JP2009141457A - Encoding system for reversible compression and information medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an encoding system for reversible compression, capable of executing high-speed processing and expecting a high compression rate. <P>SOLUTION: The encoding system for reversible compression is provided with a difference data generating means 28 for generating differential data of two paired data to be processed and storing the generated data into a memory; an encoding means 22 for encoding the data in entropy and storing the encoded data into the memory; and a data selecting means 23 for selecting the data having a high compression rate. The encoding system is further provided with a tree structure data generating means 21 for generating binary tree structure data in which data forming a parent node is differential data of the two data forming paired child data. Entropy encoding is executed for all the data, and the data selecting means 23 selects the data having the high compression rate by entropy encoding from all the data forming the nodes in the binary tree structure data so that all the data to be processed can be calculated using link information of the binary tree structure data. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an encoding system for lossless compression and an information medium.

  A memory in which data to be processed generated for each channel based on an input signal composed of a plurality of channels is stored, a control unit that can access the memory and performs various arithmetic processes, and the processing to be read from the memory Pair generation means for generating as many pairs as possible from data, difference data generation means for generating difference data of each two processed data paired by the pair generation means and storing them in the memory, and all Encoding means for entropy encoding the processed data and difference data and storing them in the memory, and data for selecting data with a high compression rate so that all the original processed data can be calculated from the processed data and the difference data An encoding system for lossless compression (reversible compression) for generating compressed data based on the data selected by the data selection means Encoding apparatus) is known conventionally (Patent Document 1, page 11, Figure 8a).

Since the encoding system for lossless compression in the above document includes data selection means for selecting data with a high compression rate from the processed data and the difference data, the data to be processed is compressed higher than simply entropy encoding. It is possible to get a rate.
Special table 2007-531012 gazette

However, the lossless encoding system described in the above document selects two pieces of data having a high compression ratio by entropy encoding from three pieces of data consisting of difference data and two pieces of processed data that are the basis of the difference data. Therefore, the difference data is often not selected, in that case, the process of forming a pair or generating the difference data becomes an extra process, in terms of compression ratio and compression processing efficiency, Issues remain.
An object of the present invention is to solve the above-mentioned problems of the present invention and to provide a lossless compression coding system capable of high-speed processing and expecting a high compression rate.

  In order to solve the above-described problem, the encoding system for lossless compression according to the present invention firstly includes a memory 3 for storing data to be processed generated for each channel based on an input signal composed of a plurality of channels, Pair generating means 27 for generating as many pairs as possible from the processed data read from the memory 3, and difference data between the two processed data paired by the pair generating means 27 to generate the memory 3 The difference data generating means 28 stored in the memory, the encoding means 22 for entropy encoding all the processed data and difference data and storing them in the memory, and all the original processed data are calculated from the processed data and the difference data. Data selection means 23 for selecting data with a high compression rate as possible, and compressing data based on the data selected by the data selection means 23 In an encoding system for lossless compression, the input signal is composed of three or more channels, the parent node has a pair of child nodes, and the data constituting the parent node consists of two pieces of data constituting the pair of child nodes. There is provided tree structure data generation means 21 for generating binary tree structure data in which the data constituting the lowest layer node that is difference data and has no child nodes is the above-mentioned processed data and stores it in the memory 3, and encoding The means 22 performs entropy coding on all the data constituting the nodes of the binary tree structure data so that the data selection means 23 can calculate all the processed data using the link information of the binary tree structure data. In addition, the compression rate by entropy coding from all the data constituting the nodes of binary tree structure data It is characterized by selecting a higher data.

  Second, the input signal is composed of eight or more channels.

  Third, the input signal is composed of 16 or less channels.

  Fourth, the data to be processed and the difference data are vector data generated by the vector data generation unit 18, the similarity calculation unit 26 for calculating the vector similarity is provided, and the pair generation unit 27 is a vector having a high vector similarity. It is characterized by generating pairs by data.

  The information medium of the present invention is characterized by being encoded by a lossless encoding system.

