JPH09130788A - Signal restoring device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To restore especially a signal of a high frequency component among packet signals lost due to a network fault. SOLUTION: The signal restoring device discriminates missing blocks a1 and blocks a0 , a2 without missing from plural received signal blocks and assumes duplication of part of signal components in the missing blocks a1 and part of signal components in the blocks a0 , a2 without missing to be duplication blocks b0 , b2 . Then discrete Fourier transformation(DFT) is applied to the blocks a0 , a2 without missing to generate each block information A1 to obtain approximate values B0 (k), B2 (k) of a coefficient B1 based on each block information A1 , inverse discrete Fourier transformation is applied to them to generate duplicate blocks b0 , b2 including error signal components e0 , e2 , the result of DFT to the error signal components e0 , e2 is obtained respectively as errors E0 , E2 and signal components of the missing blocks a1 are estimated from the errors E0 , E2 .

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal restoration device for reproducing an audio signal, an image signal and the like by improving the signal quality thereof.

[0002]

2. Description of the Related Art Recently, for example, H.264. 261 and MPEG1,
Various image coding methods such as 2 have been standardized. These encoding schemes divide the image into multiple blocks,
A discrete cosine transform (DCT) process and a quantization process are performed for each image block, and a block-by-block process for image compression is performed.

In packet communication of image coded data based on this block processing, packet loss may occur due to network congestion or the like.

In this case, when the decoder decodes the coded data received through the network, the reproduced image may be lost in blocks, and the quality of the reproduced image may deteriorate.

For example, in encoding an MPEG audio signal or the like, a block process is often performed, and when such data is packet-communicated, a similar problem occurs in the decoding process.

As a countermeasure against this, a technique for adding an error correction code to a transmission packet or a decoder for automatic reproduction (AR
Q) Techniques for adding functions and the like have been devised.

The technique using the error correction code is such that a transmitting side, that is, an encoder adds a redundant bit to a transmission information bit to encode and transmit, and an error of an information bit is received based on the redundant bit received by a receiving side, that is, a decoder. Is detected and corrected. To improve the error correction capability, it is necessary to increase the number of redundant bits to be added, which deteriorates the transmission efficiency.

On the other hand, in ARQ, a transmitting side adds a redundant bit to a transmission information bit for encoding, but the receiving side detects an error in the information bit based on the redundant bit and requests the transmitting side to retransmit the packet. It is a thing. In this ARQ, error detection is performed on received data, but error correction is not performed. Therefore, the redundant bit added on the transmission side does not have to be as large as in the case of error correction coding, but error detection is performed instead. Since the packet is retransmitted, the transmission efficiency is reduced. Therefore, the error correction coding and AR
In Q and the like, if the transmission efficiency is prioritized in both cases, the quality of the reproduced signal may deteriorate.

By the way, a signal block such as an image has a property that it has a strong similarity to a peripheral block, in other words, a strong correlation.

Therefore, in recent years, by utilizing this property,
After decoding the signal, the degraded or lost signal block is
A technique for performing restoration using interpolation or replacement using blocks around the block has been introduced.

A conventional signal restoration technique will be described below with reference to FIGS. 4 and 5.

FIG. 4 is a diagram showing a flow of signal restoration processing for a one-dimensional received signal block, and FIG. 5 is a diagram showing a flow of signal restoration processing for a two-dimensional signal.

As shown in FIG. 4A, when one-dimensional signal blocks a ₀ , a ₁ and a ₂ are continuously received and, for example, the signal block a ₁ among them is lost (hereinafter referred to as signal block a). ₁ is referred to as an erasure block a ₁ ), and as shown in FIG. 4B, the signal blocks a ₀ and a ₂ adjacent to both sides of the erasure block a ₁ are filtered in the time domain to obtain the erasure block a ₁ There is a technology to interpolate.

In this case, since the interpolation process for the lost block a ₁ is a kind of simple averaging process, when the lost block a ₁ is a low frequency component, it can be restored with an approximate value, but at a high frequency. The components cannot be interpolated.

Further, as shown in FIG. 4C, the lost block a ₁ and the lossless blocks a ₀ and a ₂ adjacent to the lost block a ₁
Is subjected to a discrete Fourier transform (DFT) to generate frequency domain signal blocks A ₀ and A ₂ , respectively, and the signal blocks A _{0 and} A ₂ are averaged to obtain the frequency domain signal A ₁ of the erasure block. There is a technique for generating a signal and performing an inverse discrete Fourier transform (IDFT) on it to restore the signal block a ₁ in the time domain.

In any case, since the block averaging process is performed, the low frequency component can be almost restored, but the high frequency component cannot be interpolated at present.

On the other hand, as shown in FIG. 5, when the signal is an image, two-dimensional processing is required. In this case, the one-dimensional signal processing is expanded as it is, and the two-dimensional block signal is processed in the time domain or the frequency domain in the same manner as described above.

A typical technique of this type is based on linear interpolation, and although the compensation of low frequency components is good, the compensation of high frequency components is also difficult in this case. The reason is that interpolation is a kind of low pass filtering.

[0019]

As described above, according to the conventional signal restoration technique, in order to recover a part of the lost signal blocks of the transmitted plurality of signal blocks, the erased blocks are deleted from the neighboring signal blocks. There are a plurality of techniques for linearly obtaining and interpolating, but all of them are effective in compensating for low frequency components, but have a problem that they are not so effective in compensating for high frequency.

The present invention has been made to solve such a problem, and an object of the present invention is to provide a signal restoration device capable of compensating for a high frequency in repairing a lost block. Column

[0021]

In order to achieve the above-mentioned object, the invention according to claim 1 is a signal block generating means for generating a plurality of signal blocks consisting of signal components of a predetermined length, and the signal block generating means. Detecting whether the signal component is lost for each signal block generated by the means,
Lossless block having loss in signal component, determination means for determining lossless block having no loss in signal component, partial signal component of the signal block determined as lossy block by the determination means, and lossless Means for assuming a block in which a part of the signal components of the block overlap as an overlapping block, and orthogonal transform processing for the signal block determined to be a lossless block by the determination means to generate each block information. Based on the block information generated by the orthogonal transforming means, an approximate value generating means for obtaining an approximate value of block information obtained by orthogonally transforming the overlapping block, and an approximate value generating means. Inverse orthogonal transformation is performed on the approximated orthogonal transformation value of the duplicated block to generate the duplicated block including the error signal component. And an estimating means for estimating the signal component of the loss block from the error signal component and the signal component of the lossless block included in the overlapping block generated by the inverse orthogonal transforming means. ing.

