JP5180782B2 - Parallel distributed information source encoding system and parallel distributed information source encoding / decoding method - Google Patents

Description

  The present invention relates to a parallel and distributed information source encoding apparatus and method, and more particularly, when encoding an information source such as an audio signal or an image signal independently, efficient encoding efficiency equivalent to that of non-independent encoding is efficiently used. The present invention relates to a parallel and distributed information source encoding device and a parallel and distributed information source encoding / decoding method.

  In recent years, digitalization of images has progressed such as the start of digital television broadcasting. Since digitized image data has a large amount of information, it is desirable to compress it for storage and transmission. Currently, international standard standards such as JPEG and MPEG are widely used as image compression methods.

  On the other hand, remarkable development of image applications in recent years is high definition. A symbolic event is the establishment of digital cinema standards. In the digital cinema, high definition images (2048 × 1080, 4096 × 2160 pixels) called “2K” and “4K” are defined as the highest standards for the purpose of obtaining sufficient resolution on a large screen. . For this reason, JPEG 2000 has been adopted as an image encoding method in digital cinema, and it has been standardized so that content with an appropriate resolution can be distributed to movie theaters having different large screens.

  Thus, network distribution of ultra-high-definition images is becoming a reality due to the standardization of digital cinema. Assuming that the digitization of movies further progresses, a new business is expected in which high-quality digital content other than movies (ODS: Other Digital Stuff / Online Digital Source) is distributed live over the optical network to the cinema. However, in order to realize a new application such as ODS, it is necessary to process high-quality content in real time. For this purpose, there is a restriction that encoding must be performed in parallel and distributed processing. In particular, in recent years, such large-scale processing is performed by an integrated circuit, and internal components of the integrated circuit cannot be observed from the outside, and it is necessary to perform control with a small number of parameters. In general, these restrictions cause a reduction in encoding efficiency in image compression.

  Bit allocation can be considered as a parameter that has a particularly large influence on the coding problem. A general optimal bit allocation problem is defined as "a problem of assigning a given bit to various encoders so that encoding performance is maximized". When considering the transform coding system using JPEG2000 as described above, the energy of the noise signal in the decoded signal is determined by how much bits are assigned to which transform coefficient with the bit amount Rc given to the whole as a constraint. It is replaced with the question “whether d (R) is minimized”. This is formulated as follows by using the Lagrange unsteady number method as a problem of minimizing the Lagrangian function J.

min {J}, where Jd (R) + λR (1)
Assuming the whiteness of the error signal, d (R) in equation (1) has the following relationship with the error energy d (R _b ) generated in each subband:

The average bit amount R allocated to the whole and the average bit _Rb amount allocated to each subband have the following relationship.

However, equation (2), the ratio between the subband number of images and the total number of pixels in ηb each sub-band in (3), G _b is the conversion gain of the decoding side in each subband.

  The optimal bit allocation to each subband in the transform coding system may be bit allocation so as to satisfy λ given by solving equation (1), and from equations (1), (2), and (3), Given as follows:

−λ = G _b d ' _b (R _b ) (4)
Also, d _b (R _b ) and subband coefficient energy values

Since the relationship of

After all, the optimal bit allocation to each subband is given by

However, ε ² in the equations (5) and (6) is a discriminative strain constant, and is defined by the following equation using the normal probability density function fx (x) of the signal.

  As described above, in recent years, parallel distributed processing has been demanded as the amount of information handled increases. However, in information source coding, these restrictions generally cause a reduction in coding efficiency. In particular, in recent years, an encoding apparatus is configured by combining a plurality of integrated circuits specialized for one process, and observation of data in each integrated circuit is limited by specifications. Therefore, it is difficult to optimize the encoding efficiency when an independent encoding process is performed.

  It is known that the methods of Non-Patent Documents 2 and 3 can be efficiently encoded even if the encoding process is performed independently if assumptions regarding the distribution and characteristics of the signal are satisfied. According to (6), bit allocation cannot be performed, or a situation occurs in which the signal distribution and characteristics do not satisfy the assumptions in Non-Patent Documents 2 and 3.

  The present invention has been made in view of the above points, and a parallel distributed information source encoding system and parallel distributed information capable of obtaining a coding efficiency equivalent to that at the time of non-independent coding although the signals are independently coded. It is an object to provide a source encoding / decoding method.