  Furthermore, the information medium of the present invention is characterized in that an image generated based on compressed data encoded by a lossless encoding system is used as the medium.

  According to the present invention configured as described above, since binary tree structure data is generated by the tree structure data generation unit and hierarchized, more difference data can be acquired. Since the number of selection data candidates increases, the compression rate increases, and the generation process of binary tree structure data is not much different from the process of generating a pair, and thus high-speed processing is possible. In addition, since the compressed data compressed by the present invention cannot be decoded without the link information of the binary tree structure data generated by the tree structure data generation means, the compressed data itself has high security.

  Moreover, the compression rate by entropy coding can be further increased by using the processed data as vector data, calculating the vector similarity by the similarity calculation means, and generating difference data by pairing similar processed data. it can.

  Further, an image is generated from the compressed data, and the data before compression can be obtained by, for example, reading a print medium on which the image is printed with a scanner or the like, converting it into a digital signal, and decoding it. As a result, large-capacity information can be exchanged by using information media such as paper on which images are printed and other image display media without using communication or existing information recording media (for example, floppy disks, CDs, DVDs, etc.). However, since the compressed data can be decoded, the convenience is improved.

  FIG. 1 is a schematic diagram showing a configuration of an encoding apparatus to which the present invention is applied. The encoding device 1 (encoding device for lossless compression, compression device) is equipped with an encoding system for lossless compression, stores a control unit 2 (CPU) composed of a microcomputer or the like, and stores temporary data. A memory 3 (RAM) that can be accessed at high speed from the unit 2, a storage unit 4 (HDD) that stores information, an input unit 6 that receives input signals and the like, an output unit 7 that outputs output signals and the like, and a floppy disk The external recording device 11 used for exchanging information between the disk 8 (information recording medium, information medium) or CD-ROM 9 (information recording medium, information medium) and the storage unit 4 and characters on the paper medium (print medium, information medium) And a printing machine 12 (printer) for printing images and the like.

  FIG. 2 is a block diagram showing the configuration of a lossless compression coding system installed in the present coding apparatus. The encoding system for lossless compression includes a quantization unit 13 that converts an input signal composed of continuous analog values into a quantized digital value, and a quantized input signal (compression target data, digital input signal). And an encoding unit 14 that performs compression processing.

  In the lossless compression encoding system, the analog input signal from the input unit 6 described above is input to the quantization unit 13, and the compressed data encoded by the encoding unit 14 is also described above. The output unit 7 outputs the output signal as an output signal. In addition, the storage unit 4 is configured to store data to be compressed. Correspondingly, the encoding unit 14 is configured to be able to acquire the compression target data stored in the storage unit 4, and can store the compressed data compressed by entropy encoding in the storage unit 4. It is configured.

  The memory 3 described above is used for information exchange between the quantization unit 13 and the encoding unit 14, information exchange between the encoding unit 14 and the output unit 7, and the like. The quantization unit 13 and the encoding unit 14 are used. Is temporarily stored, and is read by the storage unit 4, the lower code part 14, the output unit 7 and the like in the next stage. Similarly, digital input signals, vector data, residual vector data, binary tree structure data, etc., which will be described later, are also stored in the memory 3 by various means or units that generate these data, It is read from the memory 3 by each unit.

  FIG. 3A is a diagram illustrating an example of an input signal related to audio data, and FIG. 3B is a diagram illustrating an example in which the input signal is divided. For example, when the analog signal input from the input unit 6 is an audio signal, the amplitude width changes with time. In this case, the analog signal is quantized by setting n symbols at equal intervals on the horizontal axis indicating the passage of time and storing the amplitude of the analog input waveform for each symbol. Incidentally, the quantization is not limited to the above means.

  At this time, in order to accurately reproduce the waveform of the analog input signal, it is necessary to reduce the interval between the symbols to some extent. However, if the number of symbols increases due to this, division means 17 (see FIG. 5) described later. ), The input signal is divided into a plurality of frames (frame division) so that all have the same number of symbols (n in the illustrated example). One channel is assigned to each of the divided signals. That is, in the example shown in the figure, the input signal is a multi-channel signal composed of three channels.

