JPH0133992B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an orthogonal transform decoding device for concealing errors in a coding method using n-order orthogonal transform. More specifically, the sequence of received blocks hi (i=
1, 2,...n), the error is corrected by replacing the predicted value αi predicted from the value of the sequence of blocks near this block with the value of the sequence hi and decoding it. It is. Furthermore, by selecting neighboring blocks to be used for prediction according to the pattern indicated by each sequence of received blocks, the purpose is to improve the accuracy of predicted values and reduce image quality deterioration in reproduced images. .

Hadamard transform is one of the orthogonal transforms used for coded transmission. In the present invention, a case will be explained in which Hadamard transform is applied to an image signal. There is an nth-order (n: positive integer) transformation as the Hadamard transformation, but for the sake of brevity, the fourth-order Hadamard transformation will be described. First, the configuration of an encoding device and a decoding device of a system for band-compressing a television signal and transmitting the resultant television signal using a fourth-order Hadamard transform will be explained with reference to FIG.

In FIG. 1, a television signal arriving at an input terminal 1 is quantized into, for example, 8 bits by an A/D converter 2 and written into a first memory 3. For example, select four points d ₁ , d ₂ , d ₃ , and d ₄ as shown in Figure 2 as sample points of the television signal, and convert these four points into 1
Suppose we perform the Hadamard transformation as one block. In FIG. 2, A shows the K-1st horizontal scanning line, A shows the K-th horizontal scanning line, C shows the K+1-th horizontal scanning line, and D shows the K+2-th horizontal scanning line. Each sample point di (i=1, 2,
If the sample values in 3 and 4) are Xi and [H] is a 4th order Hadamard matrix, the 4th order Hadamard transformation can be expressed as the following equation.

h ₁ , h ₂ , h ₃ , and h ₄ are fourth-order sequences transformed into the frequency domain by Hadamard transform. 4
The order Hadamard matrix [H] can be expressed as follows.

In Figure 1, 4 of the blocks shown in Figure 2
The sample values X ₁ , X ₂ , X _{3 ,} and
2, 3, 4). The first quantizer 5 quantizes each sequence to obtain Qihi. Qi (i=1, 2, . . . 4) is a quantization operator for each sequence. The first quantizer 5 allocates optimal bits to each sequence and reduces the number of bits of each sequence to be transmitted, thereby achieving band compression. For example, if the value of each sequence hi is represented by 8 bits, the conversion of 1 block is 32 bits.
It is necessary, but in the quantizer 5, Q ₁ h ₁
6bit, 4bit for Q ₂ h ₂ , 2 bit each for Q ₃ h ₃ and Q ₄ h ₄
can be compressed to a total of 14 bits.

The reason why such band compression is possible can be explained as follows. Performing the Hadamard transformation is
This is because when converting time-series data of image signals into the frequency domain, it is possible to reduce the bit allocation to sequences with few components or sequences representing high frequency components with low visual sensitivity in each sequence hi. .

In FIG. 1, a transmission line encoder 6 modulates each quantized sequence Qihi into a transmission line code suitable for the transmission line to be transmitted, and sends it to a transmission line 13. The above part shows the encoding side.

On the decoding side, the transmission line code is received from the transmission line 13, decoded by the transmission line decoder 7, and the quantized signal Qihi of each sequence is reproduced. Next, the second quantizer 8 obtains Qi ^-1 Qihi. Qi ^-1 is the inverse operator of Qi. Output hi′ of second quantizer 8 =
Qi ^-1 Qihi are each decoded to 8 bits. When band compression is performed, the decoded hi' has a different value from hi. The Hadamard inverse transformer 9 performs Hadamard inverse transform by the calculation shown in the following equation 3.

In Equation 3, as can be seen from Equation 2, [H ^-1 ]=[H] ...4. Furthermore, the reproduced value Xi' is the reproduced value of the sample value Xi.

The reproduction value Xi' of the sample point obtained as the output signal of the Hadamard inverse transformer 9 is stored in the second memory 10. The second memory 10 reads out the reproduction value Xi' in synchronization with the synchronization signal of the television signal,
The signal is converted into an analog signal by a D/A converter 11 and sent out from an output terminal 12 as a television signal.