  FIG. 1 is a principle configuration diagram of the present invention.

The parallel distributed information source encoding system of the present invention (claim 1 ) is a parallel distributed information comprising a plurality of encoders that independently encode each divided signal obtained by dividing an information source signal including an audio signal and an image signal. A source encoding device;
A parallel distributed information source decoding device comprising a plurality of decoders for independently decoding each signal encoded by the parallel distributed information source encoding device;
Have
Parallel Distributed source coding KaSo location is
Data dividing means for dividing the information source signal into arbitrary blocks;
Quantization parameter deriving means for deriving a step size Δb which is a scalar quantization parameter using an arbitrary reference quantization step size Δ0 as a reference ;
A plurality of parallel distributed information source encoding means for independently encoding the divided signals divided by the data dividing means based on the scalar quantization parameter derived by the quantization parameter deriving means;
By deriving an allocation code amount to be allocated to each block while maintaining the rate amount of each block at a constant ratio, using the rate of the encoded data output from each of the plurality of parallel distributed information source encoding means as a feature amount , seen including a rate amount deriving means for feeding back to each of the parallel distributed information source coding means corresponding to the respective assigned code amount assigns the code amount of the block at any rate amounts to,
In the quantization parameter deriving means, the sum of the square norms of the filter coefficients, which are the conversion gains of the decoding side derived for each band and color component of the divided signal, is Gb, and the step size Δb is obtained by the following equation:


Each of the plurality of parallel distributed information source encoding units re-encodes the divided signal based on each allocated code amount fed back from the rate amount deriving unit.

The present invention (Claim 2 ) is a parallel distributed information source encoding / decoding method for independently encoding / decoding image data,
In a system having a parallel distributed information source encoding device and a parallel distributed information source decoding device,
A parallel distributed information source encoding device, wherein the parallel distributed information source encoding device divides the information source signal including the audio signal and the image signal by using a plurality of encoders and independently encodes each divided signal; ,
A parallel distributed information source decoding apparatus, wherein the parallel distributed information source decoding apparatus independently decodes each signal encoded in the parallel distributed information source encoding step using a plurality of decoders;
And
The parallel distributed information source encoding step is:
A data division step for dividing the information source signal into arbitrary blocks;
A quantization parameter deriving step for deriving a step size Δb, which is a scalar quantization parameter, using an arbitrary reference quantization step size Δ0 as a reference, and a scalar quantum derived by the quantization parameter deriving step in each encoder An encoding step for independently encoding the divided signals divided in the data division step based on the encoding parameters;
By deriving the allocated code amount to be allocated to each block while maintaining the rate amount of each block at a constant ratio, using the rate of the encoded data output from each of the encoding steps in each encoder as a feature amount, A rate amount derivation step that allocates a code amount in units of blocks at an arbitrary rate amount and feeds back the assigned code amount to each of the encoding steps in the corresponding encoders;

In the quantization parameter derivation step, the sum of the square norms of the filter coefficients, which are the conversion gains on the decoding side derived for each band and color component of the divided signal, is Gb, and the step size Δb is expressed by the following equation: Seeking
In each of the encoding steps in each encoder, the divided signal is re-encoded based on each allocated code amount fed back in the rate amount deriving step.

  In the present invention, by using scalar quantization capable of rate control without directly observing wavelet coefficients in a distributed environment, the wavelet coefficients can be quantized.

Can be expected, and the bit allocation system can be improved. Furthermore, unlike the optimization using EBCOT, scalar quantization can be optimized in a distributed environment that does not observe each other.

  The bit allocation ratio can maintain the coding efficiency even if the bit allocation at any target rate is approximated by linear regression with the origin or a multi-order curve passing through the origin.

  Although the present invention can be implemented without depending on input data, a case where the present invention is applied to image coding will be described as an example. When an image is used as an input signal, JPEG, JPEG2000, MPEG-2, H.264, and MPEG SVC (Scalable Video Coding) are widely used as still image coding techniques in information source coding devices. Data compression is performed and encoded data is output. The present invention can divide the processing in the information source encoding apparatus into arbitrary block units, and can encode each of the divided block units independently and with high efficiency. Further, the information source decoding apparatus can perform decoding independently according to the encoding algorithm used in the encoding apparatus. That is, although encoding processing is performed in parallel in units of blocks, encoding with less decrease in encoding efficiency is realized.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[First Embodiment]
In this embodiment, a case where color components such as a luminance signal and a color difference signal are encoded independently will be described.

