DE19758853B4 - Entropy encoder for compression and expansion of data - includes series of encoding tables which are coupled to receive state information and to encode bits of input data in response to state information - Google Patents

Abstract

The apparatus for encoding input data includes a context model which is configured to generate a series of binary decisions in response to contexts. A probability estimation model is coupled to the context model to generate probability estimates for the series of binary decisions. A series of entropy encoding tables are coupled to receive state information from a channel state storage. The entropy encoding tables are configured to encode bits of the input data in response to the state information. One of the entropy encoding tables is configured to encode n bits at a time where n is greater than or equal to two.

Description

The The present invention relates to the field of data compression and decompression systems; The present invention relates in particular to a decoder according to claim 1, a system with an encoder and a decoder according to claim 8 and a method for decoding according to claim 9th

The Data compression is extremely useful Tool for saving and transferring greater Amounts of data. For example, the required to transfer an image Time, such as a facsimile transmission a document, extremely strong decreases when compression is used to reduce the number of Reduce bits required to reconstruct the image.

at Some compression systems become an input field or record translated into a sequence of decisions under the guidance of a decision model.

each Decision has an associated probability and based on this probability, generating an output code is appended to the compressed field. To this coded system compression systems have three parts: a decision model, a probability estimation model and a bitstream generator. The decision model receives the Input data and translated the data into a decision set, the compression system used to encode the data. On the decision model becomes common referred to as a Kontektmodell. The probability estimation method is a procedure for developing the likelihood probability estimate every decision. The bitstream generator performs the final bitstream coding through to produce the output code of the compressed data set or the compressed field. The compression can be effective in one or both of the decision models and the bit generator occur.

A Compression technique, which is widely used, is because arithmetic Encode. Arithmetic coding forms a data string (i.e. a "message") on a code string in such a way that the original message from the Kodestring or the code sequence can be recovered. For a discussion of arithmetic coding, see Glenn G. Langdon, Jr., "An Introduction to Arithmetic Coding ", IBM Journal of Research and Development, Vol. 28, No. 2 (March 1984). A desirable one Characteristic of some older ones arithmetic coding systems is that the compression in one single pass sequence over the Data is performed without a fixed set of statistics, to encode the data. This is the arithmetic Coded adapted.

One binary Arithmetic coder is a kind of arithmetic coding system. In a binary Arithmetic coding system can be the selection of a symbol from a List or a set of symbols encoded as a sequence of binary decisions become. An example of a binary Arithmetic Encoder is the Q Encoder developed by IBM by Armonk, New York, has been developed.

The Final machine state (FSM) encoders are used in the art to provide efficient entropy coding for individual bits with associated probability estimation. Some of these FSM encoders are lookup tables (LUTs) has been realized. See, for example, the US patents Nos. 5,272,478 A and 5,363,099 A As an example of a final state machine, which encode channel modulation and error correction with entropy performs, Reference is made to U.S. Patent No. 5,475,388.

in the In general, final state machines that use LUTs are for multi-bit symbols not fast. If e.g. a number between 0 inclusive and 7 is to be coded, the Number can be separated into a minimum of three bits, creating three separate table lookups be required. The summed effort of three separate lookups slows down the coding process. What is needed is that avoiding multiple table lookups be while Multi-bit symbols are encoded.

Huffman coding provides m-fold coding in which a multiple symbol is encoded and / or decoded. Huffman coding produces variable length codes that are integral (not fragmentary) bit numbers are. In other words, there is no time at which the encoder contains information that causes some of the bits that are currently being output. Symbols with higher probabilities receive shorter codes.

The Data encoding and decoding are extremely time consuming operations. In many systems, the probability estimate is under Using a table performed. If both the probability estimate as also the entropy coding are realized as LUTs, become separate lookups in tables without parallelism required. It is desirable separate lookups in tables, if possible, to reduce the amount of time to the probability estimate and To perform the Entropiekodieren according to the prior art.

in the In the prior art, entropy coding has usually been done quickly, but without the best possible Compression, over for example, coding according to Huffman, or is completely adaptive or adapted been, e.g. even slower over arithmetic coding. It is desirable to have such operations or actuations to accelerate while they adapted or remain adaptive.

The The present invention provides increased speed for entropy coding using a final state machine encoder Machine encoder with finite state available, capable of doing so is, inputs to accommodate with n bits. The present invention further provides encoding m-fold symbols, such as Huffman coding, ready, with the exception of a non-integral number of bits (Sketchy or fractionated). The present invention is also more likely a single table lookup for both the probability estimate and also the bit generation as two separate operations for the Available.

The present invention solves the tasks and disadvantages mentioned in the prior art by things the independent one claims 1, 8 and 9. Advantageous embodiments this independent claims are proposed by the features listed in the subclaims.

One m-fold final state machine encoder or machine encoder with Finite state is described. In one embodiment, the encoder according to the present invention, a channel state memory device and an entropy coding lookup table on, the state information from the channel state storage device receives. The entropy coding table encodes n bits of input data a time at the same time in response to the state information from the channel state storage device, where n≥2.

The The present invention will be more fully understood from the detailed the description given below, and from the accompanying drawings of various embodiments of the invention, which, however, should not be used to to limit the invention to the specific embodiments, but for explanation only and for understanding are.

1A represents a final state machine estimator / encoder and a finite state machine estimator / encoder, respectively.