When transmitting image signals using a system having the above configuration, an error may occur in the transmission system. If an image signal is reproduced using a signal in which an error has occurred, the quality of the reproduced image will be significantly degraded.
Error correction encoding methods have been proposed to correct erroneous data. The error correction code is
Error correction is performed by providing redundancy to the original encoded signal. Therefore, the number of bits required for transmission increases, and the effect of band compression using Hadamard transform is lost. In particular, when the transmission system includes a magnetic recording system such as a VTR,
Relatively long burst errors such as dropouts also occur frequently. In order to correct a code that has caused such a burst error, a very large amount of redundancy must be provided, which further reduces the effectiveness of band compression.

Further, there is an error detection encoding method that cannot perform error correction but can detect the occurrence of an error. Since the error detection code requires less redundancy than the error correction code described above, the effect of band compression is less likely to be impaired. As an error detection code, for example, there is a method of attaching a parity signal depending on whether a code word is an even number or an odd number. In addition, in block coding, in which the code is divided into appropriate blocks and encoded, the number of “0” and “1” included in one block is determined, and during decoding, “0” and “1” are Errors can be detected by calculating the number. In addition to these codes, error detection codes suitable for the transmission system used have been proposed.

When such an error detection code is used, it is important to consider how to reduce the influence of the error after the error is detected.

As a method for concealing errors in a decoding device using a coded transmission system using the Hadamard transform, we have already proposed a method in which the value of the sequence hi in which an error has occurred is replaced with a preset correction value αi (Japanese Patent Application Laid-Open No. 57-33842). This modification method utilizes the statistical properties of sequence values. That is, by taking a value where the statistical distribution of the appearance values of each sequence is concentrated as the correction value αi, the influence of errors on the playback screen is reduced.

Although this type of retouching method can improve the deterioration of reproduced image quality due to errors to some extent, since the retouching value αi is a fixed value, the deterioration of image quality may be large when reproducing images with special patterns. be.

As a novel error correction means, the present invention predicts the value of a sequence of received blocks from blocks in the vicinity of the received block according to the pattern expressed by each sequence, and uses the predicted value of the sequence in which an error has occurred. Embodiments of the present invention will be described below with reference to the drawings.

Assume that four sample points are selected from the television signal as shown in FIG. 2, and a fourth-order Hadamard transform is performed using these four points as one block. FIG. 3 is a diagram showing a wider range on the screen of a television signal when the same blocks as shown in FIG. 2 are divided. In FIG. 3, solid lines represent horizontal scanning lines, and circles represent sample points. Also block B
Let (m, n) represent the block in the mth row and nth column on the screen.

For such sample points d ₁ , d ₂ , d ₃ , d ₄ , 1
If the fourth-order Hadamard transformation shown in equations and equations 2 and 2 is performed, each sequence hi is expressed by the following equation.

_{h 1 = X 1 + X 2 + X 3 + _ _ _ _ 1} −X ₂ +X ₃ −X ₄ ...8 Here, the sample value of each sample point di is assumed to be Xi.

FIG. 4 schematically shows the patterns expressed in block B(m, n) by the sequences shown by equations 5 and 6. In the block in Figure 4, the solid color part indicates that the coefficient applied to the sample value Xn is +
1, and the shaded area indicates that the coefficient is -1
shall indicate that.

For example, in block B(m, n), X ₁ =X ₂ =
When a signal with a large correlation in the horizontal direction arrives, such as -X ₃ = -X ₄ , the sequence h ₂
It is thought that the value of will increase. In addition, for signals with high vertical correlation such as vertical stripes, the sequence h ₄
The value of the sequence h4 becomes large, and the value of the sequence _h4 becomes large for a signal with a large correlation in the diagonal direction, such as a diagonal line. The brighter the entire block, the greater the value of the sequence _h1 .

From the above, each sequence can be considered to represent a unique pattern.

Therefore, when calculating the predicted value αi used to correct the sequence hi in which an error has occurred, for each sequence, select a nearby block with a high correlation according to the pattern expressed by each sequence, and use the value of the sequence of this block. The accuracy of prediction can be improved by making predictions based on the following information.