  FIG. 4 is a diagram illustrating an example of an input signal related to image data. When the x-axis and y-axis are defined on an imaginary plane on which an image is drawn and are orthogonal to each other, the image data includes an input signal whose amplitude changes by displacement on the x-axis and the y-axis. It can be expressed by an input signal whose amplitude changes due to displacement. That is, the image data can be represented by an analog input signal composed of two channels before being divided by the dividing unit 17.

  This input signal is quantized into a digital input signal having n symbols for each channel by means similar to the above-described audio data. At this time, in order to reduce the number of symbols of each of the two channels, each may be further divided into a plurality of parts by a dividing means. Also in this case, frame division and quantization are performed so that the number of symbols of all channels is the same.

  FIG. 5 is a block diagram illustrating a configuration of the encoding unit. The encoding unit 14 includes a data acquisition unit 16 that acquires a digital input signal from the storage unit 4 or the quantization unit 13, a division unit 17 that divides the digital input signal as necessary, and a digital signal that is passed from the division unit 17. Vector data generating means 18 for generating vector data for each channel of the input signal, and residual vector data (residual signal, processed) from vector data generated for each channel by the vector data generating means 18 based on linear prediction analysis 19 Data) generating residual vector data 20 (residual signal generating means), tree structure data generating means 21 for generating binary tree structure data from the residual vector generated for each channel, and entropy coding. Encoding means 22 to perform and data for compressed data from binary tree structure data And data selection means 23 for-option, and a data output means 24 for generating and outputting the compressed data on the basis of the data selected by the data selection means 23.

  The dividing unit 17 divides the digital input signal into a plurality of pieces as necessary. Specifically, the digital input signal is processed by the dividing means 17 so that the digital input signal is composed of channels having the same number of symbols of 3 or more. The number of channels of the final input signal that has undergone the processing by the dividing means 17 is preferably 8 to 16 according to the experimental results described later (see FIG. 9).

The vector data generating means 18 generates vector data for each channel based on the digital input signal. Since the digital input signal is composed of n pieces of data for each channel, it is regarded as an n-dimensional vector, and is specifically expressed by the following equation. In the following formula, x ₁ , x ₂ ,..., X _n correspond to values for each of n symbols.

  The residual vector data generation means 20 obtains residual vector data from the vector data generated by the vector data generation means 18 by the linear prediction analysis 19. The linear prediction analysis 19 is a technique for predicting and outputting current and future data values based on data input in the past, and is expressed by the following equation using p samples in the past.

Here, a ₁ , a ₂ ,..., _Ap are prediction coefficients obtained by linear prediction and quantized for transmission, and “·” represents integerization. Then, the residual vector data e _t is obtained by the following equation.

Residual vector data e _t required for each the channel has a larger deviation for each component (each symbol), the compression ratio at the time of the entropy coding by the coding means 22 becomes higher.

  The tree structure data generation means 21 is configured to generate binary tree structure data from residual vector data obtained for each channel, and a similarity calculation means 26 (similarity calculation means 26) for calculating similarity (vector similarity). A vector similarity calculating unit), a pair generating unit 27 for generating a pair having a high vector similarity from a plurality of residual vector data, and a difference data generating unit 28 for calculating difference data of the paired data. ing.

Similarity calculation means 26, the two vectors _{x = [x 1, x 2} , ···, x n], y = [y 1, y 2, ···, y n] similarity is two The smaller the angle θ formed by the vectors x and y, the smaller the angle θ, and the higher the angle θ formed, By using the following equation, the similarity between the two vectors x and y can be calculated with a small amount of calculation.

  The pair generation unit 27 selects residual vector data having high similarity based on the vector similarity between the residual vector data calculated based on the above formula, and generates as many pairs as possible.