According to the first aspect of the present invention, when a coded image signal or the like is received, first, a plurality of signal blocks composed of signal components of a predetermined length are generated, and the signal components are lost in each signal block. It is determined that there is a lossy block and a lossless block in which there is no loss in the signal component. Then, a block in which a part of the signal components of the lossy block and a part of the signal components of the lossless block are overlapped is assumed as an overlapped block.

Thereafter, the signal block determined to be a lossless block is subjected to orthogonal transform processing to generate block information in the frequency domain, and block information obtained by orthogonal transforming the overlapping block based on this block information. An approximate value of is obtained. An inverse orthogonal transform is applied to this approximate block information to generate a duplicate block containing an error signal component, and the loss block signal is calculated from the error signal component and the lossless block signal component included in this block. The components are estimated.

As a result, it becomes possible to interpolate the signal components from the low frequency region to the high frequency region with respect to the lost block.

According to a second aspect of the present invention, a signal block generating means for generating a plurality of signal blocks composed of signal components of a predetermined length, and a signal component is lost for each signal block generated by the signal block generating means. Determining means for detecting whether there is a loss in the signal component and a lossless block having no loss in the signal component, and a part of the signal component of the loss block determined by the determining means And means for assuming a block obtained by overlapping a part of the signal components of the lossless block as an overlapping block,
Based on the block information generated by the discrete Fourier transform means, and the discrete Fourier transform means for generating each block information by performing a discrete Fourier transform on the signal block determined as the lossless block by the determination means, Approximate value generating means for obtaining an approximate value of block information obtained by performing discrete Fourier transform on the overlapping block, and performing inverse discrete Fourier transform on the approximate block information of the overlapping block obtained by the approximate value generating means. From the inverse discrete Fourier transform means for generating the overlapping block containing the error signal component, and the error signal component and the signal component of the lossless block included in the overlapping block generated by the inverse discrete Fourier transform means , Estimating means for estimating the signal component of the lossy block.

According to the second aspect of the present invention, when a coded image signal or the like is received, first, a plurality of signal blocks composed of signal components of a predetermined length are generated, and a signal component is lost in each signal block. It is determined that there is a lossy block and a lossless block in which there is no loss in the signal component. Then, a block in which a part of the signal components of the lossy block and a part of the signal components of the lossless block are overlapped is assumed as an overlapped block.

Thereafter, the signal block determined to be a lossless block is subjected to discrete Fourier transform to generate block information in the frequency domain, and a block obtained by discrete Fourier transform of the overlapping block based on this block information. The approximate value of the information is obtained. An inverse discrete Fourier transform is applied to this approximate block information to generate an overlapping block containing an error signal component, and from the error signal component and the signal component of the lossless block included in this,
The signal component of the lost block is estimated.

This makes it possible to interpolate the signal components from the low frequency region to the high frequency region with respect to the lost block, that is, the lost block.

According to a third aspect of the present invention, in the signal restoration apparatus according to the first aspect, the symmetric component adding means for adding the symmetric component of the signal component to the signal block determined to be the lossless block by the determining means is provided. , The orthogonal transform is performed by the discrete Fourier transform or the discrete cosine transform based on the signal block to which the symmetric component is added by the symmetric component adding means, and the inverse orthogonal transform is performed by the inverse discrete Fourier transform or the inverse discrete cosine transform. It is characterized by restoring blocks.

According to the third aspect of the present invention, the above restoration process is performed by adding the symmetric component of the signal component to the lossless block. At this time, the orthogonal transform is performed by the discrete cosine transform, and the inverse orthogonal transform is performed. By performing the transform by the inverse discrete cosine transform, it is possible to use the DCT device and the IDCT device provided in the existing decoding device, and the number of calculations can be reduced because the complex number operation is not performed, and the existing device can be reused. This can contribute to cost reduction.

According to a fourth aspect of the present invention, in the signal restoration apparatus according to the second aspect, the approximation value generating means gives an approximation value of block information obtained by performing a discrete Fourier transform of the overlapping block, and the error that the overlapping block has. It is characterized in that it is calculated from an equation in which the position and magnitude of the signal are unknowns.

According to the invention of claim 4, an equation in which the position and magnitude of the error signal are unknowns, that is, the equation (1-8
), The approximate value of the block information obtained by performing the discrete Fourier transform of the overlapping block is obtained, so that it is possible to estimate and remove the unknown error, assuming that two or less of the eight components of E ₀ are not 0.

According to a fifth aspect of the present invention, in the signal restoration apparatus according to the second aspect, the approximate value generating means discretizes the loss block based on block information obtained by performing discrete Fourier transform on the lossless block. A feature is that a signal block obtained by Fourier transform is estimated, and an approximate value of the block obtained by performing discrete Fourier transform of the overlapping block is generated from the estimated signal block and the block information.

According to the fifth aspect of the present invention, the approximate value of the overlapping block after the discrete Fourier transform is obtained almost accurately.

According to a sixth aspect of the present invention, a dividing unit that divides the two-dimensional image information into a plurality of blocks, and whether or not the image component in the block is lost is determined for each block divided by the dividing unit. Then, a determination unit that determines a lossy image block having a loss in the image component and a lossless image block that has no loss in the image component, and an image of a part of the image block determined as the lossy image block by the determination unit Assuming an overlapping image block that contains the component and some image components of the lossless image block located around it,
Of the image components in the overlapping image block, an unknown image component is estimated from a known image component by a predetermined calculation process to generate the overlapping image block, and the overlapping image block generating unit generates the overlapping image block. Means for estimating the image component of the original lost image block based on the image component of the duplicated image block.

According to the sixth aspect of the present invention, an overlapping image block including a part of the image components of the image block determined to be the loss image block and a part of the image components of the lossless image blocks located around it. Assuming that, among the image components in this overlapping image block, an unknown image component is estimated from a known image component by a predetermined calculation process to generate an overlapping image block, and based on the image component of this overlapping image block. Estimate the image components of the original lost image block.

That is, since the lost image block is restored in consideration of the continuity between the blocks of the two-dimensional image, the lost block can be restored more accurately than before.

[0038]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a diagram showing a first signal processing mode of the signal restoration apparatus according to the present invention.

The present invention is intended to restore block omissions in block-coded audio signals and image signals, and the above-mentioned image signals are usually subjected to two-dimensional processing. The two-dimensional processing is an extension of the one-dimensional processing, and in this first signal processing mode, a basic one-dimensional processing algorithm will be described. 1
A signal a _i for a unit is assumed to be composed of, for example, eight data blocks, and this signal a _i is _expressed as a _i = (a _i (0), a _i (1) to a _i (7) ).

