JPH1084484A - Data compression system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To properly compress an image, including two kinds of image which are an image of a naturally continuous gradation and an image of binary/noise free/shallow pixel density by the same system. SOLUTION: A mode selection mechanism 110 discriminates whether the input image data 110 are a continuous gradation image or a binary image or which part of the image has such a characteristics so as to select the mode, and the input image data 110 gives the part to wavelet system coders (a reversible wavelet transformation block 102, an imbeded sequencing quantization block 103, and a horizontal context model block 105) or a binary system coder (a gray coding block 104). Each coder has the same entropy coder 106 in common.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

FIELD OF THE INVENTION The present invention relates to the field of data compression and decompression systems, and more particularly to lossless and lossy data in compression / decompression systems.
y) encoding and decoding method and apparatus.

[0002]

BACKGROUND OF THE INVENTION Data compression is a very useful tool for storing and transmitting large amounts of data. For example, the time required for image transmission, such as facsimile transmission of documents,
If compression is used to reduce the number of bits required for image reproduction, the time will be dramatically reduced.

[0003] There have been many different data compression techniques. Compression techniques can be broadly classified into two categories: lossy coding and lossless coding. Lossy coding is a coding that results in a loss of information and thus does not guarantee complete reproduction of the original data. The goal of lossy encoding is to keep the changes from being unpleasant or noticeable, even if they have changed from the original data. In lossless compression, all information is preserved, and the data is compressed in a way that allows full decompression.

[0004] In lossless compression, input symbols or luminance data are converted to output codewords. As input,
Image data, audio data, one-dimensional data (for example, data that changes spatially or temporally), and two-dimensional data (for example, changes in two spatial axes (or changes in one spatial dimension and one time dimension)) Data) or multi-dimensional / multi-spectral data. If compression is successful, the codeword is represented with fewer bits than were required for the input symbol (or luminance data) before encoding. The lossless coding method includes a dictionary coding method (for example, Lempel-Ziv method), a run-length coding method, a counting coding method, and an entropy coding method. In lossless image compression, compression is based on prediction or context and coding. The JBIG standard for facsimile compression and the DPCM (differential pulse code modulation-JPEG standard option) for continuous tone images are examples of lossless compression for images. In lossy compression, input symbols or luminance data are quantized and then converted to output codewords. Quantization aims at preserving important features of data while removing unimportant features. Lossy compression systems often utilize transforms to concentrate energy prior to quantization. J
PEG is an example of a lossy encoding method for image data.

[0005] Recent developments in image signal processing have focused on pursuing efficient and accurate data compression encoding schemes. Various schemes of transformation or pyramid signal processing have been proposed, including a multi-resolution pyramid processing scheme and a wavelet pyramid processing scheme. These two systems are also called a subband processing system and a hierarchical processing system. The wavelet pyramid processing method for image data is a special multi-resolution pyramid processing method for performing subband decomposition of an original image using a quadrature mirror filter (QMF). There are other non-QMF wavelet schemes. For more information on the wavelet processing method, see Antonini,
M., et al., "Imaging Coding Using Wavelet Tra.
nsform ", IEEE Transactions on Image Processin
g, Vol. 1, No. 2, April 1992, and Shapiro,
J., "An Embedded Hierarchical Image CoderU
sing Zerotrees of Wavelet Coefficients ”, Pro
c. IEEE Data Compression Conference, pgs. 214-22
See 3, 1993. To obtain information on the reversible conversion, Said, A. et al. and Pearlman, W.C. "Revers
ibleImage Compression via Multiresolution Repr
esentation and Predictive Coding ”, Dept. of El
ectrical, Computer and System Enginering, Ren
ssealaer Polytechnic Institute, Troy, NY 1993
Please refer to.

[0006] Compression is often very time consuming and requires a large amount of memory. It is desirable to perform the compression faster and / or with as little memory as possible. In some applications, compression was not used because quality could not be guaranteed, the compression ratio was insufficient, or the data rate was not controllable. However, it is desirable to use compression to reduce the amount of information to be transmitted and / or stored.

The prior art includes a compression system for handling natural continuous tone images. One example is the international standard Dis. Ten
918-1, "Digital Compression and Coding of Cont
inuous-Tone Still Images ", CCITT Recommendation T.81, which is commonly referred to as JPEG.
There are also compression systems for handling binary / noise-free / shallow pixel depth images. One example of such a system is the international standard ISO / IEC 11544, "Information Technology.
-Codded Repersentation of Picture and Audition
formation-Progressive Bi-level Image Compressi
on ”, CCITT Recommendation T.82, which is usually
Called. However, there is no system in the prior art that handles both properly. It would be desirable to have such a system.

[0008] Parsers are well known in the computer science. The parser is responsible for signifying various parts of the object whose structure is not initially known. For example, some parsers that operate as part of a compiler have one string in the program file as an "identifier", another string constitutes a reserved word, and another string as a comment. Will decide. This parser does not judge what "meaning" the character string is, but only what kind of part of the object it is.

[0009] Most image storage formats are single use. That is, only a single resolution or a single quality level is available. Other image formats are versatile. Some prior art versatile image formats support two or three resolution / quality options, but some can only specify one of resolution or quality, but not both. It is desirable to increase the available resolution and quality options.

[0010] For example, the Internet World Wide Web server currently provides the required information from a large amount of data. Normally, a user can view many images on a screen and decide to print some. Unfortunately, in the current state of browsing tools, the quality of the printed output is considerably poor if the image is mainly for monitoring, and the viewing time will be extremely long if the image is mainly for printing. Would. Acquisition of "lossless" images is either impossible or must be done as a completely independent download.

[0011]

SUMMARY OF THE INVENTION It is a general object of the present invention to provide a lossy and non-lossy data compression system that utilizes a transform that provides good energy concentration. More specifically, an object of the present invention is to provide a data processing apparatus capable of appropriately processing a natural continuous tone image, a binary / noise-free / shallow pixel depth image, and an image including both types of data. Providing a system, providing a data compression system that can support various resolution / quality image formats, and data including a parser that performs device-dependent quantization according to device characteristics provided by an image output device A compression device is provided.

[0012]

According to a first aspect of the present invention, there is provided a data compression system for compressing image data by a reversible embedding wavelet. The data compression system compresses image data by a binary encoding scheme. It consists of a binary coder and a selection control unit connected to select the wavelet method or the binary method. According to the second aspect of the present invention, the coder of the wavelet system comprises a reversible wavelet transform unit, an embedding ordering quantizer connected to the reversible wavelet transform unit, and a context model connected to the embedding ordering quantizer. Become. According to the invention described in claim 3, the coder of the wavelet system further includes an entropy coder. According to the fourth aspect of the present invention, the wavelet method performs Gray coding. According to the invention of claim 5, the wavelet method and the binary method share one encoder.
According to the invention of claim 6, the data compression system further includes an entropy coder. According to the invention of claim 7, the entropy coder comprises a finite state machine coder, and according to the invention of claim 8, the finite state machine coder comprises a look-up table. According to the ninth aspect of the present invention, the entropy coder comprises a Q coder, according to the tenth aspect, the entropy coder comprises a QM coder, and according to the eleventh aspect, a parallel coder.

[0013] A data compression system according to the present invention as defined in claims 12 to 19, further comprising: a reversible wavelet transform unit;
An embedding ordering quantizer connected to the lossless wavelet transform unit, a context model connected to the embedding ordering quantizer, an embedding binary encoding mechanism, and connected to the context model and the embedding binary encoding mechanism Wherein the reversible wavelet transform, the embedding ordered quantizer and the context model are operable to compress image data with the reversible embedding wavelet, and wherein the binary encoding scheme comprises a binary encoding scheme. And the system further includes a selection control connected to select between a wavelet format and a binary format. According to the thirteenth aspect, the binary system performs Gray encoding. According to the invention of claim 14, the entropy coder comprises a finite state machine coder, and according to the invention of claim 15, the finite state machine coder comprises a look-up table. According to the invention of claim 16, the entropy coder comprises a Q coder, according to the invention of claim 17, the entropy coder comprises a QM coder, and according to the invention of claim 18, the entropy coder is a parallel coder. Consists of a coder. According to the nineteenth aspect, the forward transform comprises a reversible wavelet.

A data compression system according to the present invention is characterized in that a histogram compression mechanism, a reversible wavelet transform section connected to the histogram compression mechanism, and an embedding-ordered quantization connected to the reversible wavelet transform section. A context modeling mechanism connected to the embedding ordering quantizer, and a coder connected to the context modeling mechanism.
According to the twenty-first aspect of the present invention, the histogram compression mechanism creates a boolean histogram, and according to the twenty-second aspect, the histogram compression mechanism maps an integer value to all possible pixel values in the image data. According to the twenty-third aspect, the apparatus further includes a notifying mechanism connected to notify the decoder of the mapping used in the histogram compression mechanism. According to the twenty-fourth aspect, the mapping is notified by a header included in the compressed data received by the decoder. According to the invention of claim 25, one bit in the header, when set, indicates to the decoder that another histogram is to be used for the current tile. According to the invention of claim 26, the decoder equals the dynamic range of the value,
Informed by sending a certain number of bits, each bit of the certain number of bits is set when the corresponding value in the dynamic range is used.

A data compression system according to any one of claims 27 to 43, comprising: a memory for storing a code stream having a header having at least one marker;
At least one output device, comprising a parser coupled to the memory and connected to receive device characteristics from the at least one output device, the parser operable to perform device-dependent quantization. is there. Claim 2
According to the invention of claim 8, the code stream comprises lossless compressed data, and according to the invention of claim 29, at least one marker comprises a number of components used for each tile in the code stream; Indicating sub-sampling and alignment, according to the invention of claim 30, the codestream comprises a main header and a local header is placed before each tile in the codestream. According to the invention of claim 31, according to the invention of claim 32, the main header is applied to all tiles in the code stream, and each local header is applied only to the associated tile.
At least one of the local headers takes precedence over the main header. According to the invention of claim 33, the parser uses a marker in the code stream to quantize the code stream, and according to the invention of claim 34, at least one of the markers indicates frequency information. . According to the invention of claim 35, the data compression system further includes a compression device for generating a code stream. According to the invention of claim 36, the parser comprises a quantization selection device,
According to the thirty-seventh aspect, the quantization selection device performs the conversion and the quantization of the set of images by discarding bit planes of various coefficients. According to the invention of claim 38, one of the tags indicates an importance level in the data in each tile, and according to the invention of claim 39, the tag indicates an importance level locator signal; , The parser stops. According to the invention of claim 40,
43. The tag indicates the number of importance levels to be stored, and
According to the described invention, the tag indicates the number of bytes to be stored. According to the invention of claim 42, the tag includes an instruction for associating the importance level with the number of bytes in each tile, and according to the invention of claim 43, at least one marker indicates the importance level of each tile. Indicates the number of bytes.

[0016]

DETAILED DESCRIPTION A method and apparatus for compression and decompression is described. In the following detailed description of the present invention, various examples are shown, such as the type of coder, the number of bits, the names of signals, etc., in order to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without such an embodiment. In other instances, well-known structures and devices are shown in block diagram form and not shown in detail in order not to unnecessarily obscure the present invention.

A significant portion of the following detailed description is provided by algorithms and symbolic representations of operations on data bits within a computer memory. Such algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most efficiently convey the substance of their work to others skilled in the art. Suppose we have an algorithm, which is generally considered to be a self-consistent sequence of steps leading to the desired result. These steps require physical processing of physical quantities. Usually, but not necessarily, these physical quantities take the form of electrical or magnetic signals that can be stored, transferred, combined, compared, and otherwise processed. It has proven convenient at times, principally for reasons of convention, to represent these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.

It should be noted, however, that such terms are to be associated with the appropriate physical quantities and are merely convenient labels attached to these quantities. As will be apparent from the following description, discussions using terms such as "processing", "operations", "calculations", "judgments", "displays", and the like, unless otherwise specified, refer to physical ( Processes data expressed as electronic quantities and converts them into other data, also expressed as physical quantities, inside a memory or register of a computer system, a similar information storage device, an information transmission device or a display device. , Computer systems or similar electronic computing devices.

The present invention also relates to an apparatus for performing the operations herein. The device may be specially made for the required purpose, or it may be a general purpose computer selectively driven or reconfigured by a built-in program. The algorithms and displays presented herein are not inherently related to any particular computer or other device. A variety of general purpose machines may be utilized in programs according to the teachings herein, or it may be advantageous to make more specialized equipment for performing the necessary method steps. The required structure for a variety of these machines will appear from the description given. Further, in describing the present invention, it is not associated with any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the present invention, as described herein.

The following terms are used in the description that follows. These various terms already have meanings. However, the defined meaning should not be considered as limiting the scope to those terms known in the art. These definitions are provided to assist in understanding the present invention.

Alignment: The degree of shift of a transform coefficient in one frequency band with respect to another frequency band.

Binary coding scheme: A coding scheme for binary, finite pixel depth, or noise-free data. In one embodiment, the binary encoding scheme consists of Gray encoding of pixels and a specific context model.

Bit-significance (bit-signific
ance): A numerical expression similar to a sign absolute value expression, in which a head bit is followed by a sign bit, and further followed by a tail bit, if any. Embedding encodes this numerical expression in bit plane order.

Context model: Information that can be used as a cause for the current bit to be coded and provides previously learned information about the current bit and enables conditional probability prediction for entropy coding. I do.

Embedded quantization: quantization included in the code stream. For example, when the importance levels are ordered from the highest level to the lowest level, quantization is performed by simply truncating the code stream. The same effect can be obtained with tags, markers, pointers, and other signals.

Entropy coder: A device that encodes or decodes a current bit based on probability prediction.

An entropy coder will also be referred to herein as a multiple context binary coder. The context of the current bit is some chosen arrangement of "neighbor" bits, allowing probability prediction for an optimal representation of the current bit (s). In one embodiment, the entropy
The coder includes a binary coder or a Huffman coder.

Fixed length: A method for converting a specific block of data into a specific block of compressed data. For example, some schemes such as BTC (block truncation coding) and VQ (vector quantization). Fixed-length codes are suitable for fixed-rate, fixed-size applications, but their rate and distortion performance is often inferior to variable-rate systems.

Fixed Rate: An application or scheme that must maintain a certain pixel rate and has a channel with limited bandwidth. To achieve this goal,
It is necessary to perform local average compression rather than global average compression. For example, MPEG requires a fixed rate.

Fixed size: An application or scheme with a limited size buffer. To this end, an overall average compression is achieved, for example a print buffer. (The application may be fixed rate and / or fixed size.) Frequency bands: Each frequency band represents a group of coefficients resulting from the same filtering sequence.

Head Bit: In the bit significance representation, a head bit is an absolute value bit from the most significant bit to the first non-zero bit, including the first non-zero bit.

Horizontal Context Model: Context model for embedded wavelet coefficients and binary entropy coder (in one embodiment).

Power: A coding that allows an image to be decompressed in a lossy format and then recompressed into the same lossy codeword.

Image tile: A rectangular area chosen to enable the definition of a grid of consecutive non-overlapping partial images, each having the same parameters. Image tiles affect the buffer size required for the calculation of the transform in wavelet coding. Image tiles can be addressed randomly. The encoding operation processes pixels and coefficient data in one image tile. Thus, image tiles can be parsed or decoded in random order. That is, the image tiles can be addressed at random or decoded according to various distortion levels of extension of the region of interest. In one embodiment, the image tiles are all the same size except those at the top and bottom. The image tile may be of any size less than or equal to the size of the entire image.