1B represents an embodiment of the FSM encoder.

1C represents a finite state machine state machine / decoder.

2 FIG. 10 illustrates one embodiment of a two-bit FSM encoder at a time capable of handling arbitrary probabilities.

3 Figure 4 illustrates a four-bit FSM encoder at a time constructed for bits of the same probability class.

4 FIG. 12 illustrates one embodiment of a multi-symbol FSM encoder (M-ary or M-fold).

5 represents an encoder that is capable of doing the same operation as after 1A . 3 and 4 by using two additional bits as a table selector.

6 FIG. 12 illustrates one embodiment of a table and coding system estimation. FIG.

7 FIG. 10 illustrates one embodiment of a finite state machine estimator / encoder in which the likelihood estimation state includes the most probable symbol (MPS).

8th Figure 13 shows a coding tree generated by the Huffman algorithm for a particular set of symbols.

One Method and device for compressing and decompressing is described. In the following, detailed Description of the present invention will be numerous specific Details have come to the fore, such as types of encoders, Numbers of bits, signal names, etc., to get a thorough understanding of the to enable the present invention. However, it will be apparent to those skilled in the art that the present invention Invention be practiced without these specific details can. In other instances, well-known structures and devices are known in the art the shape of block diagrams rather than shown in detail to avoid that present invention is made unclear.

Some Portions of the detailed descriptions that follow will become in algorithms and symbolic representations of operations that to be applied to data within a computer memory. These algorithmic descriptions and representations are the Means provided by the data processing experts in the art be used to the essential or the substance of their work to convey most effectively to others skilled in the art. An algorithm is used here and generally as a self-consistent Sequence of steps taken the one to a desired one Result. The steps are of the nature, the physical manipulations of physical quantities or quantities require. Although this is customary is not necessary, the sizes have the shape of electrical or magnetic signals that are stored, transferred, combined, compared and otherwise manipulated or changed can be. It proved to be pleasant some time ago, in principle for reasons general use, to these signals as bits, values, elements, Symbols, character, expressions, To refer numbers or numbers or the like.

It However, it should be noted that all these and similar expressions in conjunction with matching physical sizes are and only are convenient labels associated with these quantities. If it is not expressed differently, as it is made the following discussion It will be appreciated that it is preferable to be aware of the present invention away from discussions, the expressions, such as "processing" or "calculating" or "calculating" or "determining" or "displaying" or the like, use on activities and a method of a computer system or similar electronic computing device refer to the data as physical (electronic) sizes within a register or memory of the computer system or other such information storage, transmission or reproduction devices represents, manipulates or transmits. such Computer systems usually set one or more processors to process data that to one or more memories via one or more buses or data buses are coupled.

The present invention also relates to an apparatus for carrying out the operations and operations explained herein. This device may be specially constructed for the required purposes, or it may have a general purpose computer which is selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a computer readable storage medium such as a kind of disk, floppy disks, optical disks, CD-ROMs and magneto-optical disks, static memories (ROMs), random access memory (RAMs), EPROMs, EEPROMs, magnetic or optical cards, or any type of medium suitable for storing electronic commands, each coupled to a bus of the computer system, wherein the readable storage media are not limited to the aforementioned. The algorithms and reproductions presented herein are not inherently related to any particular computer or other device. Various general purpose machines may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatuses to perform the required method steps. The required structure for a number of these devices will become apparent from the following description. Furthermore, the present invention will not be described with reference to any particular programming language. It is preferable that various programming languages can be used to implement the teachings of the invention described herein to implement the act.

The present invention provides an acceleration for a Finite state machine encoder or machine encoder State available by encoding n bits at a time. With In other words, the present invention provides the coding of a Alphabets of M symbols (where M> = 2) using a machine encoder with finite state or with final state available. Encoding M symbol alphabets results in increased speed and an elevated one Compression function. The present invention also performs a run-coding, Huffman coding and other coding with multiple symbols with a FSM encoder through. The combination of the FSM encoder with the multi-symbol coder represents both the speed of the multi-symbol codes as well The compression feature is available, which typically only possible with arithmetic codes is.

Of the. FSM coder according to the present invention may have several look-up tables Use (LUTs). Every LUT is for the use has been generated in a particular situation. As e.g. described in more detail below, one of the tables for use with the last significant bits of wavelet coefficients typically 50% 1 and 50% 0. Each of the tables is used to different kinds of Perform encoders with respect to the same bitstream. The The present invention uses the encoder in which between them Tables during coding back and forward is switched while the fractional bits or the fractional or broken bits be maintained in the bitstream.

in the In the prior art, the FSM-based coding system can be three Use tables. The first of the tables contains the current probability estimate and the value of the most likely symbol (MPS) for each context, for the one appraisal carried out becomes. The inputs in this table are changed with each symbol in a context is encoded. A second table represents the updates of the probability estimates ready. This table changes not and updates inputs in the context table. The third table is the sentence or the List of states, the legal exits too describes the bit stream. This table does not change and updates the current one Condition for the Encoder variable. The table also indicates what bits are output should be.

Of the Decoder uses the same tables to store probability estimates and for updating probability estimates, but uses a different table for decoding. The third table in the decoder looks for input bits and probability estimates and determines if the MPS occurs determines the next state and determines the number of bits through which the input stream to move to the next one To prepare the decoding operation. An embodiment of the sizes and Structures of these tables are given in Table 1 below.

