JPH0445014B2 - - Google Patents

Description

Detailed Description of the Invention (Industrial Application Field) This technology is used in recording, transmission equipment, and other various equipment that process digital signals to efficiently encode signals with a smaller amount of information. This relates to efficiency coding methods.

(Prior art) When trying to digitize signals such as image signals and audio signals, the basic encoding method is to divide the signal level equally for each sample value,
Linear quantization (uniform quantization) is used in which values included in each range are replaced with one representative value.

When uniform quantization as described above is used to digitize signals, in order to make the difference between the representative point and the original value indiscernible, it is generally necessary to use 6 bits (32nd order) for natural images. Since it is said that 8 bits (256 gradations) are required from 8 bits to 16 bits (65536 gradations) for audio,
When attempting to transmit or record an image signal or an audio signal that has been digitized by uniform quantization as described above, it is necessary to handle a large amount of information as described above for each sample value. Another problem is that a larger scale device is required compared to the device used when transmitting and recording the above-mentioned image signals or audio signals in an analog system.

Therefore, even though the signal is encoded with a smaller amount of information, parts of the signal with little change are sensitive to changes, and parts of the signal with large changes are sensitive to changes, and even if there is a certain amount of error, it cannot be detected. It is well known that various high-efficiency encoding methods have been proposed in the past that take advantage of the difficult nature of human vision and hearing to reduce the amount of information for each sample.

As a specific example of the above-mentioned high-efficiency encoding method, the signal to be encoded is divided into blocks of a certain period, and the distribution of signal levels of the signals in each divided block is divided into blocks. can be converted (normalized or normalized) so that each block has a substantially common state.

According to the conventional method, the distribution of signal levels within a block that includes a large portion of signal level changes in the signal to be encoded is widely distributed within the block;
In addition, the distribution of signal levels within a block that includes parts with little change in signal level in the signal to be encoded is biased toward some signal levels, so areas with no signal are deleted. In the end, the signal to be encoded is in a compressed state, so the signal obtained with this conventional method is the quantization representative point when quantizing the signal to be encoded. Even if quantization is performed using fewer quantization representative points (grayscales) than the number of quantization representative points (grayscales), the error will be large for signals that are inversely transformed in the decoding system for parts with large changes, but decoding for parts with small changes Since the error in the signal that has been inversely transformed by the system is reduced, there are fewer problems in view of the aforementioned characteristics of human vision and hearing.

When encoding is performed using the conventional method described above, information on what kind of conversion (normalization) has been performed on the signal for each block, that is, information indicating the state of normalization (standardization However, as mentioned above, since the amount of quantization information for each sample can be reduced, it is possible to reduce the amount of information as a whole by increasing the size of the block to a certain extent. be.

By the way, depending on the type of signal, the blocks obtained by dividing the signal to be encoded into certain intervals may be one-dimensional blocks such as audio or image scanning lines that appear on the time axis, or blocks that are obtained by dividing the signal into certain intervals. Two-dimensional blocks can be considered, such as when handling from both the horizontal and vertical directions, and three-dimensional blocks can also be considered, such as in the case of moving images, where the time axis direction is also considered. Something like this is known.

“1” When normalizing the signal within a block using the maximum and minimum sample values within the block, Yi=k(Xi−Xmin)/(Xmax−Xmin)...(1) However, i=1, 2, 3,...N Xi is the i-th input sample value (1 dimension) Yi is the i-th normalized sample value (1 dimension) Xmin is the minimum sample value within the block Xmax is the minimum sample value within the block Maximum sample value N is the number of samples in the block k is the standardization constant According to equation (1) above, the i-th standard when normalization is performed using the maximum and minimum values of the sample values in the block. The converted sample value Yi is i
When the th input sample value Xi is the minimum sample value Xmin in the block, it becomes 0, and
When the th input sample value Xi is the maximum sample value Xmax in the block, it becomes k, so all normalized sample values Y1 to Yn for the N input sample values X1 to Xn in the block are between 0 and k. It will take the value of

"2" When normalizing the signal within a block using the average value of the sample values within the block, Yi=k(Xi−Xmean)/D...(2) However, i=1, 2, ^3, . . . _N

1 Maximum value of D=｜Xi−Xmean| ……(3) 2 D=(1/N) _N 〓 ⁱ⁼¹ (Xi−Xmean) ² ……(4) 3 D=(1/N) _N 〓 ⁱ⁼¹ |Xi−Xmean| ...(5) Here, it is assumed that the block is one-dimensional.