Next, the difference data generation means 28 calculates difference vector data (difference vector, difference data) between the paired residual vector data. For example, two vectors _x = the aforementioned _{[x 1, x 2, ···} , x n], y = [y 1, y 2, ···, y n] difference vector data when it becomes pairs d is obtained by the following equation.

  Then, based on the similarity calculated by the similarity calculation unit 26, the pair generation unit 27 further generates a pair of difference data and difference data, or difference data and residual vector data, whereby the parent node Has a child node of the pair and the data constituting the parent node is the difference data of the two data constituting the child node of the pair and the data constituting the lowermost node not having the child node is the data to be processed Binary tree structure data (link structure of binary tree structure data) is generated.

FIG. 6 is a schematic diagram illustrating an example of binary tree structure data. In the binary tree structure data, the lowest layer node having no child nodes has residual vector data, and in the example shown in the figure, five lowest layer nodes n _{1 to} n ₅ become residual vector data. Since the similarity between the residual vector constituting node n ₁ and the residual vector constituting node n ₂ is high, node n ₁ and node n ₂ are paired, and the residual vector constituting node n ₃ is Since the similarity of the residual vectors constituting the node n ₄ is high, the node n ₃ and the node n ₄ are paired. Among the two nodes n ₆ , n ₇ constituted by the difference data of the pair of nodes and the node n ₅ not paired, a pair is generated by the node n ₅ and the node n ₇ having high similarity, is generated at the node n ₈ by the differential data of a node pair, the node n ₉ is the root node (node with no parent node) is generated by the difference data between the node n ₆ and node n _8.

  Since the binary tree structure data is generated by the simple processing as described above, the processing speed can be increased. When there are k residual vector data (when the input signal is composed of k channels), the number of nodes of the binary tree structure data is k + (k−1 = 2k−1).

  The encoding means 22 performs entropy encoding on all residual vector data and difference data constituting each node of the binary tree structure data generated by the tree structure data generation means 21 described above. In this embodiment, the Huffman code is used as the entropy code. However, the present invention is not limited to the Huffman code, and the same effect can be obtained by using an arithmetic code, a Rangecoder, or the like.

  The data selection means 23 is the same as the number of residual vector data from all residual vector data and differential data constituting each node of the binary tree structure data so that the original digital input signal can be decoded. It is configured to select data.

  FIG. 7 is a processing flow diagram of the data selection means. When the process by the data selection unit 23 is started, the process proceeds to step S1. In step S1, all lowermost nodes (nodes constituted by residual vector data) are set to “temporarily selected nodes”, and all other nodes (nodes constituted by difference data) are set to “unprocessed nodes”. Then, the process proceeds to step S2.

  In step S2, it is detected whether or not there is still a “temporarily selected node” among the nodes of the binary tree structure data. If there is no “temporarily selected node”, the processing by the data selecting means 23 is terminated, and the data constituting the node set in the “selected node” at that time is selected as the data for compression. On the other hand, if the “temporarily selected node” still exists in step S2, the process proceeds to step S3.

  In step S3, an arbitrary “temporary selection node” whose pair node constituting the pair is not “unprocessed node” is selected as a target node, and the process proceeds to step S4. In step S4, it is detected whether or not the data constituting the target node has a lower compression ratio by entropy coding than the data constituting the parent node. If not, the process proceeds to step S5. Proceed to

  In step S5, the target node is set to “selected node” and the parent node is set to “non-selected node”, and the process returns to step S2. In step S6, it is detected whether the counter node is a “temporarily selected node”. If not, the process proceeds to step S7. In step S7, the target node is set to “non-selected node” and the parent node is set to “temporarily selected node”, and the process proceeds to step S8.

  In step S8, it is detected whether or not the parent node set to “temporarily selected node” in step S7 or steps S11, 13, and 14 to be described later is a root node. If it is a root node, the process proceeds to step S9. If it is not the root node, the process returns to step S2. In step S9, the parent node that is the root node is set to “selected node”, and the process returns to step S2.