Importance level: By defining a specific system, input data (pixel data, coefficients, error signals, etc.)
Are logically grouped into groups with the same visual effect. For example, the most significant bit plane or planes may be more visually important than the lower bit planes. Also, low frequency information is generally more important than high frequency information. Most practical definitions of "visual significance" relate to some error criteria, including the present invention, as described below. However, better visual measures may be incorporated into the systematic definition of visual importance. Different types of data have different levels of visual importance. For example, audio data has an audio importance level.

Overlap transform: A transform in which a single source sample point contributes to multiple coefficients of the same frequency. In that example,
Many wavelets and overlapping orthogonal transforms (Lapp
edOrthogonal Tansform).

Progressive: A code stream that is ordered so that consistent decompression results can be obtained from a portion of the coded data, and the accuracy can be increased by increasing the data. A codestream in which the bit planes of the data are ordered from shallow to deep; in this case,
Usually refers to wavelet coefficient data.

Progressive pixel depth: A codestream in which the bit planes of the data are ordered from shallow to deep.

Progressive pyramid: A series of resolution components whose size decreases by half (1/4 in area) each time the resolution decreases.

Reversible transformation: In one embodiment, an efficient transformation performed by integer arithmetic that can be decompressed based on the compression result.

S-transform: A special pair of reversible wavelet filters consisting of one 2-tap low-pass filter and one 2-tap high-pass filter.

Tail: In the bit significance representation, a tail bit is a lower magnitude bit that is lower than the most significant non-zero bit.

Tail information: In one embodiment, four states that can be taken for coefficients expressed in a bit significance representation. A function of the coefficients and the current bit plane, which is used for the horizontal context model.

Tail-on: In one embodiment,
2 depending on whether the state of the tail information is zero or non-zero
State. Used for horizontal context model.

Tile data segment:
The part of the codestream that completely describes one image tile. In one embodiment, the beginning of the image tile (SO
All data from the tag defining T) to the next SOT or the end of image (EOI) tag.

Conversion coefficient: the result of applying the wavelet transform. In the wavelet transform, the coefficients represent a logarithmic divided frequency scale.

TS conversion: 2.6 (Two-Six) conversion. One 2-tap low-pass analysis filter and one 6-tap
A special reversible wavelet filter pair consisting of a high-pass analysis filter. The synthesis filter is a quadrature mirror filter of the analysis filter.

TT conversion: 2 · 10 (Two-Ten) conversion. 1
A special reversible wavelet filter pair consisting of two 2-tap low-pass analysis filters and one 10-tap high-pass analysis filter. The synthesis filter is a quadrature mirror filter of the analysis filter.

[0049] Unified lossless / lossy: The same compression system provides a lossless or lossy decompressible coded data stream.

Wavelet filter: A high-pass and low-pass synthesis filter and an analysis filter used for wavelet transformation.

Wavelet transform: A transform that uses both "frequency" and "temporal (space)" domain constraints. In one embodiment to be described, the conversion consists of one high-pass filter and one low-pass filter. The resulting coefficients are decimated 2: 1 (critical filtering), and the filters are then applied to the low-pass coefficients.

Wavelet tree: A group of coefficients associated with a single coefficient in the LL part of the highest level wavelet decomposition. The number of coefficients is a function of the number of levels. The span of the wavelet tree depends on the number of decomposition levels. For example, for one-level decomposition, the span of the wavelet tree is four pixels, for two-level decomposition, 16 pixels, and so on.

The present invention provides a compression / decompression system having an encoding unit and a decoding unit. The encoder functions to encode the input data to generate compressed data, while the decoder functions to decode the already encoded data to generate reconstructed data of the original input data. The input data includes various types of data such as images (still images or moving images), sounds, and the like. In one embodiment, the data is digital signal data, but digitized analog data, text data format, and other formats are also possible. The source of the data is, for example, a memory or a communication path for the encoder and / or the decoder.

In the present invention, the components of the encoding unit and / or the decoding unit can be realized by hardware or software used on a computer system. The present invention provides a lossless compression / decompression system. The present invention may also be configured to perform lossy compression / decompression. The present invention provides parsing of compressed data,
It can be configured to execute without decompression.

Overview of the System of the Present Invention The present invention very well represents smooth edges and flat areas found in natural images. The present invention utilizes a reversible embedded wavelet to compress images with a large pixel depth. However, reversible embedded wavelets, other wavelet transform schemes and sinusoidal transform schemes,
I am not good at expressing sharp edges found in text and graphic images. This kind of image is a gray
y) Good encoding can be achieved by performing encoding and then performing context-based bit-plane encoding such as JBIG. In addition, noise-free computer-generated images are well modeled in a binary fashion.

The present invention provides a binary scheme for compressing binary and graphic images. This binary method is
It also improves compression for certain images that do not use the entire dynamic range. In this binary scheme, the present invention encodes a bit plane of an image without using a transform.

FIG. 1 is a block diagram of an embodiment of the compression system of the present invention employing a binary system. Note that the decoding unit of the system operates in the reverse order, and the data flow is the same. In FIG. 1, an input image 101 is input to a multi-component processing mechanism 111. This multi-component processing mechanism 111
Provides optional color space conversion and optional processing for subsampled image components. The method selection mechanism 110 determines whether the image is a continuous tone image or a binary image, or which part of the image has such characteristics. Image data is stored in the system selection mechanism
10, and the method selecting mechanism 110 processes the image data or the portion by the wavelet method processing (block 10).
2, 103, 105) or binary processing (block 104). In the present invention, the decision of which mode to use depends on the data. In one embodiment, scheme selector 110 comprises a multiplexer. The scheme selection mechanism 110 is not used during the decoding operation.

In the wavelet method, the reversible wavelet transform block 102 performs a reversible wavelet transform. The output of the block 102 is a series of coefficients.
The embedding ordering quantization block 103 labels the coefficients in a bit-significant representation to generate an alignment of all the coefficients in the input image 101 (generated by the lossless wavelet transform block 102).

The image data 101 is received and transformed (after appropriate multi-component processing) using a reversible wavelet as described below in a reversible wavelet transform block 102 to provide a multi-resolution image of the image. A sequence of coefficients representing the decomposition is generated. The reversible wavelet transform of the present invention is not computationally complicated. This conversion can be performed by software or hardware without causing any systematic error. Further, the wavelet of the present invention is excellent in energy concentration and compression performance. These coefficients are embedded ordering quantization block 1
03 received.

Embedding ordering quantization block 103
Performs embedding ordering quantization as described later. The result is an embedded data stream.
This embedded data stream allows quantization of the resulting code stream when encoding, transmitting or decoding.
In one embodiment, the embedding ordered quantization block 103 orders and transforms the coefficients into a sign / magnitude format.

The embedded data stream is received in horizontal context model block 105. The horizontal context model block 105 models the data in the embedded data stream based on its significance (described below). In transform mode, the "bit plane" is the transform coefficient importance level plane, and the horizontal context model block 105 arranges the wavelet coefficients into a bit significant representation.

The result of the ordering and modeling is the result of the decision (d
ecisions) (or symbols). In one embodiment, all decisions are sent to one coder. In another embodiment, the decisions are labeled by importance and the decisions at each importance level are processed by separate multiple (physical or virtual) coders. The bitstream is encoded by the entropy coder 106 in order of importance. In one embodiment, the entropy coder 10
6 comprises one or more binary entropy coder. In another embodiment, Huffman coding is used.

In the binary system, a Gray coding block 104 performs Gray coding on the pixels of the input image 101. Gray coding is a bit operation that utilizes some of the correlation between bit planes of pixels. This is because, for arbitrary values x and x + 1, gray (x) and gray (x + 1) differ only by 1 bit in their radix-2 representation.
In one embodiment, gray coding block 104 performs a point-by-point transformation on 8-bit pixels, ie, gray (x) = x XOR x / 2. The present invention is not limited to using this form of Gray coding, nor does it have to use 8-bit sized pixels. However, using the above equation has the advantage that the pixel can be reconstructed with only a part of the available most significant bits, as in the case of progressive transmission in bit plane units. In other words, this form of Gray coding preserves the ordering of the bit significance.

In the binary system, data is encoded for each bit plane using the Gray encoding block 104 and the entropy coder 106. In one embodiment,
The context model within Gray coding block 104 conditions the current bit using spatial and importance level information.

In the case of the binary system, a context model such as JBIG is used for gray coded pixels. In one embodiment, each bit plane of the image tile is encoded separately and each bit is
Using the value of 0 pixel, the image is conditioned and encoded in raster order. FIG. 2 shows the geometric relationship of the context model for each bit of each bit plane in a binary fashion. Conditioning of this bit results in adaptive probability prediction for each unique pattern. It should be noted that when a binary entropy coder is used for bit-plane entropy coding of Gray coded values, some templates may be used for the context model of the binary entropy coder. Figure 3 shows a 7 pixel and bit plane information 2 bits for 2 ^nine context bin (bin).

Using this context and the value of the current bit, entropy coder 106 generates a bitstream. This same binary entropy
A coder 106 is used to encode both conversion mode and binary data. In one embodiment, entropy coder 106 comprises a finite state machine implemented with a look-up table. It should be noted that the present invention can be used with any binary entropy coder, such as a Q coder, a QM coder, or a high-speed parallel coder.

Since the entropy coder 106 is the same for both schemes and the binary context model is simple, only a few additional resources are needed to implement the binary and conversion schemes on the same system. Not required. Furthermore, although the configuration of the context model is different, the required resources for both modes are the same. That is, both modes use the same memory for context storage and use the same binary entropy coder.

The present invention may be performed on the entire image, or more generally, on tiled segments of the image. Some tiles are better compressed by the conversion method,
Some are better with the binary method. Various selection algorithms for the mode to be used are possible. When tiles are used, random access in tile units is possible. In addition, the attention area can be separately decoded with high accuracy. Finally, whether to select the conversion method or the binary method can be determined for each tile.

The image is a dual mode image of the present invention.
Note that the system is still progressive with respect to the bit plane and can be encoded in a hierarchical format as taught by JBIG.

For decoding, 1 in the header of the tile
The bits may be used to specify the scheme used to encode the data. The method selection mechanism 110 is not used. It may be more useful if a lossless mapping from the original dynamic range to a lower dynamic range is possible, such as by histogram compression (described below). Look ahead, such as in JBIG, may be used. This look ahead may use ordinary or deterministic prediction as in JBIG.

Selection of Binary Method or Conversion Method The method selection mechanism 110 selects a binary method or a conversion method. In one embodiment, the input image is encoded in both schemes, and scheme selector 110 selects the scheme with the lower resulting bit rate (assuming lossless compression). In other words, the mode that compresses well is selected. This method
It may seem expensive, but not so much. This is because both the binary method and the conversion method are relatively fast in software and small in hardware. A method derived from this method is to bypass the coder and use the entropy value to determine the lower bit rate.

In another embodiment, the present invention generates a complete (or partial) histogram of the pixel values of an image or a histogram of the differences between pairs of adjacent pixel values. In the case of a histogram of pixel values, a statistical analysis of the data is used, such as selecting a binary method if the peak of the histogram occurs at a value much smaller than the dynamic range of the pixel depth.

In one embodiment, the present invention produces a complete (or partial) histogram of first order differences between pairs of adjacent pixels. In a standard image, such a histogram is exactly Laplacian distributed, and a wavelet scheme would be used. However, when the histogram has no Laplacian distribution peak, the binary method is used.

[0074] Both types of histograms may be generated and used together for system selection.

As will be described later, the dn filter output of the TS conversion or the TT conversion is close to the first order statistics. this is,
Suggest how the transformation will be performed and the histogram generated. A method is selected based on the histogram.
If the scheme is a transform scheme, the system continues to process the already generated transform coefficients. When the binary scheme is selected, the transform coefficients are discarded (or inversely transformed, depending on whether the pixels have been saved) and the system starts the binary scheme.

In another embodiment, previous knowledge of segmentation and / or document type may assist in deciding which method to choose.

If longer encoding times are available, the size of the tiling can be chosen to maximize the benefits of the two schemes.

It should be noted that in one embodiment, the system of the present invention does not include binary coding, and therefore utilizes only lossless embedded wavelet compression (CREW) and decompression.

Wavelet Decomposition The present invention first utilizes a reversible wavelet,
Performs decomposition of an image (as image data) or other data signal. In the present invention, the reversible wavelet transform realizes a complete reconstruction system capable of lossless restoration of a signal having an integer coefficient by an integer operation. An efficient reversible transform is due to a transform matrix with a determinant of 1 (or nearly 1).

The present invention can provide lossless compression with finite precision arithmetic by utilizing a reversible wavelet. The result produced by applying the lossless wavelet transform to the image data is a sequence of coefficients.

The reversible wavelet transform of the present invention can be realized using a set of filters. In one embodiment, the filter comprises one 2-tap low-pass filter and one 6-tap low-pass filter.
It is a tap high-pass filter. In one embodiment, these filters are addition and subtraction (and hard-wire bit shifting).
Only realized.

One embodiment of the present invention utilizing the Hadamard transform is a complete reconstruction system.

To obtain information about the Hadamard transform,
Anil K. Jain, "Fundamentals of Image Process"
ing ", page 155. The inverse of the Hadamard transform is referred to herein as the S-transform.

In the S conversion, the output can be defined as follows using the general index n.

[0085]

(Equation 1)

The factor 2 in the transform coefficient addressing is the result of implicit 1/2 subsampling.
This transformation is reversible and its inverse is as follows:

[0087]

(Equation 2)

Symbol

[0089]

[Outside 1]

Means rounding down, that is, truncation, and is sometimes called a floor function. Similarly, the ceiling function

[0091]

[Outside 2]

Means rounding up to the nearest integer.

Another example of a complete reconstruction system is 2 ·
6 (TS) conversion. The reversible TS transform is defined by the following equation for the two outputs of the low-pass and high-pass filters.

[0094]

(Equation 3)

The TS conversion is reversible, and the inverse conversion is as follows.

[0096]

(Equation 4)

Here, p (n) must first be calculated by the following equation.

[0098]

(Equation 5)

The result from the low pass filter can be used twice (in the first and second terms) in the high pass filter. Therefore, the result of the high-pass filter can be obtained only by performing two additional additions.

Another example of a complete reconstruction system is 2 ·
10 (TT) conversion. The reversible TT transform is defined by the following equation regarding two outputs of a low-pass and a high-pass filter.

[0101]

(Equation 6)

This d (n) equation can be simplified using s (n) (furthermore, integer division by 64 is
It can be rounded by adding 32 to the molecule). This gives the following equation:

[0103]

(Equation 7)

This TT transform is reversible, and its inverse transform is given by the following equation.

[0105]

(Equation 8)

Here, p (n) must first be calculated by the following equation.

[0107]

(Equation 9)

In both the TS conversion and the TT conversion,
Like the S-transform, the low-pass filter is made such that the range of the input signal x (n) is the same as the range of the output signal s (n). That is, there is no increase in the smooth output. When the input signal is b bits deep, the output signal is also b bits deep. For example, if the signal is an 8-bit image,
The output of the low-pass filter is also 8 bits. This is an important property for a pyramid system, for example, where the smoothed output is further resolved by successively applying a low-pass filter. In prior art systems, the range of the output signal is greater than the range of the input signal, which makes continuous application of the filter difficult. In addition, since there is no systematic error due to rounding when performing conversion by integer arithmetic, all errors of the lossy system can be controlled by quantization. Furthermore, the low-pass filter is 2
Having only one tap results in a non-overlapping filter. This property is important for hardware implementation.