First, let us consider three consecutive data blocks a ₀ , a ₁ , a ₂ with i = 0, 1, 2 of the signal a _i . In this case, the first data block a ₀
The data block a ₁ between the third data block a ₂ and the third data block a ₂ is the lost block (loss block) a _{1 in} which a packet loss occurs due to a network failure or the like, and normal data blocks a ₀ , a ₂ adjacent to this block Lossless blocks a ₀ , a ₂
And In such a case, this signal restoration device restores the data block a ₁ based on the data blocks a ₀ and a ₂ .

First, the discrete Fourier transform (DFT) of the signal a _i for one unit is A _i = (A _i (0), A _i
(1) to A _i (7)).

[0043] Here, the DFT of the data block a ₁ is estimated by the linear interpolation method.

There are many such interpolation methods,
Here, the averaging method will be described as the simplest method. In this averaging method, the average of two points obtained by discrete Fourier transform is obtained by the following equation (1-2).

A ₁ (k) = (A ₀ (k) + A ₂ (k)) / 2 (k = 0,1 to 7) Formula (1-2) Thus, the mutual average values A ₁ (k) after obtaining a lossless block a ₀ and a lost block a ₁ behind it is overlapped (overlapping) data blocks (overlapping blocks) as _{b 0, b 0 = (b} 0 (0), b 0 ( 1) ~ b ₀ (7)) =
(A ₀ (4), a ₀ (5), a ₀ (6), a ₀ (7), a ₁ (0), a ₁ (1),
a ₁ (2), defined as a ₁ (3))), and lost block a ₁ and block the lossless block a ₂ is overlap behind the (overlapping blocks) as b _2, b ₂ = (b
₂ (0), b ₂ (1) to b ₂ (7)) = (a ₁ (4), a ₁ (5), a ₁
(6), a ₁ (7), a ₂ (0), a ₂ (1), a ₂ (2), a ₃ (3)).

Here, the data block b _i (where i = 0,
2), a material obtained by discrete Fourier transform (DFT), B _i
= (B _i (0), B _i (1) to B _i (7)), B ₀ (k) = (A ₀ (k) + A ₁ (k)) / 2 (k = 0,1 to 7) Formula (1-3) B ₂ (k) = (A ₁ (k) + A ₂ (k)) / 2 (k = 0,1 to 7) Each approximate value B _{0 in} formula (1-4) (k) and B ₂ (k) are obtained.

Then, the coefficient B obtained by the discrete Fourier transform
the _i Then reverse DFT (IDFT), data block b _i
= (B _i (0), b _i (1) to b _i (7)) Becomes

Where b ₀ (0), b ₀ (1), b ₀ (2), b
_The correct value of ₀ (3) is a ₀ (4), according to the definition of the Fourier transform.
It is known that a ₀ (5), a ₀ (6), a ₀ (7), and the error e ₀ (n) is e ₀ (n) = b ₀ (n) −a ₀ (n + 4) (n = 0,1,2,3) Equation (1-6) can be defined.

Similarly, b ₂ (4), b ₂ (5), b ₂ (6), b ₂ (7)
Also, the correct value is _{a 2 (0), a 2} (1), a 2 (2), since we know the a ₂ (3), the error e ₂ (n + 4) is, e ₂ (n + 4 ) = B ₂ (n + 4) -a ₂ (n) (n = 0,1,2,3) Equation (1-7) can be defined.

Subsequently, these known errors e ₀ (n), e ₂ (n
+4), (n = 0,1,2,3) unknown error e ₀ (n + 4), e ₂ (n),
Estimate (n = 0,1 to 3).

Error e ₀ = (e ₀ (0), e ₀ (1) to e ₀ (7))
When the error _{e 2 = (e 2 (0} ), e 2 (1) ~e 2 7) DFT ) of
Error E ₀ = (E ₀ (0), E ₀ (1) to E ₀ (7))
And the error E ₂ = (E ₂ (0), E ₂ (1) to E ₂ (7)).
The errors E ₀ and E ₂ are A ₁ (k), B ₀ (k), B ₂ (k) where (k =
It is caused by the approximation error of 0,1 to 7).

Among the eight components of the errors e ₀ and e ₂ , four consecutive components are known.

If there are two or less non-zero components among the eight components of the errors E ₀ and E ₂ , they can be estimated. Therefore, unknown error e ₀ (n + 4), e ₂ (n)
(n = 0,12,3) can be estimated / removed. This utilizes the property of DFT, and this principle is described in the following document [1] and the like.

Reference [1] REBlahut. “Algebraic Meth
ods for Signal Processing and Communication Codin
g ”, Springer-Verlag, 1992. However, in this document [1], assuming that a certain number of signals in the frequency domain are known, a certain number of noises added in the time domain are estimated, and error of the signal in the frequency domain is removed. doing.

On the other hand, in the present invention, conversely, assuming that a certain number of signals in the time domain are known, a certain number of noises to which the frequency domain is added are estimated, and the error of the signal in the time domain is removed. The domain-frequency domain relationship is different.

Taking the above error E ₀ as an example, out of the 8 components of the error E ₀ , the number of _nonzero components is 2 or less, and these are estimated, and unknown errors e ₀ (n), (i = 4 , 5,6,7) is estimated / removed.

E ₀ = (E ₀ (0), E ₀ (1) to E ₀ (7)) 8
Of the individual components, the positions of non-zero ones are p and q,
The following equation (error locator polynomial) is defined. However, (0 ≦ p, q ≦ 7) λ ₀ (X) = (1−Xω ^−p ) (1−Xω ^−q ) = λ ₀ (0) + λ ₀ (1) X + λ ₀ (2) X ² formula ( 1-8) where ω = exp (-j2π / 8), λ ₀ (0) = 1 λ ₀ (X) coefficient vector (λ ₀ (0), λ ₀ (1), λ ₀ (2)
The DFT of ~ λ ₀ (7) is It is. From equations (1-8) and (1-9), Λ ₀ (k) = 0 for k when E ₀ (k) ≠ 0, and Λ ₀ (k ) E ₀ (k) = 0 (k = 0,1 to 7) Equation (1-10) is obtained. Therefore, And λ ₀ (0) = 1, and λ ₀ (n) = 0, where (n = 3,
4,5,6,7), It is written as. However, <nk> 8 is the remainder obtained by dividing nk by 8.

Expression (1-12) is expressed as a matrix as follows.

[0059]

(Equation 1) The matrix on the right side is called the Toeplitz matrix, and λ ₀ (1) and λ ₀ (2) are obtained by finding the inverse matrix of this matrix. In this case, the error blocks e ₀ (0), e ₀ (1), e ₀ (2),
Since e ₀ (3) is known, λ ₀ (1) and λ ₀ (2) are determined by solving the following equation (1-14).