Now, the sequential N samples corresponding to each sequential fixed interval in a one-dimensional signal are
When each sample belongs to one sequential block, according to the above equation (1), that is, Yi=k(Xi−Xmin)/(Xmax−Xmia)...(1), the above As shown in ``1'', when each sample value Xi within a block is normalized using the maximum value Xmax of the sample values within the block and the minimum value Xmin of the sample values within the block, each The sample value of is normalized to 0 if the sample value corresponds to the minimum signal level in each block before normalization, and the sample value corresponds to the maximum signal level in each block before normalization. A normalized signal in which the samples are normalized to k is obtained.

Regarding the sample value of each sample belonging to each block obtained by dividing the discrete signal to be encoded into certain intervals as described above, the equation (1) described above is calculated for each individual block. FIG. 2 shows an example of the configuration of a normalizer used to normalize the distribution in the amplitude direction of the signal according to the following, and to perform the normalization process for each block.

In FIG. 2, 1 is the input terminal of the discrete signal to be encoded, and 2 to 5 are the time intervals of the aforementioned discrete signal (sampling period used to generate the discrete signal). 6 is a maximum value detection circuit, 7 is a minimum value detection circuit, 8 and 9 are quantization circuits, 10 to 1
5 is a subtracter, 16 to 20 are dividers, 21 and 22 are output terminals for information indicating the normalization state (data indicating the normalization state), and 23 to 27 are output terminals for quantized data; The discrete signals to be encoded that are supplied to the terminal 1 are sent to the delay circuits 2 to 2, which are equal to the time interval of the discrete signals (sampling period used to generate the discrete signals).
5 series circuits.

Then, the discrete signal X1 is output from the delay circuit 5 described above.
is output and given to the maximum value detection circuit 6, the minimum value detection circuit 7, and the subtracter 11, the discrete signal X2 is output from the delay circuit 4 described above.
is outputted and given to the maximum value detection circuit 6, the minimum value detection circuit 7, and the subtracter 12, and the discrete signal X3 is outputted from the delay circuit 3,
They are maximum value detection circuit 6 and minimum value detection circuit 7.
Further, the delay circuit 2 outputs a discrete signal Furthermore, the discrete signal Xn supplied to the input terminal 1 is supplied to the maximum value detection circuit 6, the minimum value detection circuit 7, and the subtracter 15.

As a result, the maximum value detection circuit 6 detects the maximum value Xmax among the sample values X1 to Xn of N samples in one block and provides it to the quantization circuit 8. Xmax
A quantized signal obtained by quantizing with appropriate precision is supplied to the subtracter 10 as a minuend signal.

Further, the minimum value detection circuit 7 detects the minimum value Xmin among the sample values X1 to Xn of N samples in one block and provides it to the quantization circuit 9. A quantized signal obtained by quantizing with appropriate accuracy is supplied to the subtracters 10 to 15 as a subtraction signal, and is also sent to the output terminal 22.

The signal output from the subtracter 10 is the signal obtained by subtracting the quantized signal for the minimum value Xmin from the quantized signal for the maximum value Xmax, that is, the maximum of the sample values of N samples in one block. The difference between the value and the minimum value (Xmax−
Xmin), and the signal is sent to the output terminal 21 and is sent to the dividers 16 to 20.
is supplied as a divisor signal.

The above-mentioned dividers 16-20 are supplied with the outputs of corresponding ones of the above-mentioned subtractors 11-15 as dividend signals, so the output terminals 23-27 from each of the above-mentioned dividers 16-20 are supplied with the output terminals 23-27. , Yi=k(Xi−Xmin)/(Xmax−Xmin)...(1) i=1, 2,...N Normalized sample value subjected to normalized signal processing according to equation (1) above Y1 to Yn are output.

In this way, the encoder shown in the second diagram above calculates the sample value of each sample belonging to each block obtained by dividing the discrete signal to be encoded into certain intervals. A standardized signal (normalized quantized data) obtained by normalizing the distribution in the amplitude direction of each signal according to the above-mentioned equation (1) for each block is output to terminals 23 to 27. At the same time, output terminal 2
1 and 22, a signal (standardized data) indicating the standardization state of the signal to be set corresponding to the standardized signal is output.

