC using the central value of CEN is similar to that of ADRC using the MIN value. However, in the case of DR losses, the central value ADRC can provide better loss/damage data recovery performance than the MIN value ADRC.

The present invention is not limited to the above-mentioned embodiment, and various changes and modifications can still be made here, but it still belongs to the spirit and scope of the present invention. Therefore, the spirit and scope of the present invention should be defined by the scope of the following patent applications.