FIG. 2 is a basic configuration diagram of the parallel distributed coding apparatus according to the first embodiment of the present invention.

  The system shown in the figure includes a parallel / distributed information source encoding apparatus 100, a communication module 110, a communication path 120, a receiving module 210, a parallel / distributed decoding apparatus 200, and a decoding unit 220.

  The parallel / distributed information source encoding apparatus 100 shown in FIG. 1 includes a data dividing unit 101, parallel / distributed encoding units 102, 103, and 104, a quantization parameter deriving unit 105, and a rate amount measuring unit 106.

  The data dividing unit 101 divides input data into arbitrary block units.

  Based on the coding parameters derived by the quantization parameter deriving unit 105, the parallel distributed coding units 102, 103, and 104 perform coding processing independently at each parallel distributed coding unit. In this embodiment, a JPEG2000 processor is assumed as parallel distributed encoders 102, 103, and 104.

  The quantization Δ (step) parameter deriving unit 105 derives a quantization parameter necessary for the parallel distributed coding units 102, 103, and 104 to perform coding.

  The rate amount measuring unit 106 measures whether the encoded data output from the parallel distributed encoding units 102, 103, and 104 is within an arbitrary rate amount.

  The communication module 110 sends the encoded data output from the rate amount measurement unit 106 to the communication path 120.

  The communication path 120 is a channel that is a line such as the Internet.

  The receiving module 210 receives data transmitted from the communication module 110.

  The decoding unit 220 is a general non-parallel distributed decoding unit.

  The parallel distributed information source decoding apparatus 200 includes parallel distributed decoding units 201, 202, and 203 and a data synthesis unit 204.

  The parallel distributed decoding units 201, 202, and 203 perform decoding in units of blocks that are divided independently.

  The data synthesis unit 204 synthesizes the divided blocks.

  Hereafter, each said component is demonstrated in detail.

  This system is a parallel distributed processing system that can perform coding efficiently without observing each other's blocks from the input signal. A method for performing parallel distributed coding processing of components in units of blocks will be described. However, the input signal can target not only video signals but generally redundant signals such as audio signals, and the block unit is not only color components but also multi-view images and arbitrary areas of images. Can be targeted.

  Now, assuming that an RGB component image has been input, the data dividing unit 101 converts the RGB component into a YCbCr component according to the following expression and divides the image into components.

This color conversion can be expected to improve the compression rate by removing the RGB color correlation.

  The divided YCbCr color components are independently encoded by the parallel distributed encoding units 102, 103, and 104. In the parallel distributed encoders 102, 103, and 104, general encoding processing is performed. That is, each input component signal is subjected to band division such as discrete cosine transform and wavelet transform in each parallel distributed coding section 102, 103, 104, and after being converted into transform coefficients, it is quantized by a quantizer. Entropy coding is performed. As this quantizer, for example, in JPEG2000 code, it is performed by multiplying a subband coefficient by a certain step size. Specifically, scalar quantization is performed using a step size given by the following equation in a scalar quantizer (not shown).

Where ε and μ are arbitrary non-negative integers and are used for normalized subband signals. The step size, which is a quantization parameter in each of the parallel distributed encoding units 102, 103, and 104, is calculated by the quantization Δ derivation unit 105 using the following equation.

However, △ ₀ is an arbitrary quantization step size as a reference, usually those of the lowest frequency subband is used. Further, G _b is derived by a decoding-side conversion gain (sum of square norms of filter coefficients) derived for each band and color component.

  The rate amount measuring unit 108 determines whether the encoded data has a rate amount suitable for the communication capacity. If the rate amount is not suitable, the information is fed back to the quantization Δ deriving unit 105, and encoding is performed again. If it is an appropriate rate amount, packetization or the like is performed in the communication module 110 so as to be suitable for communication, and the packet is sent to the reception module 109 through the communication path 120.

  The received data is decoded by the parallel distributed information source decoding device 200 or the non-parallel information source decoding device 220. In the decoding unit 221 in the non-parallel information source decoding apparatus 220, the reverse operation of the encoding unit, that is, entropy encoding is decoded, transform coefficients are returned by inverse quantization, discrete cosine transform, wavelet transform, color transform, etc. Is solved and output as decoded data.