[0060]

(Equation 2) From this, the unknown error block e ₀ (4), e ₀ (5), e
₀ (6) and e ₀ (7) are determined.

Similarly, in the case of E ₂ , the error block e
_{2 (4), e 2 (} 5) ~e 2 (7) since it is known, lambda
₂ (1), λ ₂ (2) is determined, and from this, unknown error blocks e ₂ (0), e ₂ (1), e ₂ (2), e ₂ (3) are estimated.

A ₁ (n) = b ₀ (n + 4) −e ₀ (n + 4) where (n = 0,1,2,3) Formula (1-15) a ₁ (n) = b ₂ (n-4) -e ₂ (n-4) where (n = 4,5,6,7) a ₁ = (a ₁ (0), a ₁ (1) derived from equation (1-16) ~ A ₁ (7)) is the reproduced block.

The algorithms of this method can be summarized as the following processing procedures (1) to (6). However, here it is assumed that each block consists of eight signal components.

(1) For the lost block a ₁ and the lossless blocks a ₀ and a ₂ adjacent to the lost block a ₁ , D of the block a ₁
The A ₁ is a FT, averaged and A ₂ is a DFT of A ₀ and Block a ₂ is a DFT block a ₀ and approximated.

(2) The DFT B ₀ of the overlapping block b ₀ having the four components of the block a _{1 and} the four components of the block a ₀ is approximated by the average of A ₀ and A ₁ .

Similarly, B ₂ which is the DFT of the overlap block b ₀ having 4 components of a ₁ and 4 components of a ₂ is approximated by the average of A ₁ and A ₂ .

(3) B ₀ is composed of 8 components, and it is assumed that 6 or more components have correct values, and 2 or less components include an error.

Similarly, B ₂ is composed of 8 components, but it is assumed that correct values are obtained for 6 or more components, and that 2 or less components include an error.

[0069] (4) B ₀ to be subjected to IDFT determine the overlap block b _0. Since B ₀ contains an error, overlapping block b ₀ also contains an error.
Similarly determine the overlap block b ₂ is subjected to IDFT in B _2. Since B ₂ contains an error, this overlapping block b ₂ also contains an error.

(5) Overlap block b ₀ and a ₀
For the four parts corresponding to the component of, the error value for the true value is known, and based on that, the error value of the part that does not overlap with a ₀ at b ₀ is detected, the true value is estimated, and all The true value of b ₀ is obtained.

Similarly, in the four parts corresponding to the component of a ₂ in b ₂ , the error value with respect to the true value is known, and the error value of the part not overlapping with a ₂ in b ₂ is detected based on it. Then, the true value is estimated, and the true values of all b ₂ are obtained.

(6) The four components other than the component of a _{0 in} b ₀ corrected above and the four components of b ₂ other than the component of a ₂ are used as the restoration block of lost block a _1. To do.

In the above process, in (1) and (2), B ₀ and B ₂ which are DFTs of the overlapping blocks.
Is calculated by averaging from adjacent blocks.

Then, in the processing of (3), B ₀ , B ₂
Since there are less than 2 errors in each, (4), (5)
2 errors are estimated and removed from the 4 known components by the processing of (1), and the lost block a ₁ is finally restored in (6).

In the processes of (1) and (2), it is required that the approximation accuracy is high, but in general, a block of an image has a strong correlation with its neighboring / peripheral blocks, especially in a low frequency region. The estimation error of the low frequency component is relatively small. Therefore, it is considered that the approximation of the low frequency component can be calculated with high accuracy even in the averaging process of (1) and (2). Furthermore, in order to improve this approximation accuracy, there is also a measure such that not the average of two adjacent blocks but several neighboring blocks including adjacent blocks are weighted and averaged stepwise. As described above, the restoration of the low frequency component can be performed without a problem so far, but the restoration of the high frequency component is generally a problem.

Therefore, in this signal restoration device, (4),
In the process of (5), the overlap block is assumed, and the time component is once converted into the frequency component and further inversely converted to restore the high frequency component with a small amount of calculation and high approximation accuracy.

As described above, according to the signal restoration apparatus of the first signal processing mode, when the lost block a ₁ is restored, the lossy blocks a ₀ and a ₂ are each subjected to the discrete Fourier transform (DFT). To generate each block information (frequency domain signal blocks A ₀ and A ₂ ), and DFT the overlap block b ₀ based on this block information to obtain an approximate value B ₀ (k) of the block information B _0. Then, the inverse discrete Fourier transform (IDF) is calculated for this approximate value B ₀ (k).
T) is applied to generate an overlap block b ₀ containing an error signal component, and from the error signal component e ₀ and the lossless signal component a ₀ included therein, the signal of the loss block a ₁ is generated. Since the components are estimated, not only low frequency components but also high frequency components can be compensated.

In the first signal processing mode, one signal block has eight signal components, but this is an example, and the present method is applied to a signal block having an arbitrary number of signal components. Applicable.

Next, a second signal processing mode of the signal restoration apparatus of the present invention will be described.

The first signal processing mode is a processing mode in which the discrete Fourier transform (DFT) is performed as the orthogonal transform process and the inverse discrete Fourier transform (IDFT) is performed as the inverse orthogonal transform process. The image coding apparatus used is configured to perform a discrete cosine transform (DCT) and an inverse discrete cosine transform (IDCT) for image coding and decoding. That is, the actual image coding device includes the DCT device and the IDCT device, but D
In many cases, an FT device or IDFT is not provided.

Therefore, in the second signal processing mode, the DCT device / ID provided in the actual image coding device.
A process of restoring a signal using the CT device as it is will be described.

Generally, an N-point DCT is a 2N-point DFT.
This principle is used here because it can be calculated with. As an example, N = 8.

For example, a signal block a _i consisting of 8 points or the like
= (A _i (0), a _i (1) to a _i (7)) However, for (i = 0,1,2), a signal block consisting of 16 points a _i = (a _i (0) ' , A _i
(1) 'to _ai (15)') is expressed as _ai (n) '= _ai (n) (n = 0,1 to 7) Formula (2-1) _ai (n)' = _ai (15-n) (n = 8,9 to 15) Formula (2-2). It is a 16-point DFT of a _i (n) ', (n = 0,1 to 15)
A _{(DFT) i} (k) 'is It is.

On the other hand, an 8-point DCT of a _i (n), (n = 0, 1 to 7)
A _{(DCT) i} (k) is However, C ₀ = (1/8) ^1/2 formula (2-5) C _k = (1/4) ^1/2 (k = 1,2 to 7) formula (2-6). Therefore, A _{(DFT) i} (0) '= (32) ^1/2 A _{(DCT) i} (0) Formula (2-7) A _{(DFT) i} (8)' = 0 Formula (2-8) A _{(DFT) i} (k) '= (16) ^1/2 exp (jπk / 16) A _{(DCT) i} (k) (k = 1,2 to 7) Formula (2-9) A _{(DFT) i} ( 16-k) '= (16) ^1/2 exp (-jπk / 16) A _{(DCT) i} (k) (k = 1,2 to 7) There is a relation of formula (2-10).