In one embodiment, the multiplication by 3 and 22 is realized by shifting and adding as shown in FIG. FIG.
At 2, the s (n) input is connected to a multiplier 1501;
Multiplier 1501 multiplies the s (n) input by two. In one embodiment, this multiplication is implemented as a single digit left shift of the bits of the s (n) signal. The output of the multiplier 1501 is the adder 150
2 and is added to the s (n) signal. The output of adder 1502 is a 3s (n) signal. The output of adder 1502 is also multiplied by 2 by multiplier 1503. Multiplier 15
03 is implemented as a one-digit left shift. Multiplier 150
The output of 3 is added to the output of the multiplier 1504 by the adder 1505, and the multiplier 1504 multiplies the s (n) signal by 16 times by shifting left by four digits. The output of the adder 1505 is 22 s
(n) signal.

The stringent reversibility requirements for filters can be relaxed by noting the following:
The high-pass coefficients are encoded and decoded in a certain order. The pixel value corresponding to the previously decoded high-pass coefficient is accurately known and can be used for current high-pass filtering.

The TS transform and the TT transform have non-overlapping low-pass synthesis filters and high-pass analysis filters. Only the high-pass synthesis filter and the low-pass analysis filter are overlap filters.

The TS filter has good characteristics with respect to tile boundaries. Consider the case where the tile size is a multiple of the tree size. Then, consider applying a transform to a portion of a signal, such as would occur when an image is divided into tiles. Since the lowpass analysis filter is not an overlap filter, the lowpass coefficients are not affected by tiling. That is, if that portion of the signal has a uniform number of signals, its low-pass coefficients are the same as when the entire signal is transformed.

During decoding, if no high-pass coefficients exist for quantization and the image is reconstructed at the highest compression ratio using only SS coefficients, a low-pass synthesis filter is used across the tile boundaries. The inverse transform may be performed on the entire signal using low-pass coefficients. Since the tiling of the SS coefficients does not change, the solution is exactly the same as when tiling is not used. This removes artifacts caused by performing a forward transform on a portion of the signal.

During decoding, if the high-pass coefficients are present (but the high-pass coefficients are quantized so that their values have some uncertainty), the overlap 1D low-pass analysis filter operation The following can be done on a sample where one crosses one boundary and enters another. For samples that do not cross tile boundaries, the minimum and maximum possible reconstruction values are determined based on the filters actually used. For that sample, by using only the low-pass coefficients (and the low-pass filter), and by traversing the tile boundaries, the reconstruction values (ie, the overlap predictions) that would have been so are determined. If the overlap prediction is between the minimum and maximum possible reconstruction values (including those values), the overlap prediction is used. Otherwise, of the possible maximum and minimum reconstruction values, the one closest to the overlap prediction is used. This reduces the artifacts caused by performing a forward transform on a signal fragment.

A reconstruction value is selected each time the 1D filter operation is performed. If this is done correctly, each high-pass coefficient will be given exactly one valid reconstruction value, and the selection will not be able to propagate the error to multiple levels of the transform.

Nonlinear Image Model One embodiment of the present invention utilizes a wavelet filter, which is a reversible approximation of a linear filter such as a TS or TT transform. In one embodiment, a reversible nonlinear filter may be used. One type of nonlinear filter similar to the TS conversion and the TT conversion is as follows.

[0117]

(Equation 10)

The inverse transform is the same as in the case of the TS transform and the TT transform, except that p (n) is as follows.

[0119]

[Equation 11]

In this embodiment, q (n) is smooth (smoot
h) x from coefficient (and previous detail coefficient if necessary)
This is a prediction related to (2n) -x (2n + 1). This prediction utilizes a non-linear image model. In one embodiment, the non-linear image model is a Huber-Markov random field. This nonlinear image model is exactly the same in the forward transform and the inverse transform s. In the case of an iterative image model, the number and order of iterations are the same.

One example of the nonlinear image model is as follows. For a certain number of iterations, each value (pixel or low-pass coefficient) x (k) is adjusted to a new value x '(k), where k is 2n or 2n + 1. Any number of repetitions may be used, but in one embodiment, three repetitions are used. The value of q (n) is obtained from the last iteration.

[0122]

(Equation 12)

At each iteration, the change y (k) for each x (k) is calculated.

[0124]

(Equation 13)

Here, A is the rate of change, which may be any positive value. In one embodiment, A = 1. Bi is a difference detector. For example, in a one-dimensional case, Bi is as follows.

[0126]

[Equation 14]

In the two-dimensional case, there will be four Bi values for detecting the difference between horizontal, vertical and two diagonal directions. Another difference operator may be used as Bi.

[0128]

(Equation 15)

Here, T is a threshold, which may be any positive value. T represents a difference constituting an edge of the image. In one embodiment, T = 8.

X '(2n) + x' (2n + 1) = x (2n) + x (2n +
Under the constraint condition 1), the pair of values x '(2n) and x' (2n + 1) is adjusted by the pair of changes y (2n) and y (2n + 1). This is achieved by combining the changes y (2n), y (2n + 1) into a single change y '(2n), the largest change supported by both changes.

When y (n) and y (2n + 1) are both positive or negative, y ′ (sn) = 0. Otherwise, if | y (2n) | <| y (2n + 1) |, then y ′ (2n) = y (2n), otherwise y ′ (sn) = − y (2n + 1). Then, x '(2n) = x (2n) + y' (2n) x '(2n + 1) = x (2n + 1) -y' (2n). For more information on the Huber-Markov random field, see R. Schultz and R.S. L. Steven
son, "Improved definition imageexpansion", Pro
ceedings of IEEE International Conference on Ac
oust., Speech and Signaling Processing, vol. III,
pp. 173-176, San Francisco, March 1992.

In some embodiments, the transform is extended from one-dimensional to two-dimensional, but for that extension first q (n) =
Conversion is performed separately for each dimension as 0. Next, three q (n) values are calculated for the LH, HL, and HH values by a similar application of the image model.

Two-dimensional Wavelet Decomposition Multi-resolution decomposition is performed using the low-pass filter and the high-pass filter of the present invention. The number of decomposition levels is variable and may be any number, but currently the number of decomposition levels is 2
The levels are from level 5 to level 5. The maximum number of levels is log2
(Maximum value of length or width).

The most common way to perform a transformation on two-dimensional data, such as an image, is to apply the one-dimensional filter separately, that is, apply it along the rows and then along the columns. is there. The first level of decomposition yields four different coefficient bands, referred to herein as LL, HL, LH, HH. These characters correspond to the smooth filter defined earlier and the det (det
ail) Low (L) and high (H) meaning the application of the filter
Represents Therefore, the LL band is made up of coefficients obtained from the row and column direction smoothing filters. It is a common practice to arrange the wavelet coefficients in a form as shown in FIGS.

Each frequency sub-band of the wavelet decomposition can be further decomposed. The most common approach is to further decompose only the LL sub-blocks, which will involve further decomposing the LL frequency subbands at each decomposition level as they are generated. Such multiple decomposition is called pyramid decomposition (FIGS. 4 to 7). Each decomposition is indicated by a symbol LL, LH, HL, HH and a decomposition level number. The TS filter of the present invention, T
With any of the T filters, pyramid decomposition does not increase the coefficient size.

For example, if the reversible wavelet transform is applied recursively to an image, the first level of decomposition operates on the finest detail or resolution. First
At the decomposition level, the image is decomposed into four sub-images (ie, sub-bands). Each subband represents one spatial frequency band. The first level subband is LL
0, LH0, HL0, HH0. Since the process of decomposing the original image involves サ ブ subsampling in both the horizontal and vertical dimensions, as shown in FIG.
Each of the level sub-bands LL0, LH0, HL0, HH0 has one-fourth of the number of pixels (or coefficients) of the image of the input.

The sub-band LL0 simultaneously includes low frequency information in the horizontal direction and low frequency information in the vertical direction. Generally, most of the image energy is concentrated in that subband. The sub-band LH0 includes low-frequency information in the horizontal direction and high-frequency information in the vertical direction (for example, horizontal edge information). Subband HL0 is
It includes high frequency information in the horizontal direction and low frequency information in the vertical direction (for example, information on vertical edges). The sub-band HH0 includes high-frequency information in the horizontal direction and high-frequency information in the vertical direction (for example, information on texture or oblique edges).

The second, third, and fourth lower decomposition levels that follow are each created by decomposing the previous level low frequency LL subband. By decomposing the first level sub-band LL0, as shown in FIG. 5, the second level sub-bands LL1 and L
H1, HL1, and HH1 are created. Similarly, subband L
By dividing L1, as shown in FIG. 6, the third level sub-bands LL2, LH2, HL with coarser definition are obtained.
2, HH2 is generated. Further, as shown in FIG. 7, the sub-band LL2 is divided, so that the sub-bands LL3, LH3, HL3, and HH3 of the fourth level with lower definition are provided.
Is made. With 2: 1 subsampling, each subband at the second level is one-sixteenth the size of the original image. Each sample (or pixel) at this level represents a slightly smaller detail at the same location in the original image. Similarly, the third
Each sub-band of the level is 1 / 64th the size of the original image. Each pixel at the third level represents a fairly coarse detail of the same location in the original image. Each sub-band at the fourth level is 1 / 256th of the size of the original image.

Since the decomposed image is physically smaller than the original image due to sub-sampling, all of the decomposed sub-bands can be stored using the memory used for storing the original image. That is, in the case of three-level division, the original image and the decomposition sub-bands LL0 and LL1 are discarded and not stored.

Although only four sub-band decomposition levels have been shown, it is possible to generate more levels depending on the requirements of the particular system. Various parent-child relationships may also be defined by other transforms such as DCT or one-dimensionally arranged subbands.

Pyramid Decomposition Each frequency sub-band of the wavelet decomposition can be further decomposed. In one embodiment, only the LL frequency subband is decomposed. Such a decomposition is referred to herein as a pyramid decomposition. Symbol LL, LH, H
Each decomposition is indicated by L, HH and the decomposition level number. In addition,
According to the wavelet filter of the present invention, the pyramid decomposition does not increase the coefficient size.

In another embodiment, other sub-bands in addition to LL may be resolved. In the following description, the term “LL” may be used interchangeably with “SS” (“L” = “S”). Similarly, the term "H" is also
D "may be used interchangeably.

Wavelet Tree Structure The wavelet coefficients of the pyramid decomposition have a natural and useful tree structure. Note that there is only one LL frequency sub-block corresponding to the final decomposition level. On the other hand, the number of LH, HL, and HH bands is equal to the number of levels. This tree structure reveals that the parents of the coefficients in a certain frequency band are the coefficients in the same frequency band of lower resolution and have the same spatial positional relationship.

The root of each tree is a purely smooth coefficient. For a two-dimensional signal such as an image, the root of the tree has three "children", and the other nodes each have four children. This hierarchical tree is not limited to two-dimensional signals. For example, in the case of a one-dimensional signal, the root has one child, and nodes other than the root each have two children.
Higher dimensions are derived from the one-dimensional and two-dimensional cases.

FIG. 8 shows a parent-child relationship between two consecutive levels. In FIG. 8, the coefficient of A is a direct parent to B, C, and D, but the coefficient (E
And H, F and I, and G and J). For example,
B is the parent for 4 coefficients near E, 16 coefficients near H, and so on.

The process of multi-resolution decomposition may be performed using a filter sequence.

For an example of a one-dimensional two-level transformation implemented using a one-dimensional exemplary filter, see US patent application Ser. No. 08 / 498,695, filed Jun. 30, 1995, entitled "Method and Apparatus For Compression." Us
ing Reversible WaveletTransforms and an Embedd
ed Codestream ") and U.S. patent application Ser.
No. 036 (Received June 30, 1995, "Reversible W
avelet Transformand Embedded Code-stream Manip
ulation ”).

Execution of Forward Wavelet Transform In the present invention, the wavelet transform is executed by a 1-D (dimensional) pass in the horizontal direction and then by a 1-D pass in the vertical direction. The number of levels determines the number of iterations. FIG. 9 represents a four-level decomposition using a forward TT transform filter as previously defined.

In another embodiment, another reversible wavelet transform filter, such as an S-transform, can be used instead of the TT transform at any point in the horizontal or vertical wavelet transform at any level. In one embodiment, a four-level decomposition is performed using the TT transform in both the horizontal and vertical directions. In one embodiment, in a four-level decomposition,
Two of the TT transforms are replaced by S transforms. This has less compression loss, but has a greater effect on memory usage. The horizontal and vertical transforms will be applied alternately.

Note that a combination of the S-transform and the TT-transform may be used for performing the horizontal and vertical conversions. The order of the transformations can be mixed, but to be completely reversible, the decoder must know the order and perform the inverse operation in the reverse order. As described below, the decoder may be informed of the conversion order in the header.

Embedding Ordering In the present invention, the coefficients generated as a result of wavelet decomposition are entropy coded. In the present invention, the coefficients are first subjected to embedded coding, in which the coefficients are ordered in visually important order, or more generally some error criterion ( For example, the coefficients are ordered taking into account the distortion criterion). Error or distortion criteria include peak error and mean 2
Multiplicative error (MSE). In addition, bit-significance spatial locatio
n) may be ordered by preference (vertical, horizontal, diagonal, etc.) to prioritize validity for database queries.

The ordering of the data is performed to generate an embedded quantized version of the code stream. In the present invention, two ordering methods are used. One is for ordering the coefficients, and the other is for ordering the binary values in the coefficients. The ordering of the present invention produces a bitstream that is subsequently encoded by a binary entropy coder.

Tile In the present invention, before transformation and encoding, the image is divided into tiles. A tile is a partial image of the entire image that is completely independently encoded, and is defined by a regular rectangular grid arranged on the numbered images as shown in FIG. The tiles at the right and bottom edges have various sizes depending on the original image and the tile size.

The tiles may have any height and width equal to or less than the image size, but the selection of the tile size affects the performance. Small tiles, especially tiles with small vertical dimensions in raster-ordered images, can reduce work area memory. However, if the tiles are too small,
gnaling) overhead, loss of conversion efficiency at tile boundaries, rising adaptation of entropy coder
Two factors reduce compression efficiency. The size of the tile is a multiple of the magnitude of the lowest frequency component (CREW tree),
That is, it is advantageous to use a function of the number of levels (2 raised to the power of the number of levels). Depending on the size of the original image, 128 × 128 or 256 × 256 tiles may be suitable for many applications.

Tiles may be used for compression of a series of images. So the tiled image is
It may be a temporally different image (like a movie) or a spatially different image (like a 3D section like MRI). There is no special way to signal this, but CMT may be used.

Transforms, context models, and entropy coding operate only on pixels and coefficients of one image tile. Therefore, the image tiles can be analyzed or decoded out of order, ie, can be addressed randomly, or can be decoded to different distortion levels due to the extension of the region of interest.

All pixel data of one image tile is available to the encoder at one time, eg, buffered in memory. Once the pixel data has been transformed, all coefficient data is available for the horizontal context model. Since all coefficients can be accessed randomly, the order of embedding within the image tiles can be arbitrary as long as the encoder and decoder know. Since the entropy coder is careless about this ordering, it must be chosen carefully because its order has a large effect on the compression ratio.

Image tiles are arranged in a rectangular tree (LL).
Coefficient and all its descendants). The number of pixels in each tree is a function of the number of levels in the wavelet decomposition.

The image reference grid is a minimum grid plane in which the size of each component is an integral multiple of a grid point. This implies that for most images, the image reference grid is the same as the mode component.

If one component image or all components are the same size image, the image reference grid is the same size as the image (for example, the grid points are pixels of the image). In the case of an image in which all of the plurality of components are not the same size, the size is defined as an integral multiple of the image reference grid point. For example, CC
The IR601 YCrCb color component system is defined to have two Y components for each Cr and Cb component. Therefore, Y
The components define an image reference grid, and the Cr and Cb components cover two units in the horizontal direction and one unit in the vertical direction.

Bit Significance Representation In one embodiment, the embedding order used for the binary values in the coefficients is bit plane order. The coefficients are represented in a bit significance representation. The bit significance representation is not the most significant bit (MSB) but the code (si
gn) bits are the sign-magnitude representation encoded with the first non-zero magnitude bit.