  In addition, in the decoding units 111, 112, and 113 in the parallel distributed information source decoding device 200, the decoding processing is performed in parallel in units of divided blocks, as in the non-parallel information source decoding device 200. Each decoded block data is combined by the data combining unit 204 and output as decoded data.

[Second Embodiment]
In the present embodiment, only the parallel / distributed information source coding apparatus is different from the first embodiment. Therefore, only the parallel / distributed information source coding apparatus will be described below, and the other description will be omitted.

FIG. 3 shows the configuration of a parallel distributed information source encoding apparatus according to the second embodiment of the present invention.

  The parallel / distributed information source encoding apparatus shown in the figure includes a data dividing unit 301, parallel / distributed encoding units 302, 303, and 034, a quantization Δ deriving unit 305, and a rate amount deriving unit 306.

  Hereinafter, each component will be described.

The input data is divided from the RGB component into the YCbCr component by the data dividing unit 301 as in the first embodiment. The divided color components are independently encoded by the parallel distributed encoding units 302, 303, and 304. In the parallel distributed encoding units 302, 303, and 304, as shown in the first embodiment, a general image encoding process is performed. During the processing, the quantization is performed using the step size given by the quantization Δ deriving unit 305. At this time, the step size Δ _{0 serving as} the reference of the expression (10) in the first embodiment is as follows. A larger number is set. The encoded data is sent to the rate amount deriving unit 306.

The rate amount deriving unit 306 assigns a code amount in units of blocks at an arbitrary rate amount, using each sent rate as a feature amount. The code amount to each block at an arbitrary rate amount is derived by, for example, linear regression. That is, the rate of each block rate controlled by △ ₀ as a certain reference by maintaining a constant ratio, the code amount allocation to each block at any rate volume is carried out. This is equivalent to linear regression of the output rate of each block with the origin. Or, if the rate of each block rate controlled by different criteria become △ ₀ is obtained, it is approximated by the quadratic regression and multidimensional regression according to the data amount.

  Note that the encoding efficiency increases as the number of dimensions increases. The arbitrary bit allocation amount derived by the rate amount deriving unit 306 is fed back to the parallel distributed encoding units 302, 303, and 304, and the arbitrary code amount given by the PCRD optimization algorithm or the like is encoded. Encoded data is output.

[Third Embodiment]
In the present embodiment, only the parallel / distributed information source coding apparatus is different from the first embodiment. Therefore, only the parallel / distributed information source coding apparatus will be described below, and the other description will be omitted.

FIG. 4 shows the configuration of a parallel and distributed information source encoding apparatus according to the third embodiment of the present invention.

  The parallel / distributed information source encoding apparatus 400 shown in FIG. 1 includes a data dividing unit 401, parallel distributed S (scalable) encoding units 402, 403, and 404 that perform scalable encoding such as JPEG2000, and a quantization Δ (step) deriving unit. 405, a rate amount deriving unit 506, and a post rate control unit 307.

  Hereinafter, each component will be described.

  The input data is divided from the RGB component into the YCbCr component by the data dividing unit 401 as in the first embodiment. Each divided color component is independently encoded by parallel distributed scalable encoding sections 402, 403, and 404. For example, if encoding is performed from JPEG2000, the code amount cut-off point is marked for each layer using the scalability function of JPEG2000. At this time, rate control is performed with the step size sent from the quantization step deriving unit 405 so that the amount of codes can be known in a later process. In other words, the code truncation point is marked using the scalability function so that the total code amount in units of blocks can be controlled in the scalar quantization of JPEG2000, and any code amount below that rate amount can be controlled by post-rate control. .

  The rate amount deriving unit 406 derives bit allocation corresponding to the target rate from the encoded streams output from the respective parallel distributed scalable encoding units 402, 403, and 404. The processing of the rate amount deriving unit 406 is the same as that of the rate amount deriving unit 306 described in the second embodiment, but since only one scalar quantized value can be used in JPEG2000, linear regression (first order approximation) ) Will be allocated.

  The postrate control unit 407 uses the points marked by the parallel distributed scalable coding units 402, 403, and 404 as post quantization points so as to follow the bit rate derived by the rate amount deriving unit 406 by layer truncation. Rate control is performed in post processing, and encoded data is created.