It goes without saying that when decoding a signal encoded in this manner, inverse transformation processing of equations (1) and (2) may be performed.

On the other hand, it is known that adaptive processing is applied to various high-efficiency encoding methods to change encoding parameters, encoding method, data amount, etc. according to the characteristics of each portion of a signal. The adaptive processing means described above is effective when applied to the encoding of images and sounds in which the characteristics of the signal vary greatly depending on each part of the signal.

By the way, when trying to apply the above-mentioned adaptive processing to a high-efficiency encoding method that performs the above-mentioned normalization processing, the difference between the decoded signal and the original signal after the signal in each block is once encoded, Alternatively, after the signal in each block is encoded once, find the mean square error between the decoded signal and the original signal, for example, leave the large error obtained as described above as is, and normalize the small error. A signal processing method in which the quantization after processing is made even rougher, or, contrary to the above case, blocks with small errors are left as is, and blocks with large errors are changed to smaller ones before normalization processing. Signal processing methods that perform quantization and quantization are being considered. and,
When any of the signal processing methods described above is applied, it is possible to reasonably reduce the overall amount of data.

Now, after the signal in each block has been encoded once, the difference between the decoded signal and the original signal is l1,
Furthermore, if the mean squared error between the decoded signal and the original signal after the signal in each block has been encoded once is l2, then the above l1 and l2 can be calculated as follows (6), respectively.
It is shown by equation (7).

l1=(1/N)Σ｜Yip−Xi| ...(6) l2=(1/N)√(-) ² ...(7) However, i=1, 2, 3...N Xi is i Yip is the i-th encoded and decoded sample value (one-dimensional). Here, the size of the block that is the unit of adaptive processing does not necessarily conform to the above-mentioned standard. Although it is not necessary that the size of the block be the same as the size of the block in the standardization process, it is generally a good idea to make the size of the block that is the unit of the adaptive process the same as the size of the block in the normalization process described above. It is considered advantageous. Therefore, in the case of the example, the size of the block serving as the unit of adaptive processing is assumed to be the same as the size of the block in the normalization processing described above.

When the signal processing described above is performed, compared to non-adaptive processing that only performs simple standardization or quantization, less information is generated while keeping the signal error (quality) constant. It is known that signals can be encoded by the amount of data, and an example configuration of a signal processing circuit in the above case is described in the third section.
As shown in the figure. The signal processing circuit shown in FIG. 3 uses a non-adaptive encoder 28 that encodes signals in accordance with the high-efficiency encoding method that performs the normalization processing described above to compare the signals of each block in advance. After encoding a large amount of data, the data amount is reduced, and the normalized signal is adaptively processed by a plurality of quantizers. The signals of each block are quantized by a quantizer selected to suit the purpose of the block, and processing is performed such that each block has a different amount of data.

That is, the input signal is subjected to similar non-adaptive encoding on the signals of each block in the non-adaptive normalizer 28 in FIG. The quantized signal is supplied to the quantizer 29 in the adaptive processing determination signal forming circuit 34, and is also supplied to the adaptive quantizer 33 for performing adaptive processing.

Further, the normalized data of the output signal from the non-adaptive standardizer 28 is supplied to the decoder 30 in the adaptive processing determination signal forming circuit 34, and the standardized data is decoded by the decoder 30. The converted data is supplied to the error calculation circuit 31. An input signal is also supplied to the error calculation circuit 31, and the error calculation circuit 31 calculates the input signal (original signal) and the signal decoded by the decoder 30. The errors are compared and calculated, and the obtained error amount is supplied to a mode determination circuit 32 for determining the processing mode for each block in adaptive processing.

The mode determination circuit 32 described above determines a processing mode according to the amount of error with respect to a predetermined reference value, and supplies this processing mode data to the adaptive quantizer 33. The adaptive quantizer 33 performs quantization on a different amount of data for each block of the normalized signal supplied from the non-adaptive normalizer 28. As a result, the data finally output is rationally reduced, and a highly efficient encoding processing method with an even higher data compression rate can be obtained.