  If the counter node is “provisionally selected node” in step S6, the process proceeds to step S10. In step S10, it is detected whether or not the data constituting the counter node has a lower compression rate by entropy coding than the data constituting the parent node. move on. In step S11, the target node is set to “non-selected node”, the counter node is set to “selected node”, and the parent node is set to “temporarily selected node”, and the process proceeds to step S8.

  If it is detected in step S10 that the data constituting the opposite node of the target node has a lower compression ratio by entropy coding than the data constituting the parent node, the process proceeds to step S12. In step S12, it is detected whether or not the data constituting the target node has a higher compression ratio by entropy coding than the data constituting the counterpart node. If it is higher, the process proceeds to step S13. Proceed to

  In step S13, the target node is set to “selected node”, the counter node is set to “non-selected node”, and the parent node is set to “temporarily selected node”, and the process proceeds to step S8. On the other hand, in step S14, the target node is set to “non-selected node”, the counter node is set to “selected node”, and the parent node is set to “temporarily selected node”, and the process proceeds to step S8.

  According to the data selection means 23 configured as described above, all residual vector information is calculated by using the information related to the “selected node” and the information related to the link structure of the binary tree structure data. Is possible. If all the residual vector information is known, it is possible to acquire information about the digital input signal by the above-described equation. The configuration of the data selection unit 23 is not limited to the example shown in FIG.

8A to 8D are state transition diagrams showing an example of data selection by the data selection means. In the example shown in the figure, four lowest layer nodes n _{1 to} n ₄ are constituted by four residual vector data, and these nodes n _{1 to} n ₄ are set as “temporary selection nodes” ((A )reference). Then, the tree structure data generation means 21 configures a parent node n ₅ that child nodes the _two nodes n ₁ and n _2, and configures a parent node n ₆ that child nodes the two nodes n ₃ and n _4. it is, by the parent node n ₇ is the root node of the child node the two nodes n _5, n ₆ is constituted, the binary tree structure data is created. The numbers in the nodes indicate the order of the entropy encoding compression rate in the _seven nodes n _{1 to} n ₇ .

First, for each of the two nodes n ₁ and n ₂ , the process of Step S 2 → Step S 3 → Step S 4 → Step S 5 → Step S 2 is performed. As a result, the nodes n ₁ and n ₂ are set to the “selected node”, and the node n ₅ is set to the “non-selected node” (see FIG. 5B).

Next, the process of step S2 → step S3 → step S4 → step S6 → step S10 → step S12 → step S13 → step S8 → step S2 is performed on the node _{n 3,} as a result, the node _{n 3} "Selection is set to node ", a node n ₄ is set to" non-selected node ", node n ₆ is set to" provisionally selected nodes "(see FIG. (C)).

Next, the process of step S2 → step S3 → step S4 → step S6 → step S7 → step S8 → step S9 → step S2 is performed on the node _{n 6,} as a result, the node _{n 6} is "unselected nodes" And node n ₇ is set to “selected node” (see FIG. 4D).

As a result of these processes, node n ₁ , node n ₂ , node n ₃ and node n ₇ are selected as selection data. At this time, in order to obtain the information of the residual vector data constituting the node n ₄ obtains the information of the node n ₅ the information of the node n ₁ and node n _2, information of the node n ₅ and node n ₇ To obtain information on the node n ₆ , and obtain information on the node n _{4 based on} the information on the nodes n ₃ and n ₆ .

  The data output means 24 constitutes a node selected as a “selected node” by the data selection means 23, and is configured to generate and output compressed data with entropy-encoded data. At this time, the data output means 24 also outputs the link data related to the link information of the binary tree structure data so that the compressed data can be decoded together with the compressed data. Decoding is performed using this link data.

  FIG. 9 is a characteristic graph of processing time with respect to the number of channels in the encoding system for lossless compression. As shown in the figure, even when the number of channels is increased and the information to be handled is increased, the processing time is not increased so much while the number of channels is between 8 and 16, and the processing is performed efficiently. When the number of channels becomes 64 or 128, the processing time increases exponentially. That is, in order to perform efficient processing, the number of channels is preferably 8 to 16, and practically, the number of channels is preferably 32 or less.