The number represented in the bit significance format has three types of bits: a head bit, a tail bit, and a sign bit.
The head bits are all zero bits from the MSB to the first non-zero magnitude bit, and the first non-zero magnitude bit. The significance of the coefficients is determined by the bit plane in which the first non-zero magnitude bit resides. The bits after the first non-zero magnitude bit to the LSB are tail bits. The sign bit only represents a sign. A number of non-zero bits with MSBs, for example ± 2n, has only one head bit. Zero coefficients have neither tail bits nor sign bits. FIG.
Here is an example of a significance expression.

For values that are non-negative integers, such as those that occur in connection with pixel brightness, the possible order is bit-plane order (eg, most significant bit plane to least significant bit plane). In embodiments where two's complement negative integers are also allowed, the sign bit embedding order is the same as the first non-zero bit of the absolute value of the integer. Therefore,
Until one non-zero bit is encoded, the sign bit is not considered. For example, according to the sign / absolute value notation,
The 16-bit number of −7 is 10000000000001111. On a bit-plane basis, the first twelve decisions are "meaningless" or zero. The first one bit is found in the thirteenth decision. Next, the sign bit ("negative") is encoded. After the sign bits are encoded, the tail bits are processed. The 14th and 15th decisions are both "1".

Since the coefficients are encoded from the most significant bit plane to the least significant bit plane, the number of data bit planes must be known accurately. In the present invention, the number of bitplanes is determined by finding the upper limit of the absolute value of the coefficient value calculated from the data or derived from the image depth and the filter coefficients. For example, when the upper limit is 149, there are eight significant bits, or eight bit planes. Due to the speed of the software, bitplane encoding may not be used. In another embodiment, the bit planes are encoded only when the coefficients make sense as binary numbers.

Coefficient Alignment In the present invention, coefficients are aligned with each other before bit plane coding. This is because, like FFT and DCT, coefficients in different frequency subbands represent different frequencies. The present invention enables quantization by performing coefficient alignment. A coefficient with a smaller quantization weight is aligned to the bit plane side earlier (for example, shifted to the left). Thus, if the stream is truncated, these coefficients will have more bits defining them compared to the heavier quantized coefficients.

In one embodiment, the coefficients are aligned for the best rate-distortion performance from an SNR or MSE perspective. Many alignments are possible, including near-optimal alignments with respect to statistical error criteria such as MSE. Alternatively, the alignment may allow for physchovisual quantization of the coefficient data. Alignment has a significant effect on image quality (in other words, on the rate-distortion curve), but has little effect on the final compression ratio of a lossless system. Other alignments may correspond to special quantization, region-of-interest fidelity encoding, or resolution progressive alignment.

[0167] The alignment may be signaled in the header of the compressed data. The coefficients are encoded in bit significance order, with the highest significance level being derived from the coefficients in the coding unit. The sign bit of each coefficient is not coded to the highest significance level for which the coefficient has a non-zero magnitude bit. This has the advantage that the sign bit of the coefficient whose absolute value is zero is not coded. Also, the sign bit is not encoded to the point in the embedded code stream at which it is relevant. The alignment for the various sized coefficients has no effect on the efficiency of the entropy coder, since both the encoder and the decoder are known.

FIG. 13 shows the bit depths of the coefficients in two-level TS conversion and TT conversion decomposition of an image of b bits / pixel.
Shown in FIG. 14 is an example of a frequency band multiplier used for coefficient alignment in the present invention. For coefficient alignment, the size of the 1-HH coefficient is used as a reference, and a shift is given relative to this size.

In one embodiment, to generate an alignment of all the coefficients in the image, the coefficients are shifted in view of the absolute value of the largest coefficient. The aligned coefficients are then processed from the highest importance level (MSIL) to the lowest importance level (LSIL) in bit plane units called importance levels. Since the sign bit is not part of the MSIL, it is not coded up to the last head bit of each coefficient. Importantly, the alignment only controls the order in which bits are sent to the entropy coder. The padding, shifting, storage, and encoding of the extra zero bits are not actually performed.

Table 1 shows an example of the alignment.

[0171]

[Table 1]

Since the alignment for the various size coefficients is known to both the encoder and the decoder,
It has no effect on the efficiency of the entropy coder.

Note that the coding units of the same data set may have different alignments.

Ordering of Code Stream FIG. 15 shows the ordering of the coded stream and the ordering within the coding unit. In FIG. 15, a coding unit 1002 follows the header 1001 in order from the highest band to the lowest band. Inside the coding unit, the LL coefficient 1003 is stored in raster (line) order without being coded. After the LL coefficients, the significance levels are entropy coded, one bit plane at a time, from the most significant bit plane to the least significant bit plane. At this time, the first bit plane of all coefficients is encoded, then the second bit plane is encoded, and so on. In one embodiment, the alignment is specified in header 1001.

In one embodiment, the LL coefficients, when they are 8-bit values, are stored only in raster order without encoding. When the size of the LL coefficient is less than 8 bits, the LL coefficient is padded to 8 bits.
When the LL coefficient is longer than 8 bits, the LL coefficient is stored as follows. First, the most significant 8 bits of each coefficient are not encoded and stored in raster order. Next, the remaining lower bits of the coefficients are packed and stored in raster order. For example, in the case of a 10-bit LL coefficient, four LLs
The least significant bit of the coefficient is packed into one byte. In this way, regardless of the actual image depth, 8 per coefficient
Since bit LL data can be obtained, a simplified image or a preview image can be quickly generated.

The order in which the coefficients are processed during each bit plane period is from lower resolution to higher resolution and from lower frequency to higher frequency. The order of the coefficient subbands in each bit plane is from higher levels (low resolution, low frequency) to lower levels (high resolution, high frequency). Within each frequency subband, the encoding is performed in a certain order. In one embodiment, the order is raster order, 2x2 block order, zigzag order, Peano scan order, and so on.

In the case of four-level decomposition using the code stream of FIG. 15, the order is as follows. 4-LL, 4-HL, 4-LH, 4-HH, 3-HL, 3-LH, 3-HH, 2-HL, 2-LH, 2-HH,
1-HL, 1-LH, 1-HH.

Separating the code stream data by importance is advantageous for storing the data on a medium or transmitting it over a noisy channel. Error correcting / detecting codes with different degrees of redundancy can be used for different parts of the data. The code with the highest redundancy can be used for the header of the LL coefficient. (Based on importance)
For the less important entropy coded data, an error correction / detection code with low redundancy can be used. When an uncorrectable error occurs, the data packet must be discarded (quantized), re-transmitted from the communication channel, or re-read from the storage device to determine the data packet. Importance levels are also available. For example, a high redundancy error correction BCH code (Reed-Solomon
Codes, etc.) may be used for the most important quarter of the header data, LL data, entropy coded data. Remaining 4 of entropy encoded data
The third is a low redundancy error detection checksum or CRC.
(Cyclic Redundancy Check). In one embodiment, packets using the BCH code will always be retransmitted and not discarded, while packets with a checksum or CRC code will not be retransmitted and will be discarded after an unsuccessful attempt to transmit data. Would.

In one embodiment, the header data specifies the error correction / detection code used for each portion of the data. In other words, the information in the header indicates when to switch the error correction code. In one embodiment, the error correction / detection code is only changed between packets used for the communication path or between blocks used for the storage medium.

FIG. 26 shows a code stream in which an LL coefficient (1902) which is not coded and entropy coded data 1903 follow the header 1901 in the order of embedding.
As shown, the header 1901 and the LL coefficient 1902 use the code with the highest redundancy, while the entropy coded data 1903 uses the code with the lowest redundancy. The present invention
A sliding scale may be employed in which many different codes are used, from highest redundancy to lowest redundancy.

Horizontal Context Model An embodiment of the horizontal context model used in the present invention will be described below. This model utilizes the bits in a coding unit based on the spatial and spectral dependencies of the coefficients. The available binary values of the adjacent and parent coefficients may be used to generate the context. But,
Context affects the decodability and also influences the efficient adaptation somewhat.

Modeling Coefficients with a Horizontal Context Model The present invention provides a context model for modeling a codestream generated with coefficients in embedded bit significance order for a binary entropy coder. In one embodiment, the context model comprises a run-length count, a spatial model, a sign bit model, and a tail bit model. Run length count measures the run of bits in the same state. The spatial model includes information on the neighborhood coefficient and the parent coefficient regarding the head bit.

FIG. 16 shows adjacent coefficients of all the coefficients of the coding unit. In FIG. 16, the adjacent coefficients are represented in an easy-to-understand geographical notation (for example, N = North, NE =
Northeast, etc.). Given a coefficient, eg, P in FIG. 16, and a current bitplane, the context model can utilize any information available from all coding units prior to that bitplane. In the case of the present context model, the parent coefficient of the coefficient of interest is also used.

Horizontal Head Bit Context Model The head bit is the most compressible data. Therefore, a large amount of context or conditioning is used to increase the compression ratio. Rather than using the value of the adjacent or parent coefficient to determine the context for the bit of interest of the coefficient of interest, the information is grouped into two bits, referred to herein as tail information. This information may be stored in memory or may be calculated dynamically from neighboring or parent coefficients. The tail information indicates whether the first non-zero magnitude bit has already been found (eg, whether the first "on" bit has already been found).
And, if found, how many previous bit planes it was. Table 2 describes the tail information bits.

[0185]

[Table 2]

From the 2-bit tail information, 1-bit “tail-on” indicating whether the tail information is zero or not.
The bits are combined. In one embodiment, the tail information and tail-on bits are updated immediately after the coefficients are encoded. In another embodiment, the update occurs later to allow for parallel context creation.

Furthermore, these two bits are used to indicate the importance level to be encoded. The first two bit planes have the value 0, the second two bit planes have the value 1, the third two bit planes have the value 2, and the remaining bit planes have the value 3.
Use In addition, there is run-length encoding of all zero head bits.

The 10-bit context for the head bit is 2-bit information for each of the parent coefficient and the W coefficient,
It consists of 1-bit information of each coefficient of N, E, SW and S, and 2-bit importance level information.

In one embodiment, no tail information is used for some or all of the frequency bands. This allows the frequency band to be decoded without having to decode its parent in advance.

In another embodiment, one alignment is used to assign each frequency band to a bit plane importance level. Another alignment is used to determine the parent tail-on information, which uses fewer parent bitplanes than the actual encoded bitplanes. This allows several bit planes in a certain frequency band to be decoded without decoding corresponding parent bit planes of the same importance level (see FIG. 51). For example, an image can be encoded by pyramid alignment without using parent tail-on information based on MSE alignment (see FIG. 50).
This allows the decoder to simulate the MSE alignment, or to simulate any alignment between the pyramid alignment and the MSE alignment, and decode with the pyramid alignment.

FIG. 29 shows context dependency. Children are conditioned on their parents. Thus, a parent must be decoded before decoding its children, especially when decoding with an alignment that is different from the alignment used during encoding.

Horizontal Sign (Bit) Context Model After the last head bit, the sign bit is encoded. There are three code contexts depending on whether the N coefficient is positive or negative and whether the code has not been coded yet.

Horizontal Tail Bit Context Model There are three contexts for tail bits depending on the value of the tail information of the coefficient of interest. (When trying to encode tail bits, the tail information value is 1, 2
Or note that it is limited to 3) Steps for the horizontal context model The context model of the system uses up to 11 bits to describe the context. The whole number does not have to be specified. The meaning of each bit position depends on the previous binary value. First, only one context is used to provide some "run coding" with head bits. When there is no run of head bits, each bit is coded with adjacent and parent coefficients that contribute to a context. Specific examples of the steps are as follows.

1) It is determined whether look-ahead should be performed. If the tail information of the next N coefficients and their north adjacent coefficients is all zero, the system proceeds to step 2. Otherwise, step 3 for the next N coefficients
Proceed to. In one embodiment, N = 16.

2) Perform a look-ahead procedure. If the bit of the current bitplane to be encoded of the next N coefficients is zero, a 0 is encoded and the system moves from step 1 to the next N coefficients. Otherwise, one 1 is encoded and the system goes to step 3 for the next N coefficients.

3) The state of the coefficient of interest is determined and encoded. If the tail information of the coefficient of interest is 0, the bits of the current bit plane of the coefficient of interest are the two bits of the tail information of the west coefficient and the parent coefficient (optional), and the tail-on bits of the northwest, east, southwest, and south coefficients. And importance level information 2
Encoded with 1024 possible contexts created by the bits, the system proceeds to step 4. In one embodiment, since the parent coefficient is not used, the context is created only from the adjacent coefficient and the importance level information. If the tail information of the coefficient of interest is not 0, the bit of the current bit plane of the coefficient of interest is a tail bit, and is encoded by three contexts formed from two bits of the tail information of the coefficient of interest.

4) Determine the state of the current head bit, and encode the sign bit if necessary. If the bit in the current bit plane of the coefficient of interest is 1, the sign of the coefficient of interest is encoded by the three possible contexts created by the tail-on bit and the sign bit of the north coefficient.

FIG. 17 is a flowchart of the above-described process. In FIG. 17, white blocks have nothing to do with coding, and black blocks have no relation to coding. Although not shown, each entropy coded decision (decision)
One context is defined for The foregoing operations and flows will be understood by those skilled in the art.

One example of a horizontal context model is disclosed in US patent application Ser. No. 08 / 498,695, along with an example of a sign / magnitude unit that converts input coefficients to sign / magnitude format.
(Received June 30, 1995, “Method and Apppar
atus For CompressionUsing Reversible Wavelet
Transforms and an Embedded Codestream ") and U.S. patent application Ser. No. 08 / 498,036 (1995).
June 30, 2013, "Reversible Wavelet Transfor
m and Embedded Codestream Manipulation ").

Entropy Coding In one embodiment, the entropy coding performed by the present invention is performed by a binary entropy coder. In one embodiment, entropy coder 106 comprises a Q coder, a QM coder, a finite state machine, or a fast parallel coder. A single coder may be used to generate a single output code stream. Alternatively, multiple (physical or virtual) coders may be used to generate multiple (physical or virtual) data streams.

In one embodiment, the binary entropy coder of the present invention comprises a Q coder. For information on the Q coder, see Pennebaker, W.C. B., et al., "A
nOverview of the Basic Principles of the Q-cod
er Adaptive Binary Arithmetic, "IBM Journal of
Research and Development, Vol. 32, pg. 717-26,19
I want to read 88. In another embodiment, the binary entropy coder uses a QM coder, which is a well-known efficient binary entropy coder. QM coders are especially efficient for bits with very high probability skew. QM coders are used in both JPEG and JBIG standards.

[0202] The binary entropy coder may be a finite state machine (FSM) coder. Such a coder provides a simple conversion of probabilities and outcomes into a compressed bitstream. In one embodiment, the finite state machine coder includes both an encoder and a decoder,
This is realized by table lookup. A variety of probability prediction methods are available for such a finite state machine coder. The compression ratio for probabilities close to 0.5 is very good. The compression ratio for the large skew probability depends on the size of the lookup table used. Like a QM coder, a finite state machine coder is useful for embedded bitstreams because the decisions are encoded in chronological order. Since the output is determined by a look-up table, there is no concern about "carry over" problems. In practice, unlike a Q coder or a QM coder, there is a maximum delay between encoding and generating the compressed output bits. In one embodiment, the finite state machine coder of the present invention is disclosed in U.S. Pat.
No. 272,478 "Method and Apparatus for Entro
py Coding ".