(Problems to be Solved by the Invention) Now, in the conventional adaptive high-efficiency encoding method explained with reference to FIG. 3, what adaptive processing is performed on the signal of each block even in the decoding operation? Since it is necessary to change the mode of decoding processing according to Therefore, the smaller the block and the more types of adaptive processing there are, the more information will be included in the total data.

Therefore, if you try to perform more ideal adaptive processing by making the block smaller or increasing the number of types of adaptive processing, the amount of data in the adaptive mode will increase. Even if the original encoded data could be reduced, the overall amount of data that must be transmitted could not be reduced much due to the adaptive mode data required for each block.

In addition, since the data in the adaptive mode is different from the original encoded data, the formatting of the transmission signal becomes complicated.Furthermore, it is necessary to provide a decoding circuit and an error calculation circuit within the encoding device, which increases the scale of the device. The problem was that it ended up being quite large.

(Means for Solving the Problems) The present invention provides, for signals belonging to individual blocks obtained by dividing a discrete signal to be encoded into fixed intervals, for each of the aforementioned individual blocks.
Non-adaptive normalization means performs standardization processing regarding the degree of change in signal amplitude, and the standardized signal encoded with high efficiency for each block by the above-mentioned non-adaptive normalization means and the individual means for adaptively quantizing the normalized signal in the normalized data required for decoding of each block by an adaptive quantizer, and The above-mentioned normalized signal is obtained by converting the normalized data between the high-efficiency coded signal and the normalized data necessary for decoding each block into the above-mentioned normalized signal by the above-mentioned adaptive quantizer. The present invention provides an adaptive high-efficiency encoding system that is used as data for switching the type of quantization method that involves a change in the number of quantization bits to be applied to the data.

(Example) Hereinafter, the specific contents of the adaptive high-efficiency encoding method of the present invention will be explained in detail with reference to the accompanying drawings. FIG. 1 is a block diagram of an embodiment of the adaptive high-efficiency encoding method of the present invention. In FIG. This non-adaptive normalizer 28 encodes each block of the input signal that is targeted for high-efficiency coding, and converts the signal to be normalized and the normalized data. Outputs .

The standardized signal output from the non-adaptive normalizer 28 is supplied to the adaptive quantizer 33, and the normalized data output from the non-adaptive normalizer 28 is
It is output as standardized data and also supplied to the mode determination circuit 32.

In the adaptive high-efficiency encoding system of the present invention shown in FIG. However, among the output signals that are highly efficiently coded and output processed in the non-adaptive normalizer 28, the normalized data required for decoding is is also used to determine the mode of adaptive processing, so that adaptive processing can be performed without transmitting other mode information.

That is, in the adaptive high-efficiency encoding system of the present invention shown in FIG.
In step 8, non-adaptive encoding processing is performed on a block-by-block basis to obtain information representing the normalized state and a standardized signal of sample values that are standardized and quantized with relatively high accuracy. State information representing the normalization state is transmitted as normalization data, and is also sent to a mode determination circuit 32 that determines the mode of adaptive processing.
and compared with preset criteria.

Examples of the information indicating the normalization state described above include the following.

l3=Xmax−Xmin...(8) However, Xmax is the maximum value within the block Xmin is the minimum value within the block When the sample values converted by are uniformly quantized, the error amount is proportional to the amount of error caused by quantization. On the other hand, when the above-mentioned Xmax-Xmin is small, even if the quantization is made rough to some extent, the basic error amount is small and it is unlikely to cause a problem. Therefore, even if the adaptive processing mode is determined based on Xmax-Xmin, the adaptive processing is effective.

This determines what kind of processing is to be performed on the signal of each block, so the processing mode is set such that the data of each normalized and quantized sample value is based on Xmax - Xmin, which represents the degree of change within the block. Therefore, data is reduced and output. The data obtained as a result is only normalization information and information on each sample value, and there is no increase in the types of data. On the other hand, since the information on each sample value is rationally reduced through adaptive processing, the overall amount of data is certainly reduced by that amount.

Now, the issue here is how effective the normalization information used for determination is for adaptive processing. This is considered to be evaluated based on how close the final decoded signal is to the original data for the same amount of data, but the evaluation method is, for example, the mean square error. In this case, adaptive processing using standardized information cannot be said to be optimal, and a method such as the conventional example is a better method in that respect.