From this, A _{(DFT) i} (k) 'which is a 16-point DFT of a _i (n)', (n = 0, 1 to 15) is a _i (n), (n = 0, 1 ~ 7) 8
It can be realized by A _{(DCT) i} (k) which is a point DCT.

Next, an example in which the above principle is applied to the signal restoration device of the present invention will be described with reference to FIG. FIG. 2 is a diagram showing a second signal processing mode of the signal restoration device according to the present invention.

As shown in the figure, a plurality of adjacent signal blocks a ₀ , a ₁ , a ₂ are assumed to consist of, for example, 8 signal components.

The number of these eight signal components is merely an example, and a signal block may have an arbitrary number of signal components within a certain block length.

In this case, first, the lost block a ₁ = (a
₁ (0), a ₁ (1) to a ₁ (7)), a lossless block adjacent to this is a ₀ = (a ₀ (0), a ₀ (1) to a ₀ (7)) ) 、
If a ₂ = (a ₂ (0), a ₂ (1) to a ₂ (7)), then the above formula (2
-1) and equation (2-2), symmetric components are added to generate a ₁ ′, a ₀ ′, and a ₂ ′ consisting of 16 components.

A _i '= (a _i (0)', a _i (1) 'to a _i (7)', a _i (8) 'to a _i (14)', a _i (15) ') = (A _i (0), a _i (1) to a _i (7), a _i (7) to a _i (1), a _i (0)) where (i = 0,1,2) expression ( 2-11) Then, the 16-point DFT of a _i ', (i = 0, 2) is converted into A _i ' = (A
_{If i} (0) ', A _i (1)' ~ A _i (15) ', This formula (2-12) is an 8-point D from the relation of formulas (2-7) to (2-10).
It can be calculated by CT.

The DFT of a ₁ ′ is A ₁ ′ = (A ₁ (0), A ₁
(1) to A ₁ (15)), which is approximated by the simplest averaging method as in the first signal processing mode.

That is, A ₁ (k) '= (A ₀ (k)' + A ₂ (k) ')
/ 2 where (k = 0,1 to 15) is the equation (2-13), and this equation (2-13)
An approximate value can be obtained from

Next, the overlapping block b ₀ overlapping the lossless block a ₀ and the loss block a ₁ is replaced with b ₀
= (B ₀ (0), b ₀ (1) to b ₀ (7)) = (a ₀ (4), a ₀ (5),
It is defined as a ₀ (6), a ₀ (7), a ₁ (0), a ₁ (1), a ₁ (2), a ₁ (3)).

The loss block a ₁ and the lossless block a
_The overlapping block b ₂ overlapping with ₂ is b ₂ =
(B ₂ (0), b ₂ (1) to b ₂ (7)) = (a ₁ (4), a ₁ (5), a
₁ (6), a ₁ (7), a ₂ (0), a ₂ (1), a ₂ (2), a ₂ (3)).

Here, b ₀ ′ and b ₂ ′ consisting of 16 components are created by adding symmetric components according to the equations (2-1) and (2-2).

B _i '= (b _i (0)', b _i (1) 'to b _i (7)', b _i (8) 'to b _i (14)', b _i (15) ') = (B _i (0), b _i (1) to b _i (7), b _i (7) to b _i (1), b _i (0)) where (i = 0,2) formula (2- 14) Then, the DFT of b _i ′ is B _i ′ = (B _i (0) ′, B _i ).
(1) 'to B _i (15)'), B ₀ (k) '= (A ₀ (k)' + A ₁ (k) ') / 2 where (k = _{0, 1 to} 15) Equation (2 _{-15) B 2 (k) '} = (A 1 (k)' + A 2 (k) ') / 2 where is approximated by (k = 0,1~15) formula (2-16).

Further, the IDFT of B _i ′ is b _i ′ = (b
_i (0) ', b _i (1)' to b _i (15) ') Becomes This equation can be transformed into the following equation.

[0098] Here, the DCT of b _i = (b _i (0), b _i (1) to b _i (7))
B _{(DCT) i} = (B _{(DCT) i} (0), B _{(DCT) i} (1) to B _{(DCT) i}
(15)).

[0099] At this time, from formulas (2-7) to (2-10), B _i (0) '= (32) ^1/2 B _{(DCT) i} (0) formula (2-20) B _i (8)' = 0 Expression (2-21) B _i (k) '= (16) ^1/2 exp (jπk / 16) B _{(DCT) i} (k) (k = 1,2 to 7) Expression (2-22) B _i (16-k) '= (16) ^1/2 exp (-jπk / 16) B _{(DCT) i} (k) (k = 1,2 to 7) There is a relation of formula (2-23), Equation (2-20) to Equation (2-23) and Equation (2-18)
From Is obtained. At this time, Therefore, b _i (n) '= b _{(IDCT) i} (n) (n = 0,1 to 7) Formula (2-26) b _i (15-n)' = b _{(IDCT) i} (n ) (N = 0,1 to 7) Equation (2-27) is obtained.

Therefore, the IDFT of B _i ′ can be obtained from the IDCT of B _i . Where b ₀ (0), b
₀ (1), b ₀ (2), b ₀ (3) have correct values a ₀ (4), a
It is known that ₀ (5), a ₀ (6), a ₀ (7), and the error e ₀ (n) is e ₀ (n) = b ₀ (n) −a ₀ (n + 4) ) (N = 0,1,2,3) Formula (2-28) is defined. Similarly, b ₂ (4), b ₂ (5), b ₂ (6), b
_The correct values for ₂ (7) are a ₂ (0), a ₂ (1), a ₂ (2),
Since it is known to be a ₂ (3), the error e ₂ (n + 4)
Is defined as e ₂ (n + 4) = b ₂ (n + 4) −a ₂ (n) (n = 0,1,2,3) Formula (2-29).

Thus, the known errors e ₀ (n), e ₂ (n + 4)
However, the unknown error e ₀ (n + 4), e ₂ (n) (n = 0, 1, 2, 3) can be estimated from (n = 0, 1, 2, 3).

Here, the error block e _i '= {e _i (0), e _i (1) to e _i (1
5)} is created.