For example, in FIG. 3, block B (m, n)
If an error occurs in the sequence h ₂ , then from equation 6,
_h2 has a large correlation in the horizontal direction, so the blocks B(m, n-1) and B(m, n+1) before and after in the horizontal direction
The prediction can be made using the value of the sequence _h2 . Here, the sequence of block B(m, n)
The value of hi is hi (m, n), and its predicted value is αi (m,
n), the predicted value α ₂ (m, n) is expressed as
can be found.

α ₂ (m, n) = 1/2 {h ₂ (m, n-1) + h ₂ (m, n+1)} ...9 Similarly, the sequence h ₃ has a large correlation in the diagonal direction, so the following equation Prediction can be made by

α ₃ (m, n) = 1/4 {h ₃ (m-1, n-1) + h ₃
(m-1, n+1)+h ₃ (m+1, n-1) +h ₃ (m+1, n+1)} ...10 Since the sequence h ₄ has a large correlation in the vertical direction, prediction can be performed using the following equation.

α ₄ (m, n) = 1/2 {h ₄ (m-1, n) + h ₄ (m+1, n)} ...11 Since the sequence h ₁ represents a DC component, using the surrounding blocks, It can be predicted by the following equation.

α ₁ (m, n)=1/8 {h ₁ (m-1, n-1) +h ₁ (m-1, n)+h ₁ (m-1, n+1) +h ₁ (m, n- 1) +h ₁ (m, n+1)+h ₁ (m+1, n-1) +h ₁ (m+1, n)+h ₁ (m+1, n+1)} ...12 Which sequence is What kind of pattern is expressed varies depending on how sample points are selected within the block. In this embodiment, a case will be described in which the sample points shown in FIGS. 2 and 3 are selected.

Next, the configuration of a decoding device for carrying out the modification method using the prediction formulas shown in Equations 9 to 12 will be explained with reference to FIG. In FIG. 5, a decoder 7, a second quantizer 8, an inverse Hadamard transformer 9, a second memory 10, and a D/A converter 11 have the same symbols as those shown in FIG. It is something.

In FIG. 5, a transmission line code encoded according to the transmission line 13 is received and decoded by a transmission line decoder 7 to reproduce the quantized signal Qihi of each sequence.
Next, the second quantizer 8 obtains Qi ^-1 Qihi.
Qi ^-1 , is the inverse operator of Qi. In this embodiment, the output Qi ^-1 Qihi of the second quantizer 8 will be simply expressed as
Let's write it as hi. The modification method according to the present invention is applied to each sequence hi, and since the modification circuit is the same for each sequence, sequence h ₁ will be explained as an example.

The output h ₁ of the second quantizer 8 is sequentially stored in the memory a
Write to 20, from memory a20, block B
The value h ₁ (m, n) of the sequence h ₁ of (m, n) is read out, and the switching circuit a22 and error detection circuit a23
Send to. At the same time, the value of the sequence _h1 of eight neighboring blocks required to perform the prediction shown in equation 12 is read from the memory a20, and the prediction circuit a
Send to 21. The prediction circuit a21 performs the calculation shown in Equation 12, obtains the predicted value α ₁ (m, n), and sends it to the switching circuit a22. Error detection circuit a23
receives h ₁ (m, n), detects an error therein, and when an error is detected, generates an error detection signal e ₁ and sends it to the switching circuit a22. Switching circuit a22
normally connects the output h ₁ (m, n) of the memory a20 to the input of the Hadamard inverse transformer 9, but when receiving the error detection signal e ₁ , the prediction circuit a21
The output α ₁ (m, n) of is connected to the input of the Hadamard inverse transformer 9. Next, FIG. 6 schematically shows how the values of the sequence _h1 are stored in the memory a20. Such a memory can be easily configured with a shift register or RAM. When using a shift register, the input/output connections of the register are fixed and the stored data is configured to be shifted sequentially, and when a RAM is used, the input/output addresses are configured to be shifted sequentially. be able to.