However, in actual visual and auditory characteristics, it is known that the quality of images and sounds is not determined by the degree of mean square error. Since the purpose is to accurately grasp the properties of the block, it is considered that it can be used satisfactorily as a standard for adaptive processing. Especially mentioned above
When using Xmax or Xmin, processing is performed on the peak value of the block, and even if the average error increases, the peak of the error is suppressed, resulting in highly efficient coding where deterioration is difficult to detect. Become.

(Effects of the Invention) As is clear from the detailed explanation above, the adaptive high-efficiency encoding method of the present invention is capable of dividing the discrete signal to be encoded into individual sections. The non-adaptive normalizing means performs normalization processing on the degree of change in the amplitude of the signal for each block, and the non-adaptive normalizing means performs standardization processing on each block. means for adaptively quantizing the standardized signal in a highly efficiently encoded state and the standardized data required for decoding each individual block using an adaptive quantizer; The normalized data of the signal to be standardized that has been highly efficiently coded for each individual block by the non-adaptive normalizing means described above and the normalized data that is required for decoding each individual block is Adaptive type that is used as data for determining the switching of the type of quantization method accompanied by a change in the number of quantization bits to be applied to the above-mentioned signal to be standardized by the adaptive type quantizer of The high-efficiency encoding method is a high-efficiency encoding method that divides a discrete signal into certain blocks and performs standardization (normalization) processing so that the distribution in the amplitude direction is uniform. By effectively using information representing the state of the encoded data and adaptively reducing the amount of data for each block, it is possible to reduce the amount of data in a rational manner without requiring the transmission of data representing an adaptive processing mode different from the original encoded data. Can be processed. In this way, the adaptive high-efficiency encoding method of the present invention not only makes it possible to reasonably reduce output data, but also prevents data from increasing even if the blocks of adaptive processing are made smaller or the number of processing modes is increased. As a result, even if the signal quality is kept constant, the amount of data can be significantly reduced. On the other hand, there is no increase in the types of data to be transmitted, and there is no need for a decoder or an error calculation circuit inside the encoding system as in the conventional example. In particular, since the mean square error calculation circuit has become quite complex and large-scale, the adaptive high-efficiency encoding method of the present invention has the advantage of significantly simplifying the scale of the device. As mentioned above, the number of bits for adaptive quantization processing is reduced by adaptively quantizing the standardized signal obtained by high-efficiency encoding using a non-adaptive normalizer that normalizes the input signal in units of fixed blocks. However, this point is difficult when adaptive quantization processing such as orthogonal transformation that increases the number of signal bits is used, or when vector quantization is essential for normalization. This is particularly advantageous when adaptive quantization is performed.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the adaptive high-efficiency encoding system of the present invention, FIG. 2 is a block diagram of an example configuration of a normalizer, and FIG. 3 is a block diagram of a conventional example. 1... Input terminal for the discrete signal to be encoded, 2 to 5... Input terminal for the signal to have a delay time equal to the time interval of the discrete signal (sampling period used to generate the discrete signal) Delay circuit to give, 6... Maximum value detection circuit, 7... Minimum value detection circuit, 8, 9...
Quantization circuit, 10-15...Subtractor, 16-20
...Divider, 21, 22...Output terminal for information indicating the normalization state (normalized data), 23-27...
... Quantized data output terminal, 28 ... Non-adaptive normalizer, 29 ... Quantizer, 30 ... Decoder, 31
...Error calculation circuit, 32...Mode judgment circuit, 3
3...Adaptive quantizer, 34...Adaptive processing determination signal forming circuit.

Claims (1)

[Claims] 1 For the signals belonging to individual blocks obtained by dividing the discrete signal to be encoded into certain intervals, normalization processing is performed on the degree of change in the amplitude of each signal for each individual block. a non-adaptive standardization means for performing decoding, a signal to be standardized that has been highly efficiently coded for each block by the above-mentioned non-adaptive standardization means, and standardized data required for decoding each block. means for adaptively quantizing the standardized signal by an adaptive quantizer; Changes in the number of quantization bits to be applied to the normalized signal by the adaptive quantizer between the normalized data required for decoding each block and the normalized data An adaptive high-efficiency encoding method that is used as data for switching the type of quantization method.