E _i (n) '= e _i (n) (n = 0,1 to 7) Expression (2-30) e _i (15-n)' = e _i (n) (n = 0,1) 7) The DFT B ₀ ′ and B ₂ ′ of the overlapping blocks b ₀ ′ and b ₂ ′ of the formula (2-31) have already been obtained from the formula (2-15) and the formula (2-16) and the block a. It is approximated from A ₀ ′, A ₂ ′ which is a DFT of ₀ ′, a ₂ ′. The DFT of the error blocks e ₀ ′ and e ₂ ′ of the block b ₀ ′ b ₂ ′ is defined as approximation errors E ₀ ′ and E ₂ ′.

Approximation error E _i '= {E _i (0)', E _i (1) '~
E _i (15) '} and the error block e _i ' = {e _i (0) ', e
_i (1) '~ e _i (15)'} However, the relationship with (i = 0,2) is It is.

Therefore, The relationship of E ₁ (8) ′ = 0 is obtained.

Error of block b ₀ ′, b ₂ ′ The DFT of block e ₀ ′, e ₂ ′ is error E ₀ ′, E ₂ ′, of which the number of non-zero components is 4 or less. And the unknown error block e ₀ (n + 4), e ₂ (n) (n
= 0,1,2,3) is estimated / removed.

Next, an example when the error is E ₀ 'is described.

E ₀ '= (E ₀ (0)', E ₀ (1) 'to E ₀ (1
5) ') position of non-zero one of 16 components
The following equation (error position polynomial) (2-36) is defined as q, r, and s. λ ₀ (X) = (1−Xω ^−p ) (1−Xω ^−q ) (1−Xω ^−r ) (1−Xω ^−s ) = λ ₀ (0) + λ ₀ (1) X + to λ ₀ ( 4) X ⁴ formula (2-36) where ω = exp (-j2π / 16), λ ₀ (0) = 1, 0 ≦ p, q,
r, s ≦ 7.

The DFT of λ ₀ = (λ ₀ (0), λ ₀ (1) to λ ₀ (15)) is obtained.

[0110] Λ ₀ (k) = 0 for k when the approximation error E ₀ (k) ≠ 0
Then, for all k, Λ ₀ (k) E ₀ (k) = 0 (k = 0, 1 to 15) Equation (2-38) is obtained. Therefore, Because It is requested as follows. However, <nk> 16 means that nk is 16
It is the remainder divided by.

Here, λ ₀ (1) = 1 and λ ₀ (n) =
0, (n = 5,6 to 15), and further, considering the symmetry of the signal defined for calculating the DFT by the DCT, λ
₀ (1) = λ ₀ (3) and λ ₀ (4) = 1.

Expression (2-40) is expressed by the following determinant.
(2-41) to (2-43).

[0113]

(Equation 3) By finding the inverse matrix of the above determinant, λ ₀ (1), λ ₀
(2), λ ₀ (3) and λ ₀ (4) are determined. From this, unknown errors e ₀ (4), e ₀ (5), e ₀ (6), and e ₀ (7) are determined.

Similarly in the case of E ₂ , e ₂ (4), e
_{Since 2} (5), e ₂ (6), e ₂ (7) are known, λ ₂ (1), λ
₂ (2), λ ₂ (3), and λ ₂ (4) are determined.

From this, the unknown error e ₂ (0), e ₂ (1), e
₂ (2) and e ₂ (3) can be estimated.

A ₁ (n) = b ₀ (n + 4) −e ₀ (n + 4) (n = 0,1 to 7) Formula (2-44) a ₁ (n + 4) = b ₂ ( n) −e ₂ (n) (n = 0, 1 to 7) Formula (2-45) Erasure block a ₁ = (a ₁ (a ₁ ( 0), a ₁ (1) to a ₁ (7)} are reproduced blocks.

As described above, according to the signal restoration device of the second signal processing mode, the reproduction capability of the lost block a ₁ is almost the same as that of the first signal processing mode, but the first signal processing mode is While the calculation is performed by DFT and IDFT, in the second signal processing mode, calculation is performed by discrete cosine transform (DCT) and inverse discrete cosine transform (IDCT). That is, in order to perform DFT and IDFT, multiplication and addition operations of complex numbers are required, whereas in DCT and IDCT, multiplication and addition operations of real numbers are sufficient, so there is an advantage that the amount of operations is relatively small.

Further, the currently standardized image codec device, that is, the encoder, the decoder, etc., generally has a portion for executing processing such as DCT or IDCT, for example, D.
Equipped with CT device and IDCT device, these devices can be used as they are, and the signal restoration / interpolation function can be easily realized without externally adding (expanding) the circuit (configuration), and hardware It is also very cost effective in terms of configuration.

Next, the third signal processing mode of the signal restoration apparatus of the present invention will be described with reference to FIG. FIG. 3 is a diagram showing an example of restoring a two-dimensional image.

In the above-described first and second signal processing modes, the basic one-dimensional interpolation processing is shown for one unit of signal. However, when applied to image processing, two-dimensional processing is performed. Therefore, in this third signal processing mode, an example of two-dimensional expansion will be described.

For example, one two-dimensional image signal block a _{h, i} consisting of 8 dots × 8 dots pixels is expressed by the following equation (3-1) when expressed in a matrix.

[0122]

(Equation 4) Let _{h of} this two-dimensional image signal block a _{h, i be} 0,1,2,
When i is 0,1,2, as shown in FIG. 3, a two-dimensional image is formed by nine adjacent two-dimensional image signal blocks a _{0,0 to} a _2,2 .

[0123] In the third signal processing mode, the blocks a _0,0 ~a _{2, 2} 9 pieces of the two-dimensional image, for example, if the central block a _{1, 1} is lost (block a _1, If the ₁ and erasure block), eight blocks a _0,0 disposed adjacent the periphery of the block a _{1, 1,} a _{0, 1,}
It is assumed that the lost block a _1,1 is restored from a _0,2 , a _1,0 , a _1,2 , a _2,0 , a _2,1 , a _2,2 .

In this case, when the two-dimensional block a _{h, i} is subjected to the two-dimensional DFT, the two-dimensional block A _{h, i} is obtained. Each element of the two-dimensional block A _{h, i} is given below.

[0125] Conversely, when the two-dimensional block A _{h, i} is subjected to the two-dimensional IDFT, the two-dimensional block a _{h, i} is obtained. Each element of the two-dimensional block a _{h, i} is given below.

[0126] Then, A _1,1 which is the DFT of the lost block a _1,1 is
The DFTs of adjacent blocks are averaged to obtain an approximate value.