Although the present embodiment has been described using the sequence h ₁ as an example, the configurations of the other sequences h ₂ , h ₃ , and h ₄ are exactly the same. Figure 5 shows the sequence
An example of the configuration of _h4 is also shown. Depending on the prediction formula for each sequence, the number of neighboring blocks read out from memory and the calculations performed by the prediction circuit change for each sequence. The internal state of the memory for each sequence is expressed as the seventh
This is schematically shown in Figures 8 and 9.

In FIG. 5, the Hadamard inverse transformer 9 receives the sequence hi (m, n) or
For a sequence in which an error has been detected, the predicted value αi (m, n) is received, and Hadamard inverse transform is performed using the calculation shown in equation 3 to obtain block B (m, n).
Find the reproduction value Xi of each sample point di of n). The second memory 10 temporarily stores the reproduction value Xi and reads out the reproduction value Xi in synchronization with the synchronization signal of the television signal. The D/A converter 11 is connected to the second memory 1
Convert the signal read from 0 to an analog signal,
It is sent out from the output end 12 as a television signal.

In this example, in order to improve prediction accuracy,
A block to be used for prediction was selected from all the blocks adjacent to the block B(m, n) where the error occurred, according to the pattern shown by each sequence.
Such a method requires a lot of memory.

Therefore, as a simpler method, the blocks used for correcting the error may be selected from only the previous blocks in the chronological order. For example, the following prediction formula can be used.

α ₁ (m, n) = 1/4 {h ₁ (m-1, n-1) + h ₁
(m-1, n)+h ₁ (m-1, n+1)+h ₁ (m, n-
1)} α ₂ (m, n) = h ₂ (m, n-1) α ₃ (m, n) = 1/2 {h ₃ (m-1, n-1) + h ₃ (m-1, n + 1)} α ₄ (m, n) = h ₄ (m-1, n) Although this embodiment shows the fourth-order Hadamard transform as an example, the present invention is not limited to the fourth-order Hadamard transform. , it can also be applied to cases where higher-order Hadamard transforms and other orthogonal transforms are used.

As described above, according to the present invention, in a signal processing method using Hadamard transform, in order to correct the value of a sequence in which an error has occurred using an error correction code, a large degree of redundancy must be provided. Bandwidth compression becomes less efficient. Therefore, when errors are detected using an error detection code that has less redundancy than an error correction code, the effect of band compression is less likely to be impaired. At this time, by applying this modification method that replaces the value of the sequence where the error occurred with the predicted value predicted from the sequence value of the neighboring block selected according to the pattern expressed by this sequence, it is possible to Image quality deterioration can be effectively reduced, and band compression efficiency can also be kept high.

[Brief explanation of drawings]

Figure 1 is a block diagram for explaining the configuration of a coding transmission system using Hadamard transform, Figure 2 is a diagram showing how to select sample points in a television signal, and Figure 3 is a diagram for explaining the configuration of a coding transmission system using Hadamard transform. Figure 4 shows an example of blocking sample points.
A schematic diagram showing the pattern represented by each sequence,
FIG. 5 is a block diagram of an orthogonal transform decoding device according to an embodiment of the present invention, and FIGS.
9 are diagrams schematically showing sequence values stored in the memory in each sequence. 21, 25...Prediction circuit, 23, 27...Error detection circuit, 22, 26...Switching circuit, 9...Hadamard inverse conversion circuit, 11...Digital-to-analog conversion circuit.

Claims (1)

[Claims] 1. From an encoding device that divides an image signal into blocks, performs n-th orthogonal transformation on the blocks, and transmits a sequence hi (i=1, 2,...n) obtained as an orthogonal transformation output. In a decoding device that reproduces an image signal by reproducing a signal of a prediction circuit that calculates the predicted value αi for a given block from the value of a sequence of the same order of at least one neighboring block of that block, a switching circuit that replaces the value of the sequence in which an error has occurred with the predicted value, and the value of the replaced sequence. 1. An orthogonal transform decoding device comprising: an inverse transformer that performs orthogonal inverse transform using values of a sequence in which no error occurs. 2. The prediction circuit is a prediction circuit that obtains a predicted value by reading out the sequence value of a block at a preset position for each order of the sequence from a memory that sequentially stores the sequence value of the reproduced block. An orthogonal transform decoding device according to claim 1.