In one embodiment, the binary entropy coder of the present invention comprises a high speed parallel coder. Both QM and FSM coders need to be encoded or decoded bit by bit. A high-speed parallel coder processes several bits in parallel. In one embodiment, the high speed parallel coder is implemented on VLSI hardware or a multiprocessor computer without sacrificing compression performance. One example of a high speed parallel coder that can be used in the present invention is U.S. Pat. No. 5,381,145 issued Jan. 10, 1995, "Metho.
d and Apparatus for Paraallel Decoding and Enco
ding of Data ”.

Most efficient binary entropy
The coder is speed limited by the basic feedback loop. One possible solution is to split the input data stream into multiple streams and feed them to multiple encoders in parallel. The output of the encoders is a plurality of variable length encoded data streams. One challenge with this type of method is how to transmit the data on a single channel. US Patent No. 5,381,14
The high-speed parallel coder described in No. 5 solves this problem.
The solution is to interleave the coded data streams.

Many of the contexts utilized in the present invention are constant probabilities, which makes finite state machine coders, such as B coders, particularly useful. It should be noted that if the system utilizes a probability close to 0.5, both the fast parallel coder and the finite state machine coder disclosed in the above patent operate more efficiently than the Q coder. Thus, both of these coders have an inherent compression advantage over the context model of the present invention.

In another embodiment, both a binary entropy coder and a fast m-ary coder are used. The fast m-ary coder may be a Huffman coder.

Coding and Decoding Process of the Present Invention The flowcharts of FIGS. 18 to 20 show examples of the coding process and the decoding process of the present invention. Processing logic may be implemented by software and / or hardware.

FIG. 18 shows an example of the encoding process of the present invention. In FIG. 18, at the beginning of the encoding process, processing logic obtains one tile of input data (processing block 1201).

Processing logic then determines whether binary encoding must be performed (processing block 120).
2). If binary coding is to be performed, the process proceeds to processing block 1211 where processing logic performs Gray coding on the input data and models each bit of each coefficient with a binary context model (processing block 1212). ). Processing proceeds to processing block 1208.

[0210] If binary encoding need not be performed, the process proceeds to processing block 1203 where processing logic filters the data reversibly. After applying the reversible filter, processing logic determines whether another level of decomposition is required (processing block 1204). If another level of decomposition is required, processing logic filters the LL coefficients reversibly (processing block 1205), and processing returns to processing block 1204 to make the determination again. If another level of decomposition is not required, the process proceeds to processing block 1206, where processing logic converts the coefficients to a sign-magnitude format. Processing logic then models each bit of each coefficient with a horizontal context model (processing block 1).
207), the process proceeds to processing block 1208.

[0211] At processing block 1208, processing logic encodes each bit of each coefficient. Then, the processing block transmits or stores each encoded data (processing block 1).
209).

Next, the processing block determines whether any other tile is used in the image (processing block 121).
0). If there are other tiles in the image, processing logic loops back to processing block 1201 and the process repeats; otherwise, the process ends.

FIG. 19 shows an example of the decoding process of the present invention. In FIG. 19, the process first obtains encoded data for one tile (processing block 1301). Next, processing logic entropy decodes the encoded data (processing block 1302). Processing logic then determines whether the data must be binary decoded (processing block 1203). If the data has to be binary decoded bit by bit, the process proceeds to processing block 1311 where processing logic models each bit of each coefficient with a binary context model, and performs inverse Gray coding on the data. Is applied (processing block 1312). After this inverse Gray encoding, the process proceeds to processing block 1309
Proceed to.

If binary decoding does not need to be performed, the process proceeds to processing block 1304 where processing logic models each bit of each coefficient with a horizontal context model. Processing logic then converts each coefficient into a form suitable for filtering (processing block 13).
05), applying a reversible filter to the coefficients (processing block 1)
306).

After applying the reversible filter, processing logic determines if there is another level of decomposition (processing block 13).
07). If there is another level of decomposition, the process proceeds to processing block 1308, processing logic filters the coefficients reversibly, and the process proceeds to processing block 1307.
Loop back to If another level of decomposition is not required, the process proceeds to processing block 1309 where the reconstructed data is transmitted or stored.

Next, processing logic determines if there are any more tiles in the image (processing block 1310). If there are other tiles in the image, the process goes to processing block 13
The process is repeated by looping back to 01, but if there are no more tiles, the process ends.

FIG. 20 illustrates an example of a process for bit modeling according to the present invention. In FIG. 20, the bit modeling process first sets the coefficient variable C to the first coefficient (processing block 1401). Next, | c |>
A determination of 2 ^S is made (processing block 1402). If the answer is yes, processing proceeds to processing block 1403, where processing logic encodes bit S of coefficient C using the tail bit model, and processing block 1408.
The process proceeds to. The tail bit model may be a static (non-adaptive) model. | C | when is not greater than 2 ^S, the process proceeds to process block 1404, processing logic applies the template to the head bit (0 and the first "1" bit of the head). After applying the template, processing logic encodes bits S of coefficient C (processing block 1405). A possible template is shown in FIG.

Next, it is determined whether the bit S of the coefficient C is on (processing block 1406). If bit S of coefficient C is not on, processing continues at processing block 1408. on the other hand,
If the bit S of the coefficient C is on, the processing proceeds to processing block 1
Proceeding to 407, processing logic encodes the sign bit. Processing then proceeds to processing block 1408.

At processing block 1408, it is determined whether coefficient C is the last coefficient. If coefficient C is not the last coefficient, processing proceeds to processing block 1409, coefficient variable C is set to the next coefficient, and processing continues at processing block 1402. On the other hand, if coefficient C is the last coefficient, processing continues at processing block 1410 where it is determined whether S is the last bit plane. If S is not the last bit plane, the bit plane variable S is decremented by 1 (processing block 1411) and processing continues from processing block 1401. If S is the last bit plane, the process ends.

TS Transform Design In one embodiment, the present invention computes TS transforms in place in a buffer memory. In doing this, no extra lines of memory and extra time are needed to rearrange the calculated values. As mentioned above about TS conversion,
The present invention applies an overlap transform to the critically sampled one. In another embodiment, a TT transform is used.

FIGS. 24A to 24C illustrate the memory manipulation method employed by the present invention during the computation of the transform of the present invention. FIG. 24A shows an initial state of the memory. In FIG. 24A, the first row of the memory contains the previous value (n
-1) smoothing ("S") coefficient and detail ("D") coefficient (calculated), current value (n) smoothing ("S") coefficient and partially completed detail coefficient ("B") With the values (X2n + 2, X2n + 2, X2n + 4, X2n + 5) of the four input samples ("X"). The intermediate result of the conversion calculation is shown in the same memory row in FIG. The change in this row is that X2n + 2 and X2n in the fifth storage element and the sixth storage element.
Note that the value of +3 has only been replaced by the values of Sn + 1 and Bn + 1. Thus, the present invention saves memory space by replacing stored values that are no longer needed with the results generated during the transform calculation. FIG. 24C shows the same memory row after the conversion is completed and the detailed output Dn is generated. The only difference from FIG. 24B is that the partially completed detail coefficient Bn has been replaced by the detail output Dn.

After the detail output has been calculated for n, the transform calculation process continues down the row to calculate the detail output Dn + 1.

The following code example may be used to perform the conversion. Note that horizontal codes for forward and reverse transforms are included.

In the following, the variable soo indicates the Sn-1 value, the variable oso indicates the Sn value, and the variable os indicates Sn + 1.

A specific example of a code (C language) for an example of the forward TS conversion is as follows. / * TSFoword_1 () * / void TSForward_1 (long * x, int width) {long * start = x; long * ox = x + 2; long soo; long oso; long oos; oso = (* x + * (x + 1)) >>1; oos = (* ox + * (ox + 1)) >>1; soo = oos; * (x + 1) = * x-* (x + 1); * x = oso; while ((ox + 2) -start <width) {x = ox; os + = 2; soo = oso; oso = oos; oos = (* ox + * (ox + 1)) >>1; * (x + 1) = * x-* (x + 1) + ((oos-soo + 2) >>2); * x = oso;} x = ox; soo = oso; oso = oos; oos-soo; * (x + 1) = * x-* (x + 1); * x = oso;}.

A specific example of a code (C language) for an example of the inverse TS conversion is as follows. / * TSReverse_1 () * / void TSReverse_1 (long * x, int width) {long * start = x; long * d = x + 1; long ns = * (x + 1); long P; while (x + 2 -start <width) {p = * d-((* (x + 2) -ns + 2) >>2); ns = * x; * d = * x- (P >>1); * x + = ((p + 1) >>1); x + = 2; d = x + 1;} p = * d; * d = * x- (p >>1); * x = ((p + 1)>>1);

Although only a one-dimensional example is shown, the invention can be used in multiple dimensions and multiple levels. Note that this approach can be used for any overlap transformation that replaces values that are no longer needed for other calculations with a partial or final result one-to-one.

FIG. 25 shows a two-dimensional representation of a memory buffer for three levels. In FIG. 25, each block 180
The memory locations 1-1804 contain coefficient values. That is, each of the blocks 1801 to 1804 is an 8 × 8 block of coefficient values.

The coefficients are arranged at intervals of a natural power of 2. Given the level and the S or D offset,
Any coefficient can be accessed. Therefore, access can be made by selecting specific levels and horizontal and vertical frequencies. The buffers may be accessed in raster order.

Unit Buffer Configuration In one embodiment of the present invention, a single buffer supports transform blocks, context model blocks, and coding blocks of a compression system. This buffer is
A two-dimensional scrolling memory buffer that allows efficient access to coefficients and requires no additional memory. Each line of the buffer is accessed via a pointer stored in a line access buffer. FIGS. 23A and 23B show the configuration of this scrolling buffer. A line access buffer 1601 contains pointers to each line of the buffer 1602.

Scrolling is performed by rearranging the pointers stored in the line access buffer. Examples are shown in FIGS. 23A and 23B. FIG. 23A shows the initial state of the buffer. FIG.
Referring to FIG. 3 (B), after lines A, B, and C have been removed from the buffer and replaced by lines G, H, and I, respectively, the line access is performed to impart the effect of a scrolling buffer to the buffer. The pointers in the buffer are changed so that the first pointer points to line D in the buffer, the second pointer points to line E, and the third pointer points to line F. Next, line G,
Pointers to H and I occupy the last three positions in the line access buffer. Note that the present invention is not limited to having a buffer of 6 lines, and this is used only as an example. Buffers with a higher number of lines are commonly used and will be known to those skilled in the art. Then, line access
Access through the buffer has the appearance of scrolling the unit buffer without physically moving the storage. This allows you to sacrifice speed,
Uses minimal memory.

In the present invention, such a unit
The use of a buffer supports the application of the overlap transform to the entire image, while always storing only one band of the image in memory. To achieve this, the wavelet transform is applied to only as many lines of the image as necessary to completely compute the set of wavelet coefficients that make up the at least one band wavelet unit. In such a case, the computed set of wavelet coefficients can be modeled, entropy coded, and removed from the appropriate portion of the wavelet unit buffer. The incompletely calculated wavelet coefficients remain so as to complete the calculation in the next iteration. Then, by rearranging the line pointers, the wavelet unit buffer can be scrolled, and other data can be put in the empty portion of the wavelet unit buffer. Thus, the wavelet coefficients in the middle of the calculation can be completely calculated.

As an example, consider that a high-pass filter applies an overlap transform that depends on the current coefficient and the next low-pass filter coefficient. In this example, only two levels of decomposition would be applied to the image data, which implies that one wavelet unit is four elements long.

Wavelets of at least one band
In order to fully calculate the set of wavelet coefficients that make up the unit, the height of the wavelet unit buffer is at least 8 lines or 2 wavelet units.

When applying a wavelet transform to a two-dimensional wavelet unit buffer, first, a one-dimensional wavelet transform is applied to each row of the buffer. Then, a one-dimensional wavelet transform is applied to each column of the buffer.

When applying the one-dimensional wavelet transform to each column of the wavelet unit buffer, the calculation of the high-pass filter is performed for the last element of each column that depends on the image element not stored in the unit buffer. It can only be partially completed. This is shown in FIG.

When the second-level wavelet decomposition is performed, the calculation of the high-pass filter for the last element of each column can be only partially completed. This is shown in FIG.

In one embodiment, when many decomposition levels are used, the SS coefficient (1SS in FIG. 55 for the second decomposition level, 2S in FIG. 56 for the third level) is used.
Only in S) will the wavelet transform be applied. In such a case, the row and column locations of the unit buffer will be skipped to ensure reading and writing of the appropriate entry in the buffer.

In this example, the upper half of the buffer contains a set of wavelet coefficients for which calculation has been completed, and this wavelet coefficient set constitutes a one-band wavelet unit, which is modeled, entropy-coded, And can be removed from the buffer.

When the upper half of the buffer is free, the buffer can be sucrose by half the height. Here, the next four lines of the image can be read into the buffer.
1 for each new line stored in the buffer
A dimensional wavelet transform can be applied. In the column direction of the buffer, the partially calculated coefficients can be calculated completely, but again, the last element of each column can only be calculated partially.

The same is done for the second level wavelet decomposition. Again, the upper half of the buffer contains the computed wavelet coefficients, at which point the process repeats until there are no more lines of the image to process.

Lines in the line access buffer
Rearranging the pointers can be done in various ways. One way is to create a new line access buffer and copy the pointer from the old line access buffer into it. To scroll modulo the height of the wavelet unit buffer, the pointer stored in element i of the old line access buffer would be copied to the index (i + number of lines).

Note that the three stages of the compression system are:
It should be noted that the coefficients are generally ordered differently as the data in the buffer is performed before the data leaves the buffer. When raster-ordered data operations are performed, the scroll buffer of the present invention will suffice with minimal memory.

Software (and / or hardware)
Manages the line access buffer and operates the pointer. This software also knows which data in the buffer has completed processing and can be moved out of the buffer.

Alignment Method In the present invention, the coefficient value is shifted to the left by an arbitrary amount. In one embodiment, the alignment is performed by a virtual alignment method. This hypothetical alignment method
Do not actually shift the coefficients. Instead, while processing the coefficients one bit plane at a time, the actual bit plane that needs to be aligned for a particular coefficient is calculated. Given the level of importance and the amount of shift to be applied to a particular coefficient, the present invention accesses the desired absolute bit plane for that coefficient if it is within the range of possible bit planes. That is, the desired absolute bit plane for a particular coefficient is given by the current importance level minus the amount of shift to be applied to that coefficient. The desired bit plane is considered valid when it is greater than or equal to the smallest valid absolute bit plane and less than or equal to the largest valid absolute bit plane.

Two alignment methods are common. The first of these methods is called the mean square error (MSE) method, which reduces or minimizes the MSE when comparing a full-frame reconstructed image with an original image. , The coefficients are aligned. FIG. 50 is an example of this alignment. See also FIG.

The second method is a pyramid-type alignment method, which provides good rate-distortion performance when the image is reconstructed to a pyramid-level size.
Here, the adjacent level coefficients do not have a common significance level, ie there is no overlap. The alignment on the left side of FIG. 51 represents a strictly pyramidal alignment for a three-level TS conversion. The right side of FIG. 51 represents a level 2 pyramid alignment.
(The strictly pyramid-shaped part of FIG.
May be called a pyramid-shaped alignment. )
In each case, the coefficients inside the level are aligned with respect to the MSE.

FIG. 52 shows a typical relationship between memory storage coefficients and one alignment.

According to the present invention, since there is no need to perform an actual shift operation, there is no restriction on the memory size. Furthermore, the present invention does not require extra memory and can easily implement any alignment method.