A _1,1 (k, g) = [A _0,0 (k, g) + A _0,1 (k, g) + A _0,2 (k, g) + A _1,0 (k, g) + A _1,2 (k, g) + A _2,0 (k, g) + A _2,1 (k, g) + A _2,2 (k, g)] / 8 where (k = 0,1 to 7), (g = 0,1 to 7) Expression (3-4) Then, the DFT of the overlapping block (overlapping block) of the blocks a _0,0 , a _0,1 , a _1,0 and a _1,1 is set to B.
_{With 0,0} , the approximate value is obtained by averaging according to the following equation (3-5).

B _0,0 (k, g) = [A _0,0 (k, g) + A _0,1 (k, g) + A _1,0 (k, g) + A _1,1 (k, g) ] / 4 where (k = 0,1 to 7) and (g = 0,1 to 7) Formula (3-5) Similarly, blocks a _0,2 , a _0,1 , a _1,2 , a _An approximate value is obtained by averaging the DFT of the overlap block of _1,1 with B _0,2 by the following equation (3-6).

B _0,2 (k, g) = [A _0,2 (k, g) + A _0,1 (k, g) + A _1,2 (k, g) + A _1,1 (k, g) ] / 4 where (k = 0,1 to 7) and (g = 0,1 to 7) Formula (3-6) Similarly, blocks a _2,0 , a _2,1 , a _1,0 , a _{1 The} approximate value is obtained by averaging the DFT of the overlapping block of 1,1 with B _2,0 by the following equation (3-7).

B _2,0 (k, g) = [A _2,0 (k, g) + A _2,1 (k, g) + A _1,0 (k, g) + A _1,1 (k, g) ] / 4 where (k = 0,1 to 7) and (g = 0,1 to 7) Formula (3-7) Similarly, blocks a _2,2 , a _2,1 , a _1,2 , a _{1 The} approximated value is obtained by averaging the DFT of the overlapping block of 1,1 with B _2,2 by the following equation (3-8).

B = _2,2 (k, g) = [A _2,2 (k, g) + A _2,1 (k, g) + A _1,2 (k, g) + A _1,1 (k, g )] / 4 where (k = 0,1 to 7), (g = 0,1 to 7) Formula (3-8) block B _0,0 , B _0,2 , B _2,0 , B _2,2 Respectively include approximation errors, and let their IDFT be b _0,0 , B _0,2 , b _2,0 and b _2,2 , respectively,
Each contains an error.

Here, the error estimation / removal method will be described by taking the block B _0,0 and its IDFT b _0,0 as an example.

The error included in the block B _0,0 is _defined as an error block E _0,0 , and the error included in the IDFT b _0,0 is defined as an error block e _0,0 .

The respective elements of the two-dimensional error blocks E _0,0 and e _{0,0 have} the relationships of equations (3-9) and (3-10).

[0135] Then, of the 64 elements of the block b _0,0 , b
Elements other than the 16 elements of _0,0 (m, n) (m = 4,5,6,7), (n = 4,5,6,7) are blocks a _0,0 , a _0, It is an element of ₁ , a _1,0 and its true value is known, and the error e _0,0 (m, n) can be understood as follows.

E _0,0 (m, n) = b _0,0 (m, n) −a _0,0 (m + 4, n + 4) where (m = 0,1,2,3), ( n = 0,1,2,3) Formula (3-11) e _0,0 (m, n) = b _0,0 (m, n) -a _0,1 (m + 4, n-4) (m = 0,1,2,3), (n = 4,5,6,7) Formula (3-12) e _0,0 (m, n) = b _0,0 (m, n) −a _1,0 (m-4, n + 4) However, (m = 4,5,6,7), (n = 0,1,2,3) Formula (3-13) Based on these, block b ₀ Out of 64 elements of ₀
b _0,0 (m, n) However, the error of 16 elements of (m = 4,5,6,7), (n = 4,5,6,7) is estimated and the error component is removed. .

In the first and second signal processing modes described above, a one-variable error locator polynomial is defined, but in this third signal processing mode, two-variable error locator polynomial λ _0,0 (Y, X).
Is determined.

Two-dimensional DF for the coefficient of λ _0,0 (Y, X)
Give T, For k, g when E _0,0 (k, g) ≠ 0, Λ _0,0 (k, g)
= 0, and for all k, g Λ _0,0 (k, g) E _0,0 (k, g) = 0 where (k = 0,1 ~ 7), (g = 0,1 ~ 7) It becomes the formula (3-16). Therefore, Therefore, from these equations, a two-variable error locator polynomial λ
_0,0 (k, g) is obtained, and based on these, 16 errors e _0,0 (m, n) of the elements of b _0,0 (m = 4,5,6,7) , (n = 4,5,
6,7) is estimated. Based on this, the lost block a _1,1 6
16 out of 4 components are obtained.

A _1,1 (m, n) = b _0,0 (m + 4, n + 4) −e _0,0 (m + 4, n + 4) where (m = 0,1,2, 3), (n = 0,1,2,3) Equation (3-18) In the case of B _0,2 , B _2,0 , B _2,2 , the same operation is performed and a _1,1 (m , n) = b _0,2 (m + 4, n-4) −e _0,2 (m + 4, n-4) where (m = 0,1,2,3), (n = 4,5 , 6,7) Formula (3-19) a _1,1 (m, n) = b _2,0 (m-4, n + 4) -e _2,0 (m-4, n + 4) where ( m = 4,5,6,7), (n = 0,1,2,3) Formula (3-20) a _1,1 (m, n) = b _2,2 (m-4, n-4 ) -E _2,2 (m-4, n-4) However, (m = 4,5,6,7), (n = 4,5,6,7) Formula (3-21) is obtained. From these equations (3-18) to (3-21), 64 components of the lost block a _1,1 can be restored and regenerated.

As described above, according to the signal restoration device of the third signal processing mode, of the nine adjacent two-dimensional image signal blocks a _{0,0 to} a _2,2 , the central two-dimensional image signal block When a _1,1 is a lost block, this lost block a _1,1 and eight lossless image blocks a _0,0 , a _0,1 , a _{0, which} are arranged adjacent to the lost block a _{1,1 2} , a _1,0 , a
Overlap blocks (overlapping image blocks) b _0,0 , b _0,2 , b that overlap with _1,2 , a _2,0 , a _2,1 and a _2,2
Assuming _2,0 and b _2,2 , their DFTs are obtained, and these are averaged to obtain an approximate value, and then IDFT is performed to overlap blocks b _0,0 , b _0,2 and b _2,0. , b _2,2 ,
The errors it contains e _0,0 , e _0,2 , e _2,0 , e _2,2
Since the lost block a _1,1 is restored from the lost block a _1,1 as in the first signal processing mode,
_1,1 can be restored.

In the third signal processing mode, when the DFT of the erasure block a _{1,1 and} the DFT of the overlap block b _0,0 , b _0,2 , b _2,0 , b _2,2 are calculated. Although the simplest averaging process is performed, other processes such as weighting and averaging each data may be performed.