  In addition, the operation | movement of the component of the parallel distributed information source encoding apparatus in the said 1st-3rd embodiment and a parallel distributed information source decoding apparatus is constructed | assembled as a program, parallel distributed information source encoding apparatus, parallel distributed information It can be installed in a computer used as a source decoding device and executed, or distributed via a network.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

  The present invention is not limited to the above-described embodiments and examples, and various modifications and applications can be made within the scope of the claims.

  The present invention is applicable to image data and audio data encoding / decoding techniques.

It is a principle block diagram of this invention. 1 is a basic configuration diagram of a parallel distributed encoding / decoding system according to a first embodiment of the present invention. It is a block diagram of the parallel distributed information source encoding apparatus in the 2nd Embodiment of this invention. It is a block diagram of the parallel distributed information source encoding apparatus in the 3rd Embodiment of this invention.

Explanation of symbols

100, 300, 400 Information source encoding device 101, 301, 401 Data division unit 102, 103, 104, 302, 303, 304 Parallel distributed encoding unit 105 Quantization Δ (step) derivation unit 106 Rate amount measurement unit 110 Communication Module 120 Communication path 200 Information source decoding device 201, 202, 203 Parallel distributed decoding unit 204 Data combining unit 220 Non-parallel information source decoding device 221 Decoding unit 306, 406 Rate amount deriving unit 402, 403, 404 Parallel distributed S (scalable) Encoding unit 407 Postrate control unit

Claims (2)

A parallel distributed information source encoding device comprising a plurality of encoders that independently encode each divided signal obtained by dividing an information source signal including an audio signal and an image signal;
A parallel distributed information source decoding device comprising a plurality of decoders that independently decode each signal encoded by the parallel distributed information source encoding device;
Have
The parallel distributed information source encoding device comprises:
Data dividing means for dividing the information source signal into arbitrary blocks;
Quantization parameter deriving means for deriving a step size Δb which is a scalar quantization parameter using an arbitrary reference quantization step size Δ0 as a reference;
A plurality of parallel distributed information source encoding means for independently encoding the divided signals divided by the data dividing means based on the scalar quantization parameter derived by the quantization parameter deriving means;
Deriving an allocated code amount to be allocated to each block while maintaining the rate amount of each block at a constant ratio, using the rate of the encoded data output from each of the plurality of parallel distributed information source encoding means as a feature amount A rate amount deriving unit that allocates a code amount in units of blocks at an arbitrary rate amount and feeds back each allocated code amount to the corresponding parallel distributed information source encoding unit;
Including
The quantization parameter deriving means obtains the step size Δb by the following equation, where Gb is the sum of the square norms of the filter coefficients that are the conversion gains on the decoding side derived for each band and color component of the divided signal. ,

Each of the plurality of parallel distributed information source encoding means re-encodes the divided signal based on each allocated code amount fed back from the rate amount deriving means. .
A parallel distributed information source encoding / decoding method for independently encoding / decoding image data,
In a system having a parallel distributed information source encoding device and a parallel distributed information source decoding device,
Parallel distributed information source encoding step in which the parallel distributed information source encoding apparatus divides an information source signal including an audio signal and an image signal by using a plurality of encoders and independently encodes each divided signal. When,
A parallel distributed information source decoding step in which the parallel distributed information source decoding device independently decodes each signal encoded in the parallel distributed information source encoding step using a plurality of decoders;
And
The parallel distributed information source encoding step includes:
A data dividing step of dividing the information source signal into arbitrary blocks;
A quantization parameter derivation step for deriving a step size Δb, which is a scalar quantization parameter, using an arbitrary reference quantization step size Δ0 as a reference;
In each encoder, based on the scalar quantization parameter derived in the quantization parameter derivation step, an encoding step for independently encoding the divided signals divided in the data division step;
By deriving an allocated code amount to be assigned to each block while maintaining the rate amount of each block at a constant ratio, using the rate of the encoded data output from each of the encoding steps in each encoder as a feature amount. A rate amount derivation step of allocating a code amount in units of blocks at an arbitrary rate amount and feeding back the assigned code amount to each of the encoding steps in each corresponding encoder;
Including
In the quantization parameter deriving step, the sum of the square norms of the filter coefficients, which are transform gains on the decoding side derived for each band and color component of the divided signal, is Gb, and the step size Δb is obtained by the following equation. ,
Parallel distributed information source coding characterized by re-encoding said divided signal based on each allocated code amount fed back in said rate amount deriving step in each of the encoding steps in each encoder -Decoding method .