Histogram Compression The present invention may utilize histogram compression. In one embodiment, histogram compression is used before undergoing any transformation or binary processing. Histogram compression improves the compression ratio for some images. Such images are typically images in which some values of the dynamic range are not used for any pixels. In other words, there is a gap in the dynamic range of an image. For example, if an image takes only values 0 and 255 out of a total of 256 values, a new image having a one-to-one correspondence with the original image, but with a much smaller dynamic range can be generated. This image generation is achieved by defining an increasing function that maps an integer to the value of the image. For example, if the images are 0 and 255
When only the value of is used, 0 becomes 0 in the mapping and 1
To 255. In another embodiment, when the image has only even (or odd) pixels, the pixel value is
Remap to a value of 8.

After performing the histogram compression, the image data may be compressed by the reversible embedding wavelet of the present invention. Thus, histogram compression is used in the pre-processing mode. In one embodiment, the histogram is based on a Boolean histogram,
It's like keeping a list of whether values occur. First, all occurrences are recorded in ascending order. Next, each value is mapped in order from 0.

In one embodiment, guard pixel values are used to reduce the effects of errors. Small errors in the re-mapped values can result in large errors in the real values, as adjacent re-mapped pixel values may correspond to real pixel values separated by a large gap. By adding an extra value near the remapping value, the effects of such errors will be reduced.

The decoder is informed that the mapping has been used to reconstruct the original image. This mapping may be indicated in the header. This allows the decoder to create a similar table for post-processing. In one embodiment, the decoder is informed of the range by tile. In one embodiment, the present invention first signals that the mapping is to be performed, and then indicates the number of missing values (eg, 254 in the example above). The cost to inform whether histogram compression is used is only one bit. This bit will be followed by a table of all re-mapped values.

In one embodiment, to reduce the amount of notification when performing histogram compression on a tile-by-tile basis, one bit indicates whether the new Boolean histogram is the same as or different from the Boolean histogram used immediately before.
In such a case, the new boolean histogram is notified to the decoder only when it differs from the previous histogram (only). Even if the new boolean histogram is different from the previous one, there are usually similarities between them. More specifically, since the exclusive OR of the two histograms is more compressible by the entropy coder, it may be generated and notified to the decoder.

The histogram can be notified by transmitting the same number of bits as the dynamic range of the size (for example, 256 bits in the case of 8-bit depth).
The order of the bits corresponds to the pixel value. In this case, a certain bit is 1 when a corresponding value is used in an image. To reduce or minimize the cost of the header, this bit string is
rkov) Entropy coding may be performed under a context model.

In another embodiment, if the missing value is a majority, the values that occurred will be recorded sequentially, otherwise the missing values will be recorded sequentially.

In one embodiment, the binary scheme of the present invention is:
It will be used for compression of palletized images. The palette will be stored in the header. But,
Quantization for lossy decompression does not give reasonable results, since the palletized image will not be embedded. In another embodiment, the paletted image may be converted to a continuous tone (color or grayscale) image, and each component may be compressed in a conversion or binary manner. This allows for reasonable lossy compression.

Certain images are continuous tone images in which one designated color (or a small subset of the designated colors) has been used for a particular purpose. The special purpose color may be for notes. For example, a grayscale medical image may have computer-generated color characters to identify the image. Another special purpose color may indicate that the pixels in the overlay image are transparent and that the pixels in the lower image are displayed instead. Forbidden colors will be separated into separate component images. The continuous tone component and the special color component will then be compressed / expanded in a conversion or binary manner.

Although the conversion method and the binary method are often used for luminance data, other types of two-dimensional data such as an alpha channel for alpha mixing may be used.

Parser According to the present invention, a code stream can be parsed before being transmitted or decoded without decompression.
This parsing is performed by a parser that can abort the bitstream and convey only the amount of information needed for a particular quantization. To support this parser, markers and pointers determine the position of each bit plane of a coding unit in the bit stream.

The present invention provides device dependent quantization performed by parsing in an image compression system.
The use of markers in the compression system allows device-selective quantization after encoding. The output device reports its characteristics to the parser, which quantizes the encoded file for that particular device. This quantization is by omitting a part of the file. The use of the reversible wavelet transform allows lossless restoration of the image, or restoration of the image with various visually lossless distortions depending on the device.

According to the present invention, quantization can be performed after encoding. 27 and 28 are block diagrams of a compression system including a parser. 27 and 28, an original uncompressed image 2101 is input to a compression device 2102 of the present invention. Compression device 210
2 losslessly compresses the image 2101 into a compressed bit stream 2103,
03 and a marker is added.

A compressed bit stream 2103 is a parser 2
Input to 104, parser 2104 provides some portion of compressed bitstream 2103 as output. That part may be the entire compressed bit stream 2103 or only a part thereof. The requesting agent or device provides the device characteristics to the parser 2104 when the decompressed image is needed. In response, parser 2104 generates compressed bit stream 210
4. Select and send the appropriate part of 4. The parser 2104 performs the entropy coding /
No decoding is performed. In another embodiment, parser 2104 may perform some such functions.

The parser 2104 can provide encoded data for displaying an image on a monitor by selecting a low-resolution compression coefficient. For different requests, parser 2104 selects the compressed data to allow lossless decompression of the region of interest (ROI). In one embodiment, parser 2104, according to the request,
It sends out the bits needed to change from a preview image to a printer resolution image or a full size medical monitor image (possibly with 16 bit pixel depth).

The data provided by parser 2104 is output to a communication channel and / or storage device 2106. The decompression device 2107 accesses the data and decompresses the compressed data. The decompressed, that is, reconstructed data is output as a decompressed image 2108.

In FIG. 29, a bit plane in the 2HH frequency band is encoded using information from the 3HH frequency band. To show the bitplane more clearly,
FIG. 29 is rewritten in FIGS. 50 and 51. FIG.
When the coefficients are stored (MSE), the compression bit stream truncation is almost the same as the MSE rate-distortion optimal quantization. This truncation is indicated by the shaded marker in FIG. Looking at FIG. 30, this arrangement may be appropriate for a printer, but not enough for a monitor. As shown in FIG. 51, if the coefficients are stored "pyramidally", that is, if all bits of a certain frequency band are stored first, truncation of the bitstream provides images of various resolutions. .

With the skillful use of markers, both types of truncation will be possible and will produce a low resolution, low fidelity image. The shading changes in FIGS. 50 and 51 represent a truncation of the bitstream that would produce a low fidelity bitstream at the highest resolution. LH,
Further truncation of all HL and HH coefficients would reduce the resolution of the image.

In many image compression applications, an image is compressed only once, but may be decompressed many times. Unfortunately, most compression systems require that the amount of loss allowed and the appropriate quantization be determined at the time of encoding. Progressive systems provide a progressively more refined set of images, but generally do not allow lossless reconstruction, i.e., send out "differential images" encoded in a lossless manner independent of progressive buildup This provides a lossless reconstruction.

In the present invention, the encoder stores enough information to decompose the various coefficients into frequency and bit plane components. In one embodiment, a marker is inserted into the bitstream to indicate what is in the next entropy coded data unit. For example, the marker may indicate that the next entropy coded data unit contains HH frequency information for the third bitplane from the most significant.

If someone wants to examine an image on a monitor, they will require the information needed to generate a low resolution grayscale image. When the user wants to print the image, the information needed to generate a high resolution binary image will be required. Finally, if the user wants to perform a compression test, or do a statistical analysis of sensor noise or a medical diagnosis, a non-lossy version of the image will be required.

FIG. 31 is a block diagram relating to exchanges with a parser, a decoder, and an output device. In FIG. 31, a parser 2402 is connected to receive lossless compressed data with markers, as well as device characteristics of one or more output devices, such as the display module 2405 shown. Based on this device characteristic, parser 2402 selects the appropriate portion of the compressed data and
3 and the communication path 2403 transmits the data to the decompression device 24.
04. The decompressor 2404 decodes the data and outputs the decoded data to the display module 240
Give 5

The present invention provides improved support for data streams to the World Wide Web and other image servers. Some portions of the data stream can support low spatial resolution, high pixel depth images for monitors. Another part can support high spatial resolution, low pixel depth printers. The entire data stream provides a lossless transmission. Since these three uses are supported by the same compressed data, there is no need to send any extra data if the browser requests a monitor image, a print image, and a lossless image in order.
The transmitted monitor image information required for the print image can be reused for the print image. The transmitted information for monitor and print images can be reused for lossless images. The present invention reduces transmission time (transmission cost) for browsing and minimizes the amount of data that must be stored on the server.

In the system of the present invention, the image is compressed only once, but various markers are stored to indicate what the data is. Thereafter, the World Wide Web (WEB) server receives the request for display,
Will provide the necessary coefficients. The WEB server does not need to do any compression or decompression, but only selects the appropriate part of the bitstream to send.

Quantization with such a parsing system results in a substantial increase in bandwidth, even without a high degree of lossless compression with lossless wavelets and context models. This parsing system can also be used to select high quality regions of interest.

FIG. 32 shows a quantization selection device. In one embodiment, the selection device is configured by software to determine an appropriate quantization profile for the various devices. The image is transformed and quantized by discarding bitplanes of various frequency bands. Next, an inverse wavelet transform is performed. The reconstructed image is processed in some manner consistent with the display. For high resolution images displayed on a monitor, the process will be some kind of scaling. In the case of a printer, it may be some sort of thresholding or dithering. The same process is applied to the original image and compared to the compressed image. Although the mean square error was used as an example, any visual difference criterion may be used. A bit plane is selected and quantized using errors caused by quantization of various bit planes so that distortion due to bit rate saving is minimized. This process will continue until the desired bit rate or distortion is reached. Once the representative quantization for the various image processing operations has been determined, there is no need to simulate the quantization and the representative values will be used.

Of course, for simple image processing operations such as scaling, the effects of quantization in various frequency bands can be analytically determined. For other operations such as dithering and contrast masking, it is much easier to find a near optimal quantization by simulation.

In FIG. 32, a code stream 2500
Are the extension 2501 with quantization and the lossless extension 250
3 is given. Image processing or distortion models 2502 and 2504 are applied to the expansion result. The output is an image and a difference model such as MSE or HV5 difference model 2505 is applied. Based on the result of the difference determination, the alignment is adjusted (2504), and thus the quantization is adjusted.

To facilitate parsing, the present invention utilizes a series of header cues. In one embodiment, the codestream structure of the present invention includes a main header with one or more tag values. The tags in the main header signal information such as the number of components used for all tiles in the codestream, subsampling and alignment. In one embodiment, each tile in the codestream is preceded by its header. The information in the tile header applies only to that tile and may override the information in the main header.

[0279] The headers each include one or more tags.
In one embodiment, there are no inline markers. header·
The tag indicates the amount of compressed data from a known point to where the user resets the coder. In one embodiment, every tag is a multiple of 16 bits. Therefore,
Both the main header and the tile header are multiples of 16 bits. Note that any tag may be a multiple of a number other than 16 bits. In each tile data segment, an appropriate number of zeros is inserted so that the number of bits is a multiple of sixteen.

[0280] In one embodiment, each tile header may indicate its tile size. In another embodiment, each tile may indicate where the next tile begins. It should be noted that when backtracking of the code stream is possible, encoding may be simplified by inserting all such information into the main header. The parser can use information about the codestream to perform its quantization.

In one embodiment, the tile header may indicate whether the tile was encoded in a wavelet or binary fashion. The importance level indicator associates the data of the tile with the importance level. An importance level locator indicates possible truncation locations. For example, when the same distortion is desired for each tile, the parser can truncate the codestream at the appropriate location, knowing which importance level is equal to the desired distortion level. In one embodiment, instead of having the same number of bits, each tile has approximately the same distortion.

The present invention has an importance level locator tag so that it can have multiple tiles and an indication of where each tile should end.

Tags and Pointers Parsing markers and other information used for decoding or parsing may be included in the tags. In one embodiment, the header provides control information by tags according to the following rules.

[0284] Tags may be fixed size or variable size. Depending on the number of components, the number of tiles, the number of levels, or the number of resets or desired information, the tag may vary in length.

If the image is parsed and quantized,
Those tags are changed to represent the new image characteristics.

At the reset point in the data stream, 0 is inserted so that the number of bits is a multiple of eight. The entropy coder can be reset at some point in the codestream, but that point is determined at the time of encoding (but it can only be done at the end of one importance level of encoding). . This reset means that all state information (context and probability) in the entropy coder is returned to a known initial state. Next, the code stream is padded with zeros up to the next multiple of 8 bits.

The parser uses only codestream tags as clues when quantizing an image. In one embodiment, the tags of tile length, component length, reset, bit versus importance level, and importance level locator are used for the quantization process.

After the image has been quantized by the parser, all of its tags are modified to reflect the new codestream. This usually affects the image, tile size, number of components, span of components, all lengths and pointers, etc. Also included is an information tag that describes how the image was quantized.

Table 3 is a list of all tags in one embodiment of the present invention. The description and terminology are often different from JPEG, but the same markers and identifiers are used where possible. Every image has at least two headers, a main header at the beginning of the image and a tile header at the beginning of each tile. Three types of tags (each code stream contains at least one tile):
That is, a separation tag, a function tag, and an information tag are also used. The delimiter tag is used for framing the header and data. The function tag is used to describe the encoding function used. Information tags provide optional information about the data.

[0290]

[Table 3]

[0291] Note that "x" means that the tag is not used in the header. Either the TLM tag in the header or the TLT tag in each tile is required,
Both are not required. Component pointers are only needed when there are more than one component.

FIG. 33 shows the arrangement of delimiter tags in the code stream of the present invention. Each codestream has only one SOI tag, one SOC tag, one EOI tag (and at least one tile). Each tile is 1
It has one SOT tag and one SOS tag. Each delimiter tag is 16 bits and does not include length information.

The SOI tag indicates the start of a JPEG file and has a 16-bit JPEG magic number (magi
c number).

The SOC tag indicates the start of the file,
It comes right after the OI tag. The SOI tag and the SOC tag as a whole have 16 bits forming a unique number.

The SOT tag indicates the start of a tile. There are at least one tile in the codestream. SOT
Works as a check to ensure that the stream is still synchronized.

[0296] The SOS tag indicates the start of "scan", and is followed by the actual image data of the tile. The SOS indicates the end of the tile header, and there must be at least one SOS in the CREW codestream. The data between the SOS and the next SOT or EOI (end of image) is a multiple of 16 bits and the codestream is padded with zeros if necessary.

The EOI tag indicates the end of the image. EOI
Works as a check to ensure that the stream is still synchronized. At least one in the codestream
There are two EOIs.

The function tag describes a function used to encode the entire tile or image. Some of these tags are used in the main header, but can be overridden using the same tag with different values during the encoding of individual tiles. The SIZ tag is the width and height of the image grid, the width and height of the tile, the number of components, the color space conversion (if necessary), the size of each component (pixel depth), and how the components fill the reference grid Is defined. This tag appears only in the main header, not in the tile header. Each tile is
All of its ingredients provide the same properties. Since many of the parameters defined here are also used for other tags, the SIZ tag must immediately follow the SOC tag. The length of this tag is stored in the first field after SIZ, Lsiz, but depends on the number of components.
FIG. 34 shows the syntax of the image and tile size for the SIZ tag.

The following is a description list of the size and value of each element.

SIZ: marker.

Lsiz: Tag length in bytes, not including marker (must be even).

Xsiz: width of the image reference grid (same as the image width in the case of one-component images or images having color components by common sub-sampling).

Ysiz: Height of image reference grid (same as image height for one-component images or images having color components by common subsampling).

XTsiz: 1 tile image reference grid width.
The tile must be wide enough to hold one specimen of all components. The number of tiles in the image width is

[0305]

[Outside 3]

Is equal to

YTsiz: Height of one tile image reference grid. The tile must be tall enough to hold one specimen of all components. The number of tiles in the image height is

[0308]

[Outside 4]

Is equal to

Csiz: number of components in the image.

CSsiz: Type of color space conversion (when necessary). This tag is not comprehensive. (Many multi-component spatial transformations cannot be specified here; they need to be specified elsewhere, not within the file format of the present invention). Table 4 shows values for color space conversion.

[0312]

[Table 4]

The subsampling described in this tag is applied to an image for which the highest resolution cannot be used for each component. The system of the present invention has another way to reduce the size of less important components when the highest resolution is available.

Ssisi: Accuracy (pixel depth) of the i-th component. This parameter, XRsiz and YRsiz, is repeated for all components.

XRsisi: X-dimensional size of the i-th component. For example, the number 2 means that the component contributes to two horizontal reference grid points. This parameter and Ssi
z, YRsiz is repeated for all components.

YRsisi: Y-dimensional size of the i-th component. For example, the number 2 means that the component contributes to two vertical reference grid points. This parameter and Xsi
z, XRsiz are repeated for all components.

Res: Zero padding byte placed last when needed.

[0318]

[Table 5]

The COD tag describes an encoding method such as a binary method or a wavelet method used for an image or a tile, a transform filter, and an entropy coder. This tag is included in the main header and can also be used for tile headers. The length of this tag depends on the number of components. FIG. 35 shows the coding scheme syntax. Table 6 shows the sizes and values for the encoding scheme.

[0320]

[Table 6]

COD: marker.

Lcod: tag length in bytes, excluding marker (must be even).

Ccodi: Coding method for each component.

Res: Zero padding byte placed last when needed.

[0325]

[Table 7]

For each component, the ALG tag describes the alignment of pyramid level numbers and coefficients. ALG is used for the main header and can also be used for the tile header. The length of this tag depends on the number of components and, in some cases, on the number of levels. FIG. 36 shows an example of the component alignment syntax of the present invention. In FIG. 36, the following components are included.

ALG: This marker indicates the size and value of the component alignment parameter.

[0328]

[Table 8]

Lag: tag length in bytes (not including markers) (even number).

Palgi: Number of pyramid decomposition levels of the i-th component. This parameter and Aalg, and possibly S
alg is also repeated as one record for each component.

Aslgi: Alignment of the i-th component.
This table entry describes the alignment of the coefficients and is repeated for every component. A in Table 9
Represents the value of the alg parameter.

[0332]

[Table 9]

[0333] Palg and, in some cases, Salg are repeated as one record for each component.

Talgi: Table 10 shows a method for selecting tail information.

[0335]

[Table 10]

Salgij: Alignment value of the j-th sub-block of the ith component, and Aalg for the component
Used only when the value of i is "custom alignment". This number can be 8 or 16 bits depending on which custom alignment is chosen,
Also, for that component, it is repeated in order for every frequency band of the image. (For the binary method,
Salgij is the alignment value at the ith pyramid level.) When Salgij is used, Salgij
And Aalg and Parg are repeated as one record for each component.

Res: Zero padding byte placed last when needed.

[0338] The TLM tag describes the length of every tile in the image. The length of each tile is the first of the SOT tags
Length measured from byte to first byte of next SOT tag (for next tile) or EOI (end of image). In other words, this length is a list of pointers to tiles or a daisy chain.

The codestream contains either a single TLM tag or a TTL tag for each tile, but not both. When a TLM tag is used in the main header, no TLT tag is used. Conversely, each tile is TL
When ending with a T tag, no TLM tag is used.
The value of each tile length in the TLM header is the same value that would be used for the corresponding TLT tag if TLM were not used. The length of the TLM tag depends on the number of tiles in the image. FIG. 37 shows an example of the syntax of the tile length / main header.

TLM: Table 11 shows the sizes and values of the tile length / main header parameters.

[0341]

[Table 11]

Ltlm: Tag length in bytes, excluding markers (even number).

Ptlmi: Length in bytes from the SOT marker of the i-th tile to the next SOT (or EOI) marker. This is repeated for every tile in the image.

The TLT tag describes the length of the current tile. This length is the length measured from the first byte of the SOT tag to the first byte of the SOT tag of the next tile (or to the EOI). It is. In other words, TLT is a pointer to the next tile. FIG. 38 shows an example of the TLT syntax.

Either a TLM tag or a TLT tag is required, but not both. TLT tags are required for all tile headers when used, and TLM tags are not used. These tile length values are the same for both markers.

TLT: Table 12 shows the sizes and values of the tile length and tile header parameters.

[0347]

[Table 12]

Ltlt: Tag length in bytes (not including markers) (even number).

Ptlt: Length (number of bytes) from the SOT marker of the tile to the next SOT marker (or EOI marker).

[0350] The CPT tag is obtained from the first byte of SOT.
Points to the first byte of all components except the first component in the tile. The component coded data is arranged in a non-interleaved manner in each tile and starts from an 8-bit boundary. At this point the entropy coder is reset.

This tag is used for the tile header of every tile when the image contains more than one component. The size of the variable-length tag depends on the number of components in the image.
FIG. 39 shows an example of the syntax of the component pointer.

CPT: Table 13 shows the sizes and values of the parameters of the component pointer.

[0353]

[Table 13]

Lcpt: Tag length in bytes (not including markers) (even number).

Pcpti: The number of bytes from the SOT tag of the current tile to the start of the next component. Since the data of the first component starts immediately after the SOS tag, the number of Pcpt values is smaller than the number of components. New component data starts on an 8-bit boundary.

The IRS tag is the SOT of the current tile.
The first byte of the tag indicates a reset in the data.
These resets are found on 8-bit boundaries after the end of the significance level at which encoding is completed. The component at which a reset occurs can be determined by the relationship between the CPT tag value and the reset pointer. The length of this tag depends on the number of resets used by the decoder. An example of the importance level reset syntax is shown in FIG.

IRS: Table 14 shows the size and value of the importance level reset parameter.

[0358]

[Table 14]

Lirs: Tag length in bytes (not including markers) (even number).

[0360] Iirs ^i: current importance level number in the i-th reset. This Irs tag and the corresponding Pi
The rs tag forms a kind of record, which is repeated at each reset. These records are in order from the highest importance level with reset to the lowest importance level with reset, followed by the next component importance level, and so on to the last component. .

Pirs ⁱ : The number of bytes from the SOT tag of the current tile to the i-th reset byte. This Pi
The rs tag and the Iirs tag form a kind of record,
This is repeated for each reset. These records must be in order from the smallest pointer to the largest. That is, these records point to each reset byte in the order in which they appear in the codestream (smaller numbers indicate physically occurring bytes first).

[0362] Certain information tags are included solely for informational purposes. These information tags are not needed for the decoder, but will help the parser.

For example, the VER tag describes a major version number and a minor version number. This tag is used for the main header. This tag, although specified, does not imply a functional level required for image decoding. In fact, its purpose is to allow any decoder and parser to decode and parse any version of the codestream of the present invention. FIG. 41 shows an example of the syntax of the version number of the present invention.

VER: Table 15 shows the size and value of the version number parameter.

[0365]

[Table 15]

Lver: Table length in bytes (not including markers) (even number).

Vver: major version number.

Rver: minor version number.

A BVI tag associates a number of bits with an importance level based on the image width. This optional tag is used for the main header. The size of this variable length tag depends on the number of importance levels enumerated by the encoder. An example of bit versus importance level syntax is shown in FIG.

BVI: Table 16 shows the size and value of the tile length main header parameter.

[0371]

[Table 16]

Lbvi: Tag length in bits (not including the marker) (even number).

Cbvi ⁱ : This indicates which component data is to be described. This Cbvi parameter is Ib
Together with vi and Pbvi form a record, which is repeated for all components and importance levels described. The order must be such that all importance level descriptions for the first component, all importance level descriptions for the next component, and so on.

Ibvi ⁱ : In the current component, Pb
Importance level number encoded for the number of bytes in vii. This number (s) is selected during encoding to convey the point of interest of the rate-distortion curve. This I
The bvi parameter forms a record with Cbvi and Pbvi, which is repeated for all components and importance levels described.

Pbvi ⁱ : Number of bytes in the encoded file containing the main header and tile header, and all data related to the number of importance levels in Ibvii. This Pbv
The i-parameter together with Cbvi and Ibvi form a record, which is repeated for all components and importance levels described.

Res: Zero padding byte placed last when needed.

The ILL tag describes a pointer to a code stream corresponding to the end of the importance level of the encoded data. An ILL tag is similar to an IRS tag, but points to data without resets or insertion of bits on 8-bit boundaries. This tag allows the parser to find and cut off tiles with approximately the same distortion on an image width basis. This tag is optional and is used only in the tile header. The length of this tag depends on the number of importance levels counted. An example of the importance level locator syntax is shown in FIG.

ILL: Marker. Table 17 shows the size and value of the importance level locator parameters.

[0379]

[Table 17]

Lill: The length of the tag in bytes, not including the marker (even number).

Iill ⁱ : Importance level number encoded for the number of bytes in Pilli. Each of these numbers is selected during encoding to convey the point of interest of the rate-distortion curve. This Iill number forms one record with the Pil parameter, which is repeated in order from the highest importance level to the lowest importance level for the fastest component, and from the highest importance level to the lowest importance level for the later components. A similar record follows up to the gender level.

[0382] Pil ⁱ : from the first byte of the SOT of the current tile, Iil in the encoded data of the tile.
points to the byte where the importance level of li is completed. This number of pills together with the ill parameter forms a record, which is repeated from the highest importance level to the lowest importance level for the fastest component, and from the highest importance level to the lowest importance level for the later components. A similar record follows up to the critical level.

[0383] The RXY tag defines the X resolution and the Y resolution of the image reference grid for the actual dimensions. This tag is used only for the main header. FIG. 44 shows an example of the syntax of the resolution (pixels / unit).

RXY: Table 18 shows the resolution (pixels / unit)
Indicates the size and value of the parameter for specifying.

[0385]

[Table 18]

Lrxy: Tag length in bytes, excluding markers (even number).

Xrxy: Reference grid pixel number per unit.

Yrxy: Number of reference grid lines per unit.

RXrxy: X-dimensional unit. Therefore,
The horizontal resolution is Xrxy grid pixels / 10 (RXr
xy-128) meters.

RYrxy: Y-dimensional unit. Therefore,
The vertical resolution is Yrxy grid lines / 10 (RYr
xy-128) meters.

The CMT tag allows for unstructured data in the header. This tag can be used for both the main header and the tile header. The length of this tag depends on the length of the comment. FIG. 45 shows an example of the syntax of the comment.

CMT: Table 19 shows the size and value of the comment parameter.

[0393]

[Table 19]

Lcmt: tag length in bytes, excluding markers (even number).

Rcmt: Tag registration (re
gistration) value. Table 20 shows the values of the registration parameters.

[0396]

[Table 20]

Ccmti: byte of unstructured data. It is repeated arbitrarily.

Res: 0 placed at the end when necessary
Filling bytes.

The QCS tag describes how much the quantized code data has been quantized. When the quantization is performed by the parser or encoder, this tag helps the encoder roughly determine how much to encode in terms of importance level. This tag is optional and is used only for the tile header. FIG. 46 shows an example of the syntax of the quantization code stream.

QCS: Table 21 shows the sizes and values of the parameters of the quantized code stream.

[0401]

[Table 21]

Lqcs: Tag length in bytes (not including the marker) (even number).

[0404] Cill ⁱ : Number of the current component. This C
The ill number forms a record with Iqcs, which is repeated in order from the highest importance level to the lowest importance level for the fastest component, and from the highest importance level to the lowest importance level for the later components. A similar record to identify follows.

[0404] Iqcsi: This is the level of importance in which at least a portion of the encoded data remains. All data remaining from that point until the next reset is already censored (quantized).

Res: Zero padding byte placed at the end if necessary.

In one embodiment, the present invention performs lossy reconstruction by rounding values to a set of predetermined integer values. For example, 0 and 31
Are quantized to 0, all coefficients between 32-63 are quantized to 32, and so on. FIG.
Indicates a typical distribution of coefficients when not quantized. Such quantization may occur if the lowest bit of each coefficient is not known. In another embodiment, the value in the middle of each bin may provide a more accurate value that represents the set of coefficients. For example, all coefficients between 64 and 127 are quantized to 95. When a value is quantized to a point, that point is called a reconstruction point.

The resulting distribution is distorted by the differences between the images. For example, curves 2701 and 270 in FIG.
Compare two.

In the present invention, the reconstruction points are selected based on their distribution. In one embodiment, the distribution is estimated,
A reconstruction point is selected based on the estimated distribution. The estimated distribution is generated based on already known data. Prior to collecting the data, a default reconstruction point will be used. Thus, the present invention provides an adaptive lossy reconstruction method. Further, the present invention is a non-iterative method for improving coefficient reconstruction. To compensate for the non-uniform use of the range due to differences in distribution, the present invention defines:

[0409]

(Equation 16)

Here, 2 ^S is the sample variance measured by the decoder based on the available data, and Q is the quantization notified to the decoder. Next, it corrects the non-zero coefficients by moving them away from zero.

[0411]

[Equation 17]

[0412] Here, i is an arbitrary integer.

In one embodiment, after all decoding is complete, all non-zero coefficients are adjusted to a certain reconstruction level. To make this adjustment, it is necessary to read, possibly modify, and write each coefficient.

In another embodiment, when each bit plane of each coefficient is processed, if the coefficient is non-zero, the appropriate reconstructed value of the coefficient is stored. When decoding stops, all coefficients are set to their appropriate reconstruction values. By doing so, there is no need to go through a separate memory for setting the reconstruction level.

Color The present invention can be applied to color images (and data). The multi-component processing mechanism 101 shown in FIG. 1 performs processing required for color data. For example, the YUV color space has three components, a Y component, a U component, and a V component, and each component is separately encoded.

[0416] In one embodiment, the entropy coded data for each component is separated from the entropy coded data for the other components. In this example, there is no interleaving of the components. Separating the data by component, when combined with pyramid alignment, helps the decoder or parser to easily quantize different components separately.

In another embodiment, different components of entropy coded data are interleaved on a per frequency band or per importance level basis. This is beneficial when combined with MSE alignment, since a common truncation can be used to quantize the data of all components. For this interleaving scheme, the encoder needs to provide a relationship between frequency bands or importance levels of the different components. Since the frequency band or importance level would be a fairly large amount of encoded data, the parser or decoder could use the markers to quantize the components independently.

In still another embodiment, entropy encoded data of different components is interleaved on a pixel-by-pixel or coefficient-by-coefficient basis. This is beneficial when combined with the MSE alignment since a common truncation acts on all components. For pixel-by-pixel interleaving, the decoder and parser must use the same inter-component relationships defined in the encoder.

According to the present invention, subsampling can be performed by the same system.

In one embodiment, each component is stored separately. By using a decompressor and parser, when generating a lossy output image, only a selected one of the decomposition levels and components will be obtained from each of the separate component memories. For example, in the YUV color space, the full decomposition level would be obtained for the Y color component,
For the U and Y components, all decomposition levels other than the first decomposition level will be obtained. The resulting image combination is a 4: 1: 1 image. Note that different types of images can be obtained by using different parts of the data stored in the memory.

Many types of multi-component images can be processed. The image data is RGB (red, green,
Blue), CMY (cyan, magenta, yellow), CMYK (cyan, magenta, yellow, black) or CCIR601 YCrCb
May be. Multispectral image data (eg, remote sensing data) may also be used. For visual data such as RGB and CMY, US patent application Ser.
No. 436,662 (received May 8, 1995, "Metho
d and Apparatus for Reversible ColorCompressio
n "can be used.

Bit Extraction The present invention can compute a context model and encode bits to enhance bit extraction. Specifically, the context model for the head bit is based on information provided by neighboring pixels. often,
This context is zero, especially when performing lossy compression. Because of the approximate statistics of the headbit context,
The present invention provides a mechanism for maintaining a context for a head bit.

In one embodiment, the memory is cleared prior to encoding. The context remains until its parent, one of its neighbors, or the pixel of interest changes. When changed, the context memory is updated for all affected contexts. When using tail information, only the adjacent pixels and children are updated. When the head bit is on, the memory is updated only once per coefficient.

In one embodiment, each coefficient is one of the sign
4 bits for tail-on information, 8 for context
The bit is stored as a 32-bit integer consisting of 19 bits of the following coefficient. FIG. 48 shows an example of the coefficient.

In one embodiment, five different cases are generated using four bits of tail-on information.

In the case where the 4-bit value of the tail-on information is 0, the bit of the current bit plane of the absolute value bit of the current coefficient is encoded using the context bit. If the bit is 0, the process ends. If the bit is 1, the sign of the coefficient is encoded. Then, the first bit of the tail-on information is inverted, the north, northeast, west, south, east and four child contexts are updated, and the process ends.

In the case where the 4-bit value of the tail-on information is 1, the bits of the current bit plane of the absolute value bits of the current coefficient are encoded using a certain context for the case. The second bit of the tail-on information is inverted. The east of the current coefficient and the child context are updated. The process ends.

In the case where the 4-bit value of the tail-on information is 7, the bits of the current bit plane of the absolute value bits of the current coefficient are encoded using a certain context for the case. The third bit of the tail-on information is inverted. All contexts do not need to be updated. The process ends.

In the case where the 4-bit value of the tail-on information is 3, the bits of the current bit plane of the absolute value bits of the current coefficient are encoded using a certain context for the case. The fourth bit of the tail-on information is inverted. The east of the current coefficient and the child context are updated. The process ends.

In the case where the 4-bit value of the tail-on information is 15, the bits of the current bit plane of the absolute value bits of the current coefficient are encoded using a certain context for the case. None of the bits of the tail-on information need be inverted. The process ends.

FIG. 48 shows an example of coefficients according to the present invention. FIG.
, The coefficient 2801 is composed of a sign bit 2802, a tail neon information bit 2803 that follows, a context bit 2804 that follows, and a coefficient absolute value bit 2805 that follows. The foregoing process is illustrated in the flowchart of FIG.

By using this approach to update all contexts when a change occurs, context modeling works faster as long as the head bits are overwhelmingly zero. This is especially true for lossy coding.

Huffman Coding of Lossless Wavelet Coefficients In one embodiment, the invention encodes wavelet coefficients using Huffman coding. The alphabet for Huffman coding consists of two parts. The first part is equal to the length of the zero coefficient run, and the second part is the hash value of the non-zero terminator coefficient.
FIG. 53 shows the alphabet format, which consists of the number of 0 coefficients, in other words, 4 bits indicating the length of the run, and 4 bits indicating the hash value from 0 to 15 that follows.

The hash value is a value N, where N is 0
Is the integer part of the log base 2 of the absolute value of the non-terminator coefficient. In one embodiment, this hash value is the number of bits required to represent the value N. For example, N =-
In the case of 1,1, the hash value is 1. On the other hand, N =-
In the case of 3, -2, 2, and 3, the number of bits required to represent the value N is two. A similar correspondence is used for JPEG.

In such a situation, the maximum allowed length of a zero coefficient run is fifteen. When the run exceeds 15, a special token will be used to indicate that 0 is followed by 16 runs and a new run. In one such exception token, both the first four bits and the last four bits are all zeros. In one embodiment, all 16 tokens with the second 4 bits set to 0 are:
Used for exception cases. Thus, there are 256 8-bit Huffman tokens.

In one embodiment, a table is created for Huffman tokens. In one embodiment, the table is used for all images. In another embodiment, many tables are created and one particular table is chosen depending on the quantization. Each table is selected based on the number of bits to be quantized. That is, the table is selected depending on whether the number of bits to be quantized is 1 bit, 2 bits, 3 bits, or the like.
In another embodiment, the Huffman code is image specific and is stored / transmitted with the image.

To use the table, one Huffman token is generated. And this token is
It is sent to the table to be encoded.

A Huffman token specifies a run length of 0 and a hash value of a non-zero terminator symbol, but requires extra bits to uniquely identify the terminator symbol. One embodiment of the present invention provides these extra bits. After the Huffman token has been replaced with a Huffman codeword (e.g., obtained from a table or the like), an extra bit equal to the hash value of the terminator symbol is written. For example, in the case of -1, 1 extra 1 bit is written, but in the case of -3, -2, 2, 3 extra 2 bits are written. Thus, the present invention provides Huffman coding that is variable in size with extra bits that uniquely identifies the terminator symbol.

Note that another m-ary coder may be used. For example, one alphabet and the m-ary code may be used for the 0 coefficient and another alphabet and the m-ary code may be used for the hash value.

In one embodiment, a set of Huffman tables for each quantization level is pre-computed and used for most images. Compression will be monitored while using one table to select between the various tables. Based on the results of using that table, a switch to a table with more or less skew will be made.

All of the coefficients of the present invention are placed in a buffer. For each buffer, a decision will be made as to which table to use. Three bits and one header may be used to indicate which of the eight Huffman tables to use. Thus, by informing the header, a table selection will be made.

The order in which the coefficients are encoded is important. Note that in prior art coefficient coding, for example in JPEG, the coefficients are compressed in zigzag order. In the present invention, the coefficients cannot be placed in zigzag order because they are all in a buffer. If the zigzag order is understood as an order from low to high frequencies, it can be extended to compression (tree order) with embedded wavelets.

In one embodiment, the coefficients are encoded in a linear order for the entire buffer. FIG. 54 shows such an example.
Shown in Note that in this embodiment, the first block of smoothing coefficients is excluded.

In another embodiment, all blocks are encoded in raster order into higher frequency blocks than lower frequency blocks. Such an example is shown in FIG.
Shown in Due to memory constraints, all of one frequency path may not be completed before another frequency path begins. If limited by memory, another method is to encode one tree at a time. Starting from the root, all trees are coded horizontally. However, a route that is a smoothing coefficient is not included. This method is shown in FIG. FIG. 54 (C) shows the first tree, where one line is taken from the first set of sub-blocks, two lines are taken from the next set of sub-blocks, and the next set of sub-blocks is taken. More four lines are taken. Such an embodiment is possible because these lines are available before other lines become available.

An exception token may be stored to indicate that the rest of the tree consists of zero coefficients. This ensures that the same token representing 16 zeros is not used again and again.

In one embodiment, all importance levels are encoded by Huffman coding. In another embodiment, one or more groups of importance levels are encoded by Huffman coding. All importance levels may be encoded by Huffman coding for each separate group, or some importance levels may be encoded by Huffman coding, and the remaining importance levels may be encoded by horizontal context models and binary coding. It may be encoded by an entropy coder.

The encoding by Huffman encoding for one group of importance levels is performed as follows. If the significance level coefficient bits in the group are all head bits, the coefficient is Huffman coded as a 0 coefficient (perhaps as part of a run count). If the bits of the coefficient are all tail bits, then those bits are encoded as extra bits (probably terminating the run). No Huffman codeword is used. The bits of the coefficient are signed (in addition to the head or tail bits)
When ign) bits are included, both the Huffman codeword (possibly terminating the run) and the extra bits are encoded.

[0448] Huffman coding of multiple importance levels reduces execution costs. However, discontinuation of Huffman coded data in the middle causes deterioration of rate and distortion. Huffman coding of groups of importance levels gives
Censoring at the beginning / end of the group is possible so that the rate and distortion are good. In some applications, a limited number of required quantization points are known at the time of encoding. Importance levels without quantization points can be Huffman coded together with subsequent levels.

Applications The present invention can be used for many applications. Some of such applications are described below by way of example. Specifically, the present invention can be used for high-end applications with high resolution and large pixel depth and applications that do not tolerate artifacts. In accordance with the present invention, high-end applications can maintain the highest quality in a high-quality environment, while at the same time utilizing the same compressed data in applications where bandwidth, data storage or display capabilities are further limited. This is exactly the device independent representation commonly required in modern imaging applications such as web browsers.

The excellent lossless compression performance of the present invention for images with deep pixel depth (10 bits to 16 bits / pixel) is ideal for medical images. In addition to lossless compression, the present invention is a true lossless compressor without the many artifacts known in block-based compressors. The lossy artifacts resulting from utilizing the present invention tend to be along sharp edges and are often invisible due to the visual masking phenomenon of the human visual system.

The present invention can be used in applications related to the pre-press industry where images often have very high resolution and high pixel depth. The pyramid decomposition of the present invention facilitates a prepress operator performing image processing operations on a low resolution lossy version of an image (on a monitor). Once the operation is over, the same operation can be performed on the non-lossy version.

The present invention is also applicable to facsimile document applications in which the time required for transmission is likely to be too long without compression. According to the present invention, a very high-quality image can be output from a facsimile apparatus having various spatial resolutions and pixel resolutions.

The present invention can also be used in image archiving systems that require compression, especially to increase storage capacity. The device-independent output of the present invention advantageously allows an image archiving system to be accessed by systems with different bandwidth resources, memories and displays. The progressive transmission feature of the present invention is also beneficial for browsing. Finally, the present invention provides a desirable lossless compression for an output device of an image archiving system.

The hierarchical progressive nature of the lossless or high quality lossy data stream of the present invention makes the present invention ideal for the World Wide Web, especially where device independence, progressive transmission and high quality are essential. It is.

The present invention is also applicable to satellite images, especially satellite images that tend to have high pixel depth and high resolution. Further, the use of satellite images limits the bandwidth of the communication path.
Because the present invention is flexible and has progressive transmission characteristics, the present invention will allow humans to browse or preview images.

Applications where bandwidth is limited at a "fixed rate", such as ATM networks, require a means to reduce data when data overflows the available bandwidth. However, when there is sufficient bandwidth (or when the data is highly compressible), there should be no quality penalty. Similarly, "fixed-size" applications, such as memory-limited frame stores in computers and other imaging devices, also require a means to reduce data when memory is full. Again, there should be no penalty for images that can be losslessly compressed to a reasonable amount of memory.

The embedded code stream of the present invention serves both of these uses. The embedding operation unconditionally allows the codestream to be truncated or truncated for transmission or storage of the lossy image. If no truncation or truncation is needed, the image arrives lossless.

In sum, the present invention provides a single continuous tone image compression system. The system of the present invention is lossless and lossy for the same code stream and utilizes embedded quantization (included in the code stream).
The system of the present invention is also pyramidal, progressive, provides interpolation means, and is easy to implement. Thus, the present invention provides a flexible "device independent" compression system.

An integrated lossy and lossless compression system is very useful. State-of-the-art lossy and lossless compression can be performed on the same system, plus the same codestream. The system can determine whether to preserve or abort the lossless code of the image into a lossy version during encoding, during storage or transmission of the codestream, or during decoding.

The lossy compression provided by the present invention is achieved by embedded quantization. That is, the code stream includes quantization. The actual quantization (or visual significance) level may be a correlation point factor with the decoder or channel, and not necessarily with the encoder. If bandwidth, storage and display resources allow, the image is restored losslessly. If not,
The image is quantized as required for the most constrained resources.

The wavelet used in the present invention is of a pyramid type, and performs １ ／ decomposition of an image without a difference image. This is a very special hierarchical decomposition. The pyramidality of the present invention is ideal for applications requiring thumbnails for image browsing or display on low resolution devices.

The usage of embedding in the present invention is progressive, more specifically, in the order of bit planes, ie, in the order of MSB followed by lower bits.
Specifically, the present invention is progressive in the wavelet domain, but both the spatial domain and the wavelet domain may be progressively decomposed. For applications with spatial resolution but low pixel resolution, such as printers, the progressive ordering of the bits in the present invention is ideal. These features are obtained in the same code stream.

The present invention can be implemented relatively easily by software or hardware. Wavelet transformation is
It can be calculated with only four addition / subtraction operations for each high-pass and low-pass coefficient pair and only a few shifts. The embedding and encoding is performed by a simple "context model" and a binary or m-ary "entropy coder". This entropy coder can be implemented by a finite state machine, a parallel coder or a Huffman coder.

[0464]

As is apparent from the above description, according to the data compression system of the present invention, efficient lossy or non-lossy compression of image data and the like can be performed by using a conversion that provides good energy concentration. And a natural continuous tone image, a binary / noise-free / shallow pixel depth image, and
Images containing both types of data can be properly processed by the same system, and it can flexibly support image formats of various resolutions / quality,
Furthermore, the parser does not decompress the code data,
This has effects such as appropriate quantization of the code data stream according to the characteristics of the image output device.

[Brief description of the drawings]

FIG. 1 is a block diagram of a compression system according to an embodiment of the present invention.

FIG. 2 illustrates an example of a possible geometric relationship of a context model for each bit of each bit plane in a binary scheme.

FIG. 3 is a diagram illustrating an example of a possible geometric relationship of a context model for each bit of each bit plane in a binary scheme.

FIG. 4 illustrates a first level decomposition.

FIG. 5 illustrates a second level decomposition.

FIG. 6 illustrates a third level of decomposition.

FIG. 7 illustrates a fourth level decomposition.

FIG. 8 shows a parent-child relationship between two levels before and after.

FIG. 9 is a diagram showing an example of a wavelet decomposition process using only the TT transform.

FIG. 10 is a diagram showing an example of a wavelet decomposition process using a TT transform and an S transform.

FIG. 11 is an explanatory diagram of tiling of an image.

FIG. 12 is a diagram illustrating an example of a bit significance expression.

FIG. 13 is a diagram showing a coefficient size in the present invention.

FIG. 14 is a diagram showing an example of a frequency band multiplier used for coefficient alignment in the present invention.

FIG. 15 is a diagram illustrating an example of a configuration of a code stream.

FIG. 16 is a diagram showing an adjacency relationship between coefficients (or pixels).

FIG. 17 is a flowchart of a tail bit processing process.

FIG. 18 is a flowchart of an example of the encoding process of the present invention.

FIG. 19 is a flowchart of an example of the decoding process of the present invention.

FIG. 20 is a flowchart of the modeling process of the present invention.

FIG. 21 illustrates templates available for the modeling process.

FIG. 22 is a block diagram illustrating an example of a part of a TT conversion filter.

FIG. 23 is an explanatory diagram of a scroll buffer according to the present invention.

FIG. 24 is an explanatory diagram of a memory operation employed in the present invention.

FIG. 25 is a diagram showing a two-dimensional representation of a three-level memory buffer.

FIG. 26 is a diagram illustrating an example of a code stream according to the present invention.

FIG. 27 is a block diagram of a compression system with a parser.

FIG. 28 is a block diagram of a decompression system corresponding to the compression system of FIG. 27.

FIG. 29 is a diagram showing a context dependency relationship.

FIG. 30 is a diagram showing applications defined in terms of pixel depth and spatial resolution.

FIG. 31 is a block diagram illustrating an example of a parser, a decoder, and their interaction with an output device.

FIG. 32 is a block diagram illustrating an example of a quantization selection device.

FIG. 33 is a diagram illustrating an arrangement of delimiter tags in a code stream.

FIG. 34 is an explanatory diagram of an SIZ tag.

FIG. 35 is an explanatory diagram of a COD tag.

FIG. 36 is an explanatory diagram of an ALG tag.

FIG. 37 is an explanatory diagram of a TLM tag.

FIG. 38 is an explanatory diagram of a TLT tag.

FIG. 39 is an explanatory diagram of a CPT tag.

FIG. 40 is an explanatory diagram of an IRS tag.

FIG. 41 is an explanatory diagram of a VER tag.

FIG. 42 is an explanatory diagram of a BVI tag.

FIG. 43 is an explanatory diagram of an ILL tag.

FIG. 