Further, the signal restoration apparatus of the third signal processing form is the calculation processing by the DFT, which is a two-dimensional extension of the first signal processing form, but it is two-dimensional like the second signal processing form. The calculation by the DCT extended to is also possible.
In this case, similarly to the second signal processing mode, the restoration process is performed on the block to which the symmetrical component is added.

[0143]

As described above, according to the present invention, when a blocked signal is transmitted from the transmission side to the reception side through the network, the reproduction block may be lost at the reception side due to a network failure or the like. However, assuming an overlapping block from the erasure block and the lossless block adjacent to this, the signal component of the lossless block is orthogonally transformed, the approximate value of the orthogonal transformation of the overlapping block is obtained, and it is Inverse orthogonal transform is performed to generate a duplicate block containing an error component, and the signal component of the lost block is estimated from the signal component of this duplicate block, so the signal component of the lost block can be almost certainly interpolated and received and lost. Not only low frequency components but also high frequency components can be compensated for the signal block.

Further, according to the present invention, a discrete cosine transform device (DCT device) or an inverse discrete cosine transform device (IDCT) provided in a general image coding device, image decoding device, or the like.
In addition to the above effects, it is possible to efficiently perform signal restoration with a small amount of calculation in addition to the above effects.

That is, since the signal block is calculated with respect to the signal component in the time domain-frequency domain, and the lost block is estimated by performing the linear interpolation while considering the continuity with the signal components in the surroundings, the low frequency to the high frequency are calculated. The signal up to the component can be restored.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a first signal processing mode of a signal restoration device of the present invention.

FIG. 2 is a diagram for explaining a second signal processing mode of the signal restoration device of the present invention.

FIG. 3 is a diagram for explaining a third signal processing mode of the signal restoration device of the present invention.

FIG. 4 (a) is a diagram showing a plurality of received signal blocks. (B) is a conventional signal restoration technique,
It is a schematic diagram at the time of restoring a lost block by filtering. (C) is a conventional signal restoration technique,
It is a schematic diagram at the time of restoring an erasure block from an erasure block.

FIG. 5 is a diagram for explaining a restoration process of a two-dimensional image signal in the conventional signal restoration technique.

[Explanation of symbols]

_{a i, a 0, a 2} ... 1 -dimensional signal block, a ₁ ... lost block, b _i ... overlapping blocks (overlapping blocks), A _i ... a _i to a discrete Fourier transform (DFT) of the frequency region subjected Block information, block information (coefficients) obtained by applying DFT to B _i ... b _i , block obtained by _performing inverse discrete cosine transform (IDCT) on b _{(IDCT) i} ... B _i , B _{(DCT) i}
... b _i subjected to DCT, e _i ... error block, E _i ... e _i subjected to DFT error, a _{h, i} ... two-dimensional image signal block, A _{h, i} ... a _{h, i} 2 A block that has undergone a three-dimensional DFT.

Claims (6)

[Claims] 1. A signal block generating means for generating a plurality of signal blocks having signal components of a predetermined length, and detecting whether or not the signal component is lost for each signal block generated by the signal block generating means. A determination block for determining a loss block having a loss in the signal component and a lossless block having no loss in the signal component; a signal component of a part of the signal block determined to be the loss block by the determination unit; Means for assuming a block in which a part of the signal components of the lossy block overlaps as an overlapping block, and each block information is subjected to orthogonal transform processing on the signal block determined to be a lossless block by the determination means. An orthogonal transform unit for generating, and a block obtained by orthogonal transforming the overlapping block based on the block information generated by the orthogonal transform unit. Approximate value generation means for obtaining an approximate value of the check information, and the overlapping block containing an error signal component by performing an inverse orthogonal transform on the approximate block information of the overlapping block obtained by the approximate value generation means. Inverse orthogonal transform means for generating, and an estimating means for estimating the signal component of the loss block from the error signal component and the signal component of the lossless block included in the overlapping block generated by the inverse orthogonal transform means. A signal restoration device comprising: 2. A signal block generating means for generating a plurality of signal blocks having signal components of a predetermined length, and detecting whether or not the signal component is lost for each signal block generated by the signal block generating means. A lossless block having a loss in the signal component and a lossless block having no loss in the signal component, a part of the lossy block determined by the determination means, and the lossless block Means for assuming a block in which a part of the signal components are overlapped as an overlapped block, and each block information is generated by applying a discrete Fourier transform to the signal block determined as the lossless block by the determination means. Discrete Fourier transform means, and based on the block information generated by the discrete Fourier transform means, the overlapping block is subjected to discrete Fourier transform. The approximate value generating means for obtaining the approximate value of the block information obtained by converting the approximate block information of the overlapping block obtained by the approximate value generating means is subjected to inverse discrete Fourier transform to include an error signal component. And a signal of the loss block from an error signal component and a signal component of the lossless block included in the overlap block generated by the inverse discrete Fourier transform unit. A signal restoration device comprising: an estimation unit for estimating a component. 3. A symmetric component adding means for adding a symmetric component of a signal component to a signal block determined to be a lossless block by the determining means, the signal block having a symmetric component added by the symmetric component adding means. The signal restoration according to claim 1, wherein the lossy block is restored by performing orthogonal transform by discrete Fourier transform or discrete cosine transform and inverse orthogonal transform by inverse discrete Fourier transform or inverse discrete cosine transform. apparatus. 4. The approximate value generation means calculates an approximate value of block information obtained by performing discrete Fourier transform on the overlapping block from an equation in which the position and magnitude of the error signal of the overlapping block are unknowns. The signal restoration device according to claim 2, which is characterized in that. 5. The approximate value generation means estimates a signal block obtained by performing a discrete Fourier transform of the loss block based on block information obtained by performing a discrete Fourier transform of the lossless block, and estimates the estimated signal block. The signal restoration device according to claim 2, wherein an approximate value of a block obtained by performing a discrete Fourier transform of the overlapping block is generated from the signal block and the block information. 6. A dividing unit that divides the two-dimensional image information into a plurality of blocks, and it is determined whether an image component in each block is lost for each of the blocks divided by the dividing unit, and the image component is lost. A lossy image block with a determination unit for determining a lossless image block with no loss in image components, a partial image component of the image block determined as the lossy image block by the determination unit, and its surroundings. Assuming an overlapping image block that includes some image components of the lossless image blocks located, of the image components in this overlapping image block, an unknown image component is estimated from a known image component by a predetermined calculation process. Then, an overlapping image block generating means for generating the overlapping image block, and an original based on the image component of the overlapping image block generated by the overlapping image block generating means. And a means for estimating the image component of the lossy image block.