Classifications

• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
  • H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
  • H03M7/40—Conversion to or from variable length codes, e.g. Shannon-Fano code, Huffman code, Morse code
  • H03M7/42—Conversion to or from variable length codes, e.g. Shannon-Fano code, Huffman code, Morse code using table look-up for the coding or decoding process, e.g. using read-only memory

Definitions

• the invention generally relates to data compression, and more specifically relates to a form of entropy coding.
• input data is encoded by an encoder, transmitted over a communication channel (or simply stored), and decoded by a decoder.
• an input signal is typically pre-processed, sampled, converted, compressed or otherwise manipulated into a form for transmission or storage. After transmission or storage, the decoder attempts to reconstruct the original input.
• One fundamental limitation of this simple model is that a given communication channel has a certain capacity or bandwidth. Consequently, it is frequently necessary to reduce the information content of input data in order to allow it to be reliably transmitted, if at all, over the communication channel.
• an optimal encoding is to use equal length code words, where each bit of an ⁇ -bit code allows distinguishing among 2 n equally probable input possibilities.
• a single bit i.e., single entry code book
• two bits can distinguish four possibilities, etc.
• Entropy encoding operates by assigning variable-length codes (e.g., code book entries) to fixed-sized blocks of input. That is, a random variable X, which is known to take on values x ⁇ .. ⁇ m with corresponding probability p ⁇ ..pTM, is mapped to an entry within a set of code words ⁇ Y ⁇ .
• y ⁇ k will be referenced simply as V ⁇ , with k implied.
• the code alphabet is likely to be a series of binary digits ⁇ 0, 1 ⁇ , with code lengths measured in bits. It is assumed code words are constructed so only a single scan of a compressed representation needs to be inspected in order to reconstruct appropriate output.
• the difficulty in entropy encoding the source signal depends on the number m of possible values X may take. For small m, there are few possible messages, and therefore the code book for the messages can be very small (e.g ., only a few bits need to be used to unambiguously represent all possible messages) .
• One possible set of code book entries is assigning "1 " to represent message xi , "01 " for message X2, "000” for message X3, and "001 " for message X4. This gives an average code length of 1 .56 bits instead of 2 for encoding the random variable X -- a significant savings.
• Coding sequences of X's produces further savings as it is known from information theory studies that the entropy of a coherent series X ⁇ ..X ⁇ is less than or equal to the sum of each individual X's entropy.
• vector Huffman coding can compress a coherent source much more efficiently than scalar Huffman coding .
• the efficiency of vector Huffman coding is only limited by practical concerns. In order to achieve higher compression ratios, bigger vector dimensions are needed. Higher dimension, however, increases code book sizes beyond practical limits. For example, for source symbols having 30 possible values, a dimension of only 6 corresponds to a code book of 729 million entries.
• entropy coding are characterized as fixed-to-variable length coding as the source symbols have fixed length and the code words have variable length depending on the probability of the corresponding source symbol.
• Another methods of entropy coding have been attempted which attempt the opposite approach, where a variable number of source symbols are grouped together and then translated into code words having equal length.
• the source is composed of independent X's, and symbol groupings achieve equal probability, such a reverse scheme is provably optimal .
• such solutions require resources exceeding resources practically (if at all) available.
• variable-to-fixed length approach is not useful.
• the invention relates to a method of assigning variable length codes to variable length input sequences.
• entropy-type codes are assigned to probable input sequences, thus allowing a particular input stream to be encoded in a compressed format.
• the invention may be configured so as to reduce the size of the code book required for performing encoding and decoding.
• variable length code words might only be assigned to inputs that are highly probable, and where default codes can be assigned to less probable sequences.
• the degree of probability required for assignment of a specific code to a specific input is adjusted according to a desired code book size.
• the input stream to encode can be of any data type, such as numbers, characters, or a binary data stream which encodes audio, video or other types of data.
• the input stream is referenced herein as a series of symbols, where each "symbol” refers to the appropriate measurement unit for the particular input.
• a code book is constructed for groupings of symbols, in which variable-sized groups of symbols are each assigned a variable length code based on probability of occurrence of symbol groupings.
• possible groupings of symbols are generated and compared against the probability of the generated grouping occurring in exemplary input used to generate the code book.
• Such exemplary input is assumed to approximate arbitrary input likely to be received and require encoding.
• a data structure may be used to track symbols combinations (e.g., the groupings) . This structure is used to associate the new symbol with previously received symbols, so that arbitrarily long groupings of previously received symbols are tracked.
• One possible configuration for the data structure is a tree-type data structure, in which successive symbol groupings form new leaf nodes. These nodes may contain an entire grouping or just the single symbol extension to a previous parent node. In this latter configuration, the path from the root of the tree corresponds to a particular grouping.
• one or more trivial groupings are selected, such as single symbol "groups" containing symbols from the input alphabet.
• the probability of these initial groupings is evaluated to determine the grouping most likely to occur as input, where such probability is necessarily computed with respect to exemplary inputs.
• the most probable grouping is then expanded with symbols from the alphabet to form tentative groupings.
• the probability of these tentative groupings is then evaluated to identify the most probable tentative expansions, and the least probable groupings combined into a single grouping.
• the concept of a code book is to assign code words to symbol groupings.
• the invention can be configured so that code book size is restricted. One method of doing so is avoiding assigning codes to all input sequences. Instead, only probable input sequences are stored in the code book and assigned an entropy-type code. Improbable sequences are represented in the code book as an input sequence prefix followed by a special expansion character suffix. This suffix character represents all possible input sequence extensions to the prefix. The prefix-suffix pairing represents all possible input sequences beginning with the prefix that do not have an entry in the code book. Thus, after evaluating the tentative extensions, two code book entries result, one for the most probable extension, and one to represent all other extensions (again, assuming only keeping one most probable extension) .
• FIG. 1 is a block diagram of a computer system that may be used to implement a variable to variable entropy encoding.
• FIG . 2 shows a basic communication model
• FIG. 3 is a flowchart showing creation of a code book having variable length entries for variable length symbol groupings.
• FIGS. 4-10 illustrate creation of a code book pursuant to FIG . 3 for an alphabet ⁇ A, B, C ⁇ .
• the invention has been implemented in an audio/visual codec. This is only one example of how the invention may be implemented.
• the invention is designed to be utilized wherever entropy-type coding may be utilized, and is applicable to compression of any type of data. Briefly described, optimal entropy encoding requires excessive resources, and the illustrated embodiments provide a nearly optimal encoding solution requiring far fewer resources.
• FIG. 1 and the following discussion are intended to provide a brief, general description of a suitable computing environment in which the invention may be implemented. While the invention will be described in the general context of computer- executable instructions of a computer program that runs on a personal computer, those skilled in the art will recognize that the invention also may be implemented in combination with other program modules. Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Moreover, those skilled in the art will appreciate that the invention may be practiced with other computer system configurations, including hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframe computers, and the like.
• the illustrated embodiment of the invention also is practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. But, some embodiments of the invention can be practiced on stand alone computers. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.
• an exemplary system for implementing the invention includes a computer 20, including a processing unit 21 , a system memory 22, and a system bus 23 that couples various system components including the system memory to the processing unit 21 .
• the processing unit may be any of various commercially available processors, including Intel x86, Pentium and compatible microprocessors from Intel and others, the Alpha processor by Digital, and the PowerPC from IBM and Motorola. Dual microprocessors and other multi-processor architectures also can be used as the processing unit 21 .
• the system bus may be any of several types of bus structure including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of conventional bus architectures such as PCI, AGP, VESA, MicroChannel, ISA and EISA, to name a few.
• the system memory includes read only memory (ROM) 24 and random access memory (RAM) 25.
• ROM read only memory
• RAM random access memory
• BIOS basic input/output system
• BIOS basic routines that help to transfer information between elements within the computer 20, such as during start-up, is stored in ROM 24.
• the computer 20 further includes a hard disk drive 27, a magnetic disk drive 28, e.g., to read from or write to a removable disk 29, and an optical disk drive 30, e.g., for reading a CD-ROM disk 31 or to read from or write to other optical media.
• the hard disk drive 27, magnetic disk drive 28, and optical disk drive 30 are connected to the system bus 23 by a hard disk drive interface 32, a magnetic disk drive interface 33, and an optical drive interface 34, respectively.
• the drives and their associated computer- readable media provide nonvolatile storage of data, data structures, computer-executable instructions, etc. for the computer 20.
• a number of program modules may be stored in the drives and RAM 25, including an operating system 35, one or more application programs (e.g., Internet browser software) 36, other program modules 37, and program data 38.
• application programs e.g., Internet browser software
• a user may enter commands and information into the computer 20 through a keyboard 40 and pointing device, such as a mouse 42.
• Other input devices may include a microphone, joystick, game pad, satellite dish, scanner, or the like.
• These and other input devices are often connected to the processing unit 21 through a serial port interface 46 that is coupled to the system bus, but may be connected by other interfaces, such as a parallel port, game port or a universal serial bus (USB) .
• a monitor 47 or other type of display device is also connected to the system bus 23 via an interface, such as a video adapter 48.
• personal computers typically include other peripheral output devices (not shown), such as speakers and printers.
• the computer 20 is expected to operate in a networked environment using logical connections to one or more remote computers, such as a remote computer 49.
• the remote computer 49 may be a web server, a router, a peer device or other common network node, and typically includes many or all of the elements described relative to the computer 20, although only a memory storage device 50 has been illustrated in FIG. 1 .
• the computer 20 can contact the remote computer 49 over an Internet connection established through a Gateway 55 (e.g., a router, dedicated-line, or other network link), a modem 54 link, or by an intra-office local area network (LAN) 51 or wide area network (WAN) 52.
• LAN local area network
• WAN wide area network
• the present invention is described below with reference to acts and symbolic representations of operations that are performed by the computer 20, unless indicated otherwise. Such acts and operations are sometimes referred to as being computer-executed. It will be appreciated that the acts and symbolically represented operations include the manipulation by the processing unit 21 of electrical signals representing data bits which causes a resulting transformation or reduction of the electrical signal representation, and the maintenance of data bits at memory locations in the memory system (including the system memory 22, hard drive 27, floppy disks 29, and CD-ROM 31 ) to thereby reconfigure or otherwise alter the computer system's operation, as well as other processing of signals.
• the memory locations where data bits are maintained are physical locations that have particular electrical, magnetic, or optical properties corresponding to the data bits.
• a basic communication model there is a data source or sender 200, a communication channel 204, and a data receiver 208.
• the source may be someone speaking on a telephone, over telephone wires, to another person.
• the source may be a television or radio broadcast transmitted by wireless methods to television or radio receivers.
• the source may be a digital encoding of some data, whether audio, visual, or other, transmitted over a wired or wireless communication link (e.g., a LAN or the internet) to a corresponding decoder for the information.
• an encoder 202 is used to prepare the data source for transmission over the communication channel 204.
• the encoder is responsible for converting the source data into a format appropriate for the channel 204.
• a decoder 206 is required to reverse the encoding process so as to present sensible data to the receiver.
• the encoder 202 is frequently designed to utilize compression schemes for the transmission of the data. Compression is desirable since, except for unusual circumstances, communication bandwidth is limited. Therefore, for complex data sources, such as audio or video data, the source data needs to be compressed to allow its transmission over conventional transmission paths.
• a particularly effective coding method is known as entropy encoding, which utilizes a "code book" containing short code words that have been pre-assigned to highly probable input data.
• An effective coding method is entropy encoding. Such coders can capitalize on the data coherency, and are particularly effective when symbols have non- uniform probability distribution.
• FIG. 3 is a flowchart showing a preferred method for generating a code book.
• FIG. 3 illustrates how to create a code book having variable length code assignments for variable length symbol groupings.
• prior art techniques either require fixed-length codes or fixed blocks of input.
• Preferred implementations overcome the resource requirements of large dimension vector encoding, and the inapplicability of coding into words of equal lengths, by providing an entropy based variable-to-variable code, where variable length code words are used to encode variable length X sequences.
• yi represent each source symbol group ⁇ XJ ⁇ , for 1 ⁇ / ⁇ N ⁇ , having probability Pi of occurring within the input stream (FIG . 2 channel 204), and that each group is assigned a corresponding code word having L bits. It is presumed that each x f is drawn from a fixed alphabet of predetermined size. The objective is to minimize the
• equation L instead of finding a general solution to the problem, the problem is separated into two different tasks.
• the first task is identification of a (sub-optimal) grouping of a set of input symbols ⁇ xi ⁇ through an empirical approach described below.
• the second task is assigning a entropy-type code for the grouped symbols ⁇ yi ⁇ . Note that it is known that if the source is not coherent (i.e., the input is independent or without memory), any grouping that has the same configuration of ⁇ Nj ⁇ can achieve the same coding efficiency. In this situation, the first task becomes inconsequential.
• This initial configuration assumes that an exemplary input stream is being used to train creation of the code book.
• a computer may be programmed with software constructions such as data structures to track receipt of each symbol from an input.
• data structures may be implemented as a binary-type tree structure, hash table, or some combination of the two. Other equivalent structures may also be used.
• the probability of occurrence for each y. is computed 302. Such probability is determined with respect to any exemplary input used to train code book generation. As further symbols are added to the symbol data structure, the probabilities are dynamically adjusted.
• the most probable grouping y. is identified 304 (denoted as y m ) . If 306 the highest probability symbol is a grouping of previously lower probability symbols, then the grouping is split 308 into its constituent symbols, and processing restarted from step 302. (Although symbols may be combined, the group retains memory of all symbols therein so that symbols can be extracted.)
• step 310 in which the most probable grouping is then tentatively extended 310 with single symbol extensions x.'s.
• y mp is extended with each symbol from the X alphabet is used.
• a predictor can be used to only generate an extension set containing only probable extensions, if the alphabet is very large and it is known many extensions are unlikely. For example, such a predictor may be based on semantic or contextual meaning, so that very improbable extensions can be ignored a priori.
• the probability for each tentative expansion of y mp is then computed 31 2, and only the most probable extension retained 314.
• the rest of the lower probability extensions are collapsed together 31 6 as a combined grouping and stored in code book with a special symbol to indicate a combined grouping .
• This wild-card symbol represents any arbitrary symbol grouping having y m as a prefix, but with an extension (suffix) different from the most probable extension. That is, if ymp + X p is the most probable root and extension, then the other less probable extensions are represented as ym P * , * ⁇ x mp . (Note that this discussion presumes, for clarity, serial processing of single-symbol extensions; however, parallel execution of multiple symbol extensions is contemplated.)
• Code book construction is completed by repeating 31 8 steps 302-31 6 until all extensions have been made, or the number of code book entries reaches a predetermined limit.
• the effect of repeatedly applying the above operations is to automatically collect symbol groupings having high correlation, so that inter-group
• One structure for a code book is traversal and storage of a N-ary (e.g., binary, tertiary, etc.) tree, where symbol groupings guide a traversal of the tree structure.
• Leaf nodes of the tree represent the end of a recognized symbol sequence, where an entropy- type code is associated with the sequence.
• Nodes can be coded in software as a structure, class definition, or other structure allowing storage of a symbol or symbols associated with the node.
• the code book may be structured as a table having each string of input symbol sorted by probability of occurrence, with highly probable input at the top of the table.
• the table can be sorted according to the first symbol, i.e., all symbol series beginning with "A” are grouped together, followed by series starting with "B", etc. With this arrangement, all entries within the grouping are sorted according to their probabilities of occurrence. The position of the beginning of each section is marked/tracked so that a hash-type function (e.g., a look-up based on the first symbol) can be used to locate the correct portion of the code book table, in this look-up table approach to storing the code book, once the first symbol is hashed, look-up simply requires a search of the corresponding table section until a matching entry is located.
• FIGS. 4-1 0 illustrate creation of a code book pursuant to FIG.
• the code book is defined with respect to an exemplary input stream "A A A B B A A C A B A B B A B".
• one or more exemplary inputs may be used to generate a code book that is then used by encoders and decoders to process arbitrary inputs.
• the code book is presented as a tree structure, although it may in fact be implemented as a linear table, hash table, database, etc.
• the tree is oriented left-to-right, where the left column (e.g., "A” and "X0") represents the top row of a tree-type structure, and successively indented rows represent the "children" of the previous row's node (e.g., in a top-down tree for FIG. 5, node "A” is a first-row parent node for a second-row middle-child node “B”.) .
• FIG. 4 shows an initial grouping for the input stream "A A A B B A A - C A B A B B A B".
• This initial trivial grouping can be created based on different criteria, the simplest being having a first-level node for every character in the alphabet.
• FIG. 4 illustrates this technique by starting with only two initial groups, group A 400 having probability 8/1 5, and group XO 402 having probability 7/1 5, where X0 represents all remaining low probability symbols in the alphabet, e.g., B and C.
• the leaf-node having highest probability is selected for extension (see also FIG. 3 discussion regarding processing sequence) .
• group A 400 is tentatively expanded by each character in the alphabet (or one may limit the expansion to some subset thereof as described for creating the initial grouping) .
• Probabilities are then recomputed with respect to the input stream "A A A B B A A C A B A B B A B” to determine values for the tentative extensions A 406, B 408, and C 410.
• the result is nine parsing groups, where "A A” appears 2/9, "A B” appears 4/9, and "A C" appears 0/9.
• FIG. 6 shows the collapse into X1 41 2 for FIG. 5. Processing repeats with identification of the node having highest probability, e.g., node B 408 at probability 4/9
• this node 408 is tentatively extended with symbols A 414, B 41 6, C 41 8, and as discussed above, the tentative grouping with highest probability is retained.
• the result is eight parsing groups in which the symbol sequence "A B A” 414 appears once, "A B B” 41 6 appears once, and "A B C” 41 8 does not appear at all. Since tentative extensions A 41 4 and B 41 6 have the same probability of occurrence, a rule needs to be defined to choose which symbol to retain.
• the highest row node e.g., the left-most child node in a top-down tree
• the left-most row's node e.g., the node closest to the root of a top-down tree
• code book entries can be created to account for such end of input sequences, or the input having no entry can be escaped with a special character and inserted in the encoded output stream For example, a special symbol can be used to indicate end of input, therefore implying how to handle the trailing characters on decoding
• the next step is to expand the node currently having highest probability with respect to the input stream.
• the highest node in the tree (X0 402) is extended (Although it is only necessary to be consistent, it is preferable to expand higher level nodes since this may increase coding efficiency by increasing the number of long code words.)
• X0 402 is a combined node, so it must be split instead of extended.
• FIG . 9 illustrates the result of splitting node XO into its constituent symbols B 422 and C 424.
• FIG 1 0 shows the result of retaining high-probability node B 422. Note that grouping XO now only represents a single symbol "C”. After revising probabilities, the node having highest probability must be identified and split or extended.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)

Priority Applications (4)

Applications Claiming Priority (2)

Publications (1)

Family

ID=22787319

Family Applications (1)

Country Status (6)

Cited By (1)

Families Citing this family (34)

Citations (5)

Family Cites Families (39)

• 1998
  • 1998-12-14 US US09/211,532 patent/US6404931B1/en not_active Expired - Lifetime
• 1999
  • 1999-12-07 WO PCT/US1999/029003 patent/WO2000036751A1/en not_active Ceased
  • 1999-12-07 JP JP2000588899A patent/JP4559631B2/ja not_active Expired - Lifetime
  • 1999-12-07 DE DE69916661T patent/DE69916661T2/de not_active Expired - Lifetime
  • 1999-12-07 EP EP99968085A patent/EP1147612B1/en not_active Expired - Lifetime
  • 1999-12-07 AT AT99968085T patent/ATE265106T1/de not_active IP Right Cessation

Patent Citations (6)

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events