Table 1

The 1A Figure 12 is a block diagram of one embodiment of a finite state machine estimator / encoder with two tables representing connections between the tables therein. Referring to 1A becomes a context memory 101 that stores the present probability estimate the MPS value for each context for which an estimate is made coupled to a context 110 , a next_mps ad 111 Exclusive-Or (XOR) 102 and a next_p state indicator 112 to recieve. The next_mps ad 111 provides an ad for the context memory 101 available, which is the next MPS. That is, the next_mps ad 111 indicates if the MPS has changed. The next_p state indicator 112 provides an indication of the context memory 101 in terms of what the next probability estimate is. Based on these inputs is on the context memory 101 accessed and the current probability state (p-state) 113 and the current MPS are output.

The current MPS 114 becomes an input of the XOR gate 102 coupled. The other input of the XOR gate 102 becomes an output, the Kipp_mps or the Toggle_mps 115 , the probability estimation table 102 coupled. In one embodiment, if the Kipp_mps signal 115 is equal to "1" and the value of the p-state 113 is less than 215, or an output (output 122 > 0) There is the value of the MPS stored in the context store 101 is changed from "0" to "1" or changed from "1" to "0".

A comparator block 104 is coupled or connected to the current MPS 114 to receive and compare the current MPS 114 with the input bit 116 , If the result of the comparison indicates that the MPS 114 the input bit 116 equals, becomes the "equality" indicator 117 specified. If the result of the comparison indicates that the MPS 114 not equal to the input bit 116 is, becomes the "equality" or "probability" indicator 117 not displayed.

The probability estimation table 103 is connected to the p-state 113 and the equality / probability indicator 117 receive and gives based on these inputs a probability class (p-class) 118 , the Kipp_ or Toggle_mps 115 and the next_p status indicator 112 out. The Kipp_mps 115 indicates whether the MPS should be changed from 1 to 0 or vice versa. The actual probability estimate is represented by a class referred to herein as a P-class. Each P-class is used for a range of probabilities. In one embodiment, each P-class is used for a probability range, rather than using different bits to specify an accurate probability. In this way, only a few Bits are needed to specify in which class area the probability lies. In one embodiment, four bits are used to specify the P-class.

An embodiment of the probability estimation table 103 is shown in Table 2 below:

Table 2 - Admission PEM

The entropy coding table 105 , which represents a finite state machine, is coupled to the P-class 118 , the probability indicator 117 and an output from the channel state 106 and generates 0 or more output bits such as the compressed bit stream. The channel state 106 stores broken bits representing the current state in which the output channel is. In one embodiment, the channel state 106 a register or other storage element that stores state information. In one embodiment, the channel state stores 106 six bits of channel information. The output of the channel state 106 occurs in response to a next status indication, the next_state 119 , on, the from the entropy encoding table 105 is issued.

Based on their inputs gives the entropy coding table 105 one or more bits than the compressed data stream, here as an output bit 121 Reference is made, and an indication of the number of bits in the output, here as an output 122 Reference is made. Note that the output size 122 may only be an indication as to which bits in the output stream have significance. The entropy coding table 1.05 also generates one or more bits to indicate the next state, the next_state 119 Reference is made. The next_state 119 becomes in the channel state 106 as the state input for the next bit (s).

In one embodiment, the entropy coding table 105 a lookup table (LUT) having multiple states, each of the states having one or more pairs of legal transitions. Each transition is defined to cause the storage of zero or more bits to be omitted when the transition occurs and a target state to which the transition transfers during the transition. It is because of this target state that the LUTs continue to process the next symbol.

The 1B FIG. 12 illustrates one embodiment of the FSM encoder. Referring to FIG 1B becomes the entropy coder lookup table 105 connected to six bits of state information from the state register 106 to recieve. Four bits represent the P class and one bit provides the probability indication. in succession, the entropy coder look-up table 105 8 output bits (OUTBITS) ready, 4 bits to indicate the output size (OUTPUT) and 6 bits to indicate the next state.

A bunch of buffers can be used to get the output bits 121 to buffer or buffer. These buffers are not shown to avoid obscuring the present invention.

A embodiment the probability estimation and the FSM encoding table used to describe the entropy coding perform, are shown in Tables 3 and 4 below. Table 3 shows an excerpt of the probability estimation table while the Table 4 extracts from the FSM encoding table.

Table 3 - Excerpts from the Probability Estimate Table

The Table 4 shows only the first and last few rows and is determined by the state, the probability class and LPS / MPF commanded or ordered. Each entry into the table is threefold (next State, output, output bits). The third element leaves the number of the output bits that correspond to the compressed output stream should be added, making it possible to distinguish 0 and 00.

Table 4 - Excerpts from the coding table

The 1C Figure 12 is a block diagram of a finite state machine decoder. The context memory is exactly the same as the probability estimation table for the decoder. The entropy decode table has a different set of inputs and outputs as described below. Referring to 1C become compressed bits 130 through a shift register 131 receive which compressed bits for the entropy coding table 115 provides. The entropy coding table 115 emp also captures the channel state from the channel state storage area 106 and a p-class 118 from the probability estimation table 103 , In response to these inputs, the entropy coding table generates 115 an LPS 132 , the next_state 119 (as in the encoder) and the output 121 ,

The LPS 132 is applied to an input of the exclusive OR (XOR) gate 137 connected, that too the current MPS 114 receives. On the basis of these two inputs will be one bit 140 output from the decoder.

The LPS 132 also with the Kipp_MPS 115 that from the probability estimation table 103 is output through the AND gate 136 summed. The output of the AND gate 136 will be with the LPS 132 through the XOR gate 138 an exclusive-or operation to the next_p state indication 112 to generate an input to the context memory 101 represents.

The LPS 132 is also used as a select signal for a multiplexer (mux) 135 used to get either the Next_LPS signal 133 that indicates the next LPS or the Next_MPS signal 132 indicating the next MPS to issue, that of the probability estimation table 103 be issued. The output from the Mux 135 is the next_MPS 111 , which is an input to the context memory 101 is.

The output size or size 121 is from the entropy coding table 115 to control the number of bits to shift the compressed bitstream to the next input to the entropy coding table 115 provide.

Of the Decoder can be removed from the encoder using the below Codes are derived. Given the coding table of the Form of Table 4, where the function maketable (table, 62, 16) too return to a decoding table is determined by the state, the probability class and the compressed Bits (8 bits for the compressed data stream) in the form of Table 5 is.

Dekodererzeugungskode

One Excerpt for the decoding table is shown in Table 5 below.

Table 5 Part of the FSM decoding table

Several bits at several probabilities

An entropy coding table can be constructed that accepts multiple bits (same indications) and multiple probabilities (probability class) at a time. An example of such a table is given in FIG 2 shown. Referring to 2 becomes an entropy-coding lookup table 205 connected to receive 16 input bits representing the 6 bits of state information from the status register 106 , 4 bits representing a first probability class P-class-1, a bit providing a first probability indication Likely1, four bits representing a second probability class P-class-2, and a bit providing a second probability indication Likely2 , The entropy coding table 205 outputs 27 bits having 6 bits representing the next state to the state register 106 16 output bits (OUTBITS) and five bits indicating the size of the output (SIZE). Thus, the entropy coding table encodes 205 two bits for any probabilities at a time.

One Problem with using a single table that has multiple bits for many Can accommodate probability classes at a time, is that one Such table with the number of larger inputs very gets big. Also requires a very big one Table more bits to address the table, indicating its speed reduced.

Several bits at a fixed probability

The present invention provides an increased speed for a Encoder available, by turning multiple bits into a single one at a fixed probability Entropiekiertiertabelle be sent. The fixed probability can have a probability class. However, each brings Table, the n bit inputs (n ≥ 2), only one such class or probability below. It should be noticed that one Table can be built to accommodate different P classes, However, the accommodation of an increased input size requires.

An example of a table that accommodates multiple bits at the same time is a table that handles bits that are at a probability of 0.5 (or 50%). The 3 FIG. 12 illustrates one embodiment of such an entropy coding table. Referring to FIG 3 is the entropy coding table 305 connected to receive 10 input bits, the 6 bits of state information from the state register 306 along with four input bits, Likelyl-4, and outputs 21 bits, the six bits of the next state information corresponding to the state register 306 Therefore, using the fixed likelihood, the size of the input and the output will decrease compared to the table 2 reduced.

at another embodiment receives the coding table several inputs, who have a fixed probability, which is strongly increased, such as e.g. in the upper range of 90% (for example, approximately 98, approximately 99%, etc.).

The FSM encoder of the present invention may be incorporated into a coding system such as that disclosed in U.S. Pat 1A is shown. In one embodiment, the estimation state is updated at pseudorandom intervals. By only updating at pseudo-state intervals, complete counts of all LPSs and MPSs need not be stored. In one embodiment, the estimation state is updated when the entropy encoding table 105 Output bits generated. In one embodiment, the output may be 122 are compared with zero to determine when an output has occurred. This requires that the table contain extra states to compensate for changes in the likelihood that occur between the updates. Likewise, the FSM decoder may also be included in the decoder associated with 1A has been discussed.

table creation

Tables for handling multiple bits at a time, as opposed to individual bits one at a time, can be generated by applying individual bits and monitoring the output bits and the next states that occur when a one-bit-at-a-table is used , Then, new outputs can be generated by concatenating the output bits from a series of states. The next state for the series is the next state generated by the last output bits. This process is made clear by the examples discussed below. The table is often larger and contains inputs that can be determined by sending bits one at a time by the current entropy encoder. That is, specific tables can be constructed by the finite state machine estimator / decoder 1A is used. In the entropy coding tables constructed with multiple bits or multiple bits in this manner, the present invention maintains compatibility with binary coding tables and allows the tables to be switched at any point, even with fractions of bits.

To the An example of an exemplary FSM encoder is shown in Table 6 below shown.

Table 6 - FSM Encoder

FSM encoder Two-two-bit encoder at a time generates a next state and an output in response to inputs representing the current state, have two probability classes and two results. Around to determine what the next one State and the output should be the output of one Encoder with one bit at a time and the last state after used to encode two bits. In the case of Table 1 e.g. is to assume that the Encoder with one bit at a time in the state is 0 and a MPS is to be coded in class 1 followed by an MPS in class 0. The first bit causes the encoder with one at a time, no bits to spend and to move to state 1. The second bit is encoded in state 1 and causes a 0-bit to be issued and returns to the 0 state. Of the Exit the new table for this input would be with two bits 0, which is linked to the concatenation of the non-bit the first bit and the 0-bit output, that with the second Bit linked becomes. The new next State in the encoder encoding two at a time would be for the input with two bits, the state is 0. A new table, several (2) Handle bits at a time supplied by the FSM encoder with a State as shown in Table 6 is shown in Table 7 below.

Table 7

One An example of a 3-bit class 0 encoder is shown in the table below 8.

Table 8 - 3-bit Class 0 Encoder

It should be noted that the Table production method of the present invention used can be used to create tables of different sized inputs. However, it will a condition for each possible Input bit combination needed.

It is often desirable to make tables, such as Table 7, faster. However, to accelerate a 64-state table having 16 probability classes to four times as fast, the table would need to have 2 ²⁶ inputs instead of 2 ¹¹ inputs 2 used by the single-quick table and is not in many situations practical.

M-ary encoding

Laufkodes

at an embodiment be for a quick coding for 50% bits (or 60% bits) and a Golumb or other start step stop codes provided using single bits and 50% bits. Such a Code is in the book or essay by Bell, Cleary, Witten, text Compression, described. Table 9 shows an example of one Start-stop-stop codes:

Table 9

One typical run length code is shown in Table 10. Referring to Table 10 This code is good for a sequence of zeros with a probability from approximate 0.917 to encode.

Table 10

at Such a system usually has two table lookups for up to eight coded bits. In the present invention, the run code in Table 10 using the FSM coding table for 1 and 4 50% bits are realized

An embodiment of the code for implementing the run coder is:

If the maximum sound length (8 in this case), then only one bit is to be output. This is coded by coding a 0 at a fixed probability of 50% done (done by the function FSMCode_bei_50). This encoding becomes approximate However, using a bit will result in arbitrary bits of fractions of maintained prior operations. If a run occurs, which is less than the maximum, then a value of 4 bits must be encoded that starts with a 1. This is done using a lookup table done with one bit at a time, giving access over the Function FSMCode_4_Bits_ is 50. The 50 in the function name refers to the fact that this function assumes that the probability fixed at half or 50%.

One embodiment of the code used to implement decoding is as follows:

Of the Decoder knows in Do not predict when a 4-bit value or a one-bit value must be decoded. So he first decodes a bit at 50% with the call FSMDekodiere_bei_50 (). If this bit is a number 1, then the encoder actually encodes four Bits to describe a run. The remaining three bits are decoded with the call FSMDecode_in_50 () and their value is the encoded run length. If the first decoded bit was a 0, then the maximum occurred Run on and it is not necessary any additional ones To decode bits.

Huffman codes and Huffman-like codes

The The present invention provides a FSM encoder which has a Huffman coding table. The coding table after Huffman may be the only table of the FSM or one of several entropy coding tables to be in the FSM. In one embodiment the Huffman bits have a 50% chance and they will coded with a fixed probability of 50%. This can be one Compression advantage over Huffman similar Codes available put. In one embodiment the bits are increased or more likely instead of 50%.

you notice, however, that in this case this compression function would not get better than at a time trial.

wobei jeder Ast des Huffman-Baums näherungsweise 98% sein sollte, jedoch die Wahrscheinlichkeitsklasse nur eine Erhöhungseffizient von 90 % ermöglichen kann. Ein Huffman-Baum mit diesen Wahrscheinlichkeiten wird in 8 gezeigt. Der Huffman-Kode kann verwendet werden, um die Anzahl von binären Entscheidungen zu verringern oder sogar zu minimieren und um die stark erhöhte Wahrscheinlichkeiten zu verringern. Zum Beispiel hat das erste kodierte Bit (nur) 90% anstatt daß jedes Bit bei 93 % Erhöhung bzw. Versatz kodiert wird.. Ein FSM-Kodierer, der Huffman-Ausgänge kodiert, wird effizienter sein als ein Huffman oder ein FSM-Kodierer, der eine Tabelle mit einem Bit zu einer Zeit einsetzt. Indem einfach der FSM-Kodierer mit einem Bit zu einer Zeit mit den wahrlichen Wahrscheinlichkeiten nach 8 verwendet wird und das Ergebnis aufgezeichnet wird, kann eine Tabelle mit einem Bit zu einer Zeit aufgebaut werden.Assume that a run or a series of strongly raised or probable bits should be coded. For example, at 98%, it is likely that bits are 0 and the encoder is to be configured to encode a run of length 0-5, as shown in Table 11. Table 11

where each branch of the Huffman tree should be approximately 98%, but the probability class can only provide an increase efficiency of 90%. A Huffman tree with these probabilities will be in 8th shown. The Huffman code can be used to reduce or even minimize the number of binary decisions and to reduce the greatly increased probabilities. For example, the first coded bit has (only) 90% instead of each bit being coded at 93% increment. An FSM encoder that encodes Huffman outputs will be more efficient than a Huffman or a FSM encoder, which inserts a table with one bit at a time. By simply hitting the FSM encoder one bit at a time with the true chances 8th is used and the result is recorded, a table can be built with one bit at a time.

If the FSM is using two states, where p-class 0 is used for probabilities less than 0.618, and the huffman tree is after 8th is used, table 12 can be constructed. Table 10 works better than the Huffman code because 80% of the time is in one row or two row Be runs of a length of 5 and this will save a bit.

Table 12

The 4 FIG. 12 illustrates one embodiment of a multi-symbol entropy encoding table that may be used to perform coding, such as by Huffman, or run-code encoding. Referring to 4 becomes an entropy coding table 405 coupled to receive a 14-bit input, the six bits of state information from the state register 406 and eight bits from a symbol or expiration count. The eight bits may indicate any ASCII character or run length from 0 to 255. In succession is the entropy coding table 405 25 bits off, the six bits state information for the status register 406 that represent the next state. The entropy coding table 405 Also outputs 15 output bits (OUTBITS) and four bits indicating the size of the output (SIZE).

An entropy coding table with several tables

The The present invention provides different subsets of a larger acceleration table to disposal, the useful are. This means, in one embodiment For example, the n-bit entropy encoding table becomes only one at a time a number of lookup tables generated by the entropy coding FSM is used. The FSM can use several tables that are constructed are to make a different number of bits (two or more) for individually accommodate fixed probabilities. A system can have one Table with one at a time and a table included to different Encode bits with the same probability. To this Way, multiple bits are encoded with the same probability such as when the probability is 50% or very steeply sloped or increased is (e.g., 99%). A special table for any fixed probability or even a fixed pattern of probabilities could be easy generated (e.g., a 50% bit, then a 99% bit, then two 60% bits).

In an embodiment For example, the present invention uses a one bit LUT to one Time associated with one or more n bits with n bits to one Fixed-time time to encode wavelet coefficients. The last two significant bits of the wavelet coefficients are typically 50% chance or 50% chance 1 or 0 (i.e., the probabilities are close to 0.5). The upper bits in the coefficient contained in the article Edward L. Schwartz, Ahmad Zandi, Martin Boliek, "Implementation of Compression with Reversible Embedded Wavelets ", Proc. Of SPIE 40th Annual Meeting, Volume 2564, San Diego, CA, July 1995, are heavily sloped, approximate at 99% zeros. A separate LUT brings several bits of head to one Time under.

When multiple tables are used to implement the FSM encoder, the present invention maintains the same channel states as one input to all of the tables. In one embodiment, a single channel state register outputs the same state information as an input to all of the tables. In one embodiment, the next state output from the table is used as feedback to the channel state registers for all tables. Note that this requires all tables to have the same state. The control logic can be used to control which table is accessed based on the inputs by switching between the tables. The control logic may have a response via the fit for the inputs. In another embodiment, all of the tables are accessed at all times, but the output is multiplexed by all of the tables. In this case, the control logic selects only one set of inputs as the outputs of the FSM encoder. By using the same channel state for each of the tables, no fractional bits are lost when switching between the encoders.

at an embodiment, in which the head bit or the leading one Bit and the last significant bits of the coefficients (tail bits) with an n-bit table (n> 2) can be a table with one bit at a time for those bits between the leading and the tail bits.

you note that such Tables are not overly large, because the number of bits that are required to get the probability represent (e.g., 4 bits) does not change for each of the input bits. The Probability only needs to be sent once, instead of with every bit that needs to be encoded.

Similar to the one discussed above, the multi-symbol table or multiple symbol table may after 4 one of the several tables used by the FSM encoder. The 5 represents an encoder capable of performing the same operation as in 1A (a FSM table with one bit at a time), 3 (a 4-bit FSM table at a time operating at a 50% fixed probability), and 4 (a Huffman code FSM table) by using two additional bits as a table selector. The remainder of the bits would remain the same as used in the previous table. The input and output quantities are determined based on the maximum size required by the subtables (plus the selector bits). Referring to 5 receives the entropy coding table 505 18 bits of input, with 6 bits over the state information, 2 bits of the table selector, and 10 table specific bits. For run coding, these may be one run length, four of which could be four 50% bits and so on. The entropy coding table 505 outputs 27 bits, the 6 state information bits, which feed back to the state register 506 are 16 output bits and 5 bits that indicate the size of the output.

Combine of tables

In one embodiment, the number of tables in a system, such as the tables in FIG 1A , are reduced by combining tables. Since the table for updating the estimation state has an output which is used as an input to the entropy-encoding state table, the inputs and outputs of these two tables can be combined. The integration of the FSM and the Probability Estimation Engine (PEM) provides an improvement in speed when one bit is encoded at a time.

An embodiment having this combined table is described in FIG 6 shown. Referring to 6 becomes a context memory 101 coupled or connected to a context 110 from a context model (not shown to avoid obscuring the present invention) and a next_np state indication 612 to recieve. As described above, the next_np state indicator represents 612 an ad for the context memory 101 with respect to which the next probability estimation state holds. Because of these inputs is on the context memory 101 accessed and gives the current probability state (np state) 613 out.

The connected probability estimation and entropy coding table 603 is connected to the NP status 613 , the current bit 616 and an output of the channel state 606 to recieve. The output of the channel state 606 is in response to a next status indication (next_state) from the associated probability estimation and entropy coding table 601 is issued. Based on their inputs, the linked probability estimation and entropy coding table generates 601 Output bits as output bits 121 are shown, and a display of the output, which as Ausausgröße 122 is shown. The connected probability estimation and entropy coding table 601 also generates the next_np status indicator 612 indicating the next probability estimation state.

An embodiment of the connected probability estimation and entropy coding table 601 is shown below in Table 13.

Table 13 Connected probability estimation and entropy coding table

The Decoder table is different because the encoder and decoder encoder tables are different. For example, the decoder includes the following Inputs. An entrance for compressed bits representing the compressed bitstream, a Entrance for the channel state, which represents the channel state, and an input for the p state representing the probability estimation state. The Outputs of the Decoders include the following: a shift amount that includes a Indicates amount to shift the input compressed bitstream, an MPS indicator, a new_p state, which is the next likely instruction state and a next_status indicator, which represents the new probability estimation state.

For a pseudorandom probability estimation table, the linked table encoder is of a realizable size. Using a single table can take some computing time at the expense of Save memory.

The 7 FIG. 12 illustrates a modified embodiment of an encoder in which the likelihood state includes the MPS. The use of this embodiment represents a computational enhancement with lower storage costs than that of a table method associated with 6 has been described. This embodiment requires an estimation table that is twice as large as the estimation table described above with respect to 1A however, eliminates the need for comparison and exclusive-or-operation (XOR).

Referring to 7 becomes a context memory 101 connected to a context 110 and a next_np status indicator 712 to recieve. The next_np state indicator 712 provides an ad for the context memory 101 with respect to which the next probability estimation state applies, and also includes the MPS. Based on these inputs is on the context memory 101 accessed and gives the current probability state (np state) 713 out.

The probability estimation table 703 is connected to the p-state 713 and the current bit 116 receive and gives based on these inputs a probability class (p-class) 718 out. One embodiment of the probability estimation table is shown in Table 14 below.

Table 14 - PEM with XOR and containing comparison

The entropy coding table 705 is connected to the p-class 718 and an output from the channel state 106 to recieve. The output of the channel state 106 is a next_state in response to a next status indication 119 that of the entropy coding table 705 is issued. Based on their inputs, the entropy coding table generates 705 Output bits as output bits 121 are shown, and an indication of the output, which as Ausgröße or output 122 will be shown.

In the form of software, it is reasonable to combine a fixed field with a bit with a field of nine bits, so that the context memory 101 4 bytes per context (2 bytes estimate state + 1 byte MPS indication + one byte padding) can be reduced to 2 bytes per context.

system applications

The entropy coder described herein may be used in a variety of compression systems. For example, the entropy coder could be used in the JPEG system. In an off For example, the SSM encoder and the probability estimator can replace the QM encoder of the standard JPEG system. This replacement would probably only benefit from the combined estimation and entropy coding tables. An alternative embodiment could replace the Huffman code from the JPEG standard with the M-fold version of the FSM described herein. In fact, the JPEG standard uses "mantissa bits" and "make-up bits" in the codes. The "mantissa bits" could be coded by the M-times FSM and the "make-bits" coded by the fixed N at a 50% time-related bitmap.

Similarly, the Entropy coder used with the CREW wavelet compression system as described in the article by Edward L. Schwartz, Ahmad Zandi, Martin Boliek, "Implementation of Compression with Reversible Embedded Wavelets ", Proc. of SPIE 40th Annual Meeting, Vol. 2564, San Diego, CA, July 1995. In the CREW system there are leading ones Bits that are very likely to be 0 are. These leading Bits or head bits could be a code having a fixed probability of e.g. 95% through uses an FSM encoder with n bits at a time. Many are similar the "tail bits" equally likely equal to 0 or 1 and could using an FSM coder with n bits at a time for a 50 % probability be coded. There are different ways about which the head or tail bits could be coded. For example, four could Tail bits are coded together by a single coefficient. Alternatively, could the last four significant bits of four different coefficients coded together. There are many other options available to the professionals can be seen in the prior art. It is clear that others Wavelet image compression systems and even text or sound compression systems the different coding possibilities of the FSM encoder the can use present invention.

While many amendments and modifications of the present invention to a person skilled in the art The technology will no doubt be brought to mind, he said the preceding Description, it is to be understood that the particular embodiment, which has been shown and described by way of illustration not intended to be considered restrictive become. Therefore, it is not intended that references to details the present embodiment the scope of the claims restrict, which in themselves reflect only those features that are essential to the invention considered essential.

The The present invention relates to a device for coding and / or Decoding for compression or expansion or decompression used by data. A machine with finite state or finite state has a number of tables together have several states. One of the tables encodes multi-bit symbols for fixed probabilities.

The following embodiments of the invention are also part of the description:
An encoder for encoding data inputs, the encoder comprising:
a channel state storage device; and
an entropy coding table coupled to receive state information from the channel state storage device and configured to encode n bits of the input data at a time in response to the state information, where n is greater than or equal to 2.
B coder according to feature A, wherein the entropy coding table encodes n bits at a fixed probability.
C Encoder according to one of the features A or B, wherein the n bits have a symbol.
D coder according to one of the features A to C, wherein the entropy coding table comprises a Huffman coder.
An encoder according to any of features A to D, wherein the entropy coding table comprises a run-length encoder.
F coder according to any of features A to E, wherein the entropy coding table encodes last significant bits of coefficients.
A coder according to any one of features A to F, wherein the entropy coding table encodes leading bits of coefficients in the data input.
H system for encoding input data, in particular with the encoder according to one of the features A to G, the system comprising:
a context model configured to generate a series of binary decisions in response to contexts;
a probability estimation model coupled to the context model to generate probability estimates for the series of binary decisions;
a channel state storage device; and
a plurality of entropy coding tables coupled to state information of the channel state memory and configured to encode bits from the input data in response to the state information, wherein one of the plurality of entropy encoding tables is configured to encode n bits at a time, wherein n is greater than or equal to 2 is.
I System according to feature H, wherein the one of the plurality of entropy coding tables encodes n bits at a fixed probability.
A system according to any of features H to I, wherein the one of the plurality of entropy coding tables comprises a Huffman encoder.
K system according to one of the features H to J, wherein the one of the plurality of Entropiekoderiertabellen has a run-length encoder.
L system according to one of the features H to K, wherein the one of the plurality of Entropiekoderiertabellen encodes the last significant bits of the coefficient tails and ends.
M system according to one of the features H to L, wherein the one of the plurality of Entropiekodiertabellen coded front or bits of the coefficients in the data input.
N system according to one of the features H to M, wherein at least one of the plurality of Entropiekodiertabellen has a combined probability estimation and entropy coding.
The system of any one of features H to N, further comprising a switch configured to select individual tables of the plurality of entropy coding tables to encode one or more bits of the input data based on the likelihood.
P system according to one of the features H to O, wherein the probability estimation model comprises a table.
Q System of feature P, wherein the context model provides a probability state indicator having a Most Probable Icon (MPS) indicator therein.
R encoder for encoding data inputs, in particular for a system according to one of the features H to Q, wherein the encoder comprises:
a channel state storage device; and
a linked probability estimation and entropy coding table coupled to receive state information from the channel state memory device and configured to encode n bits of the input data at a time in response to the state information, where n is greater than or equal to 2.
S Encoder according to feature R, wherein the Entropiekodiertabeller encodes n bits at a fixed probability.
T encoder according to feature R or S, wherein the n bits have a symbol.
U coder according to one of the features R to T, wherein the entropy coding table comprises a Huffman coder.
V coder according to one of the features R to U, wherein the entropy coding table comprises a run length coder.
W coder according to one of the features R to V or feature H, wherein the entropy coding table encodes the least significant bits of coefficients.

Claims (15)

Decoder, comprising - a state memory device ( 106 ) to store a channel state; A probability estimation table ( 103 ) to provide an indication of a probability estimate; An entropy coding table ( 115 ) to shift bits to be shifted from input compressed bits based on the state information supplied by the state memory device ( 106 ) and the likelihood estimation display provided by the likelihood estimation table ( 103 ) to decode the decoded compressed bits; - a shift register ( 131 ) to shift the received compressed bits based on information received from the entropy coding table. The decoder of claim 1, wherein the entropy coding table generates an output, which is coupled to control the shift register. Decoder according to one of claims 1 or 2, wherein the entropy coding table the next State, which is returned to the state storage device, to decode the next one or more bits of compressed data. The decoder of claim 1, further comprising an XOR gate, wherein the entropy coding table generates a Last Likelihood Symbol (LPS) indication that performs an XOR operation is subjected to the Most Probability Symbol (MPS) XOR gate for the current probability estimate state to generate one or more bits. The decoder of claim 4, further comprising a context memory which is coupled to a probability estimation state for the Probability estimation table to provide in response to a context where the Context memory also the MPS for displays the context. Decoder according to one of claims 1 to 5, further comprising a multiplexer coupled to one of the next LPS and the next Receive MPS from the probability estimation table, wherein the MUX provides an indication of the next MPS in response to the LPS output from the entropy coding table outputs, being between the next LPS and the next MPS input to the MUX selected becomes. Decoder according to one of claims 1 to 6, wherein the state memory means is coupled to the state information to each of the several Entropy coding tables enter. System with coder and decoder, wherein the decoder is constructed according to one of claims 1 to 7, wherein the compressed bits ( 130 ) are output from an entropy coding table in the encoder that receives state information from a channel state memory device and that is configured to code n bits of the input data simultaneously in response to the state information, where n is greater than or equal to two. Method for decoding, comprising the following steps: - a channel state is stored in a state memory device ( 106 ) saved; A probability estimation display is provided by a probability estimation table ( 103 ) made available; Bits to be shifted from input compressed bits are determined based on the state information supplied by the state memory device (FIG. 106 ) and the likelihood estimation display provided by the likelihood estimation table ( 103 ) have been received by means of an entropy coding table ( 115 ) to decode the compressed bits; The bits of the compressed bits are stored in a shift register on the basis of information received from the entropy coding table ( 131 ) postponed. The method of claim 9, wherein from the entropy coding table generates an output which is coupled to control the shift register. Method according to one of claims 9 or 10, wherein of the Entropy coding table the next state is generated and returned to the state storage device, to decode the next one or more bits of the compressed data. Method according to one of claims 9 to 11, wherein the entropy coding table generates a Last Likelihood Symbol (LPS) display that an XOR operation through an XOR gate with the most probable symbol (MPS) for the current one Subjected to probability estimation state is to generate one or more bits. The method of claim 12, further comprising a context memory providing a probability estimation state for the Probability estimation table in response to a context, wherein the context memory also the MPS for displays the context. Method according to one of claims 9 to 13, wherein a multiplexer (MUX) is coupled to one of the next LPS and the next MPS from the probability estimation table receive, wherein the MUX is an indication of the next MPS in response to the LPS output from the entropy coding table, with between the. next LPS and the next MPS input to the MUX selected becomes. Method according to one of claims 9 to 14, wherein the state memory device is coupled to the state information to each of the several Entropy coding tables enter.