  In the above example, the residual signal is calculated by the linear prediction analysis 19, but the lossless compression coding system can obtain a sufficient compression rate without using the linear prediction analysis 19, and in some cases The processing by the residual vector data generation means 20 may be omitted. In this case, the data generated by the vector data generating means 18 becomes the data to be processed.

  In addition, the compressed data can be converted into image data by a method reverse to the method shown in FIG. 4 and a method reverse to the dividing unit 17 (in some cases, the method of the dividing unit 17). It is. Compressed data containing a large volume of music, moving image data, etc. by printing it on a paper medium (information storage medium, information medium) or displaying it on a computer display screen (information medium) via the printer 12 Can be exchanged without using a dedicated information recording medium (floppy disk, CD-ROM). For this reason, it is possible to greatly reduce the cost and it can be applied to various scenes in the media industry. Incidentally, the compressed data can be stored in the floppy disk 8 or the CD-ROM 9 via the external recording device 11.

It is the schematic which shows the structure of the encoding apparatus to which this invention is applied. It is a block diagram which shows the structure of the encoding system for lossless compression mounted in this encoding apparatus. (A) is drawing which shows an example of the input signal regarding audio | voice data, (B) is drawing which shows an example in which an input signal is divided | segmented. It is drawing which shows an example of the input signal regarding image data. It is a block diagram which shows the structure of an encoding part. It is a schematic diagram which shows an example of binary tree structure data. It is a processing flow figure of a data selection means. (A)-(D) are state transition diagrams which show an example of the data selection by a data selection means. It is a characteristic graph of processing time with respect to the number of channels of the encoding system for lossless compression.

Explanation of symbols

3 Memory (RAM)
18 Vector data generation means 21 Tree structure data generation means 22 Encoding means 23 Data selection means 26 Similarity calculation means (vector similarity calculation means)
27 Pair generation means 28 Difference data generation means

Claims (6)

  A memory (3) in which processed data generated for each channel based on an input signal composed of a plurality of channels is stored, and as many pairs as possible from the processed data read from the memory (3). Pair generating means (27) for generating the difference data generating means (28) for generating difference data of each of the two processed data paired by the pair generating means (27) and storing them in the memory (3) And an encoding means (22) for entropy encoding all the processed data and difference data and storing them in the memory, and a compression rate so that all the original processed data can be calculated from the processed data and the difference data A data selecting means (23) for selecting high-quality data, and generating a compressed data based on the data selected by the data selecting means (23). The input signal is composed of three or more channels, the parent node has a pair of child nodes, and the data constituting the parent node is the difference data of the two data constituting the pair of child nodes and the child node There is provided tree structure data generation means (21) for generating binary tree structure data in which the data constituting the lowermost layer node having no data is the above-mentioned processed data and storing it in the memory (3), and encoding means (22 ) Performs entropy coding on all the data constituting the nodes of the binary tree structure data, and the data selection means (23) can calculate all the processed data using the link information of the binary tree structure data. Like, data with a high compression rate by entropy coding from all the data constituting the nodes of binary tree structure data Lossless compression encoding system to select.   The lossless compression coding system according to claim 1, wherein the input signal is composed of eight or more channels.   The encoding system for lossless compression according to claim 2, wherein the input signal is composed of 16 or less channels.   The processed data and the difference data are vector data generated by the vector data generating means (18), provided with similarity calculating means (26) for calculating the vector similarity, and the pair generating means (27) The encoding system for lossless compression according to claim 1, 2 or 3, wherein a pair is generated by high vector data.   An information medium storing compressed data encoded by the lossless encoding system according to claim 1, 2, 3, or 4.   An information medium using an image generated based on compressed data encoded by the lossless encoding system according to claim 1, 2, 3, or 4 as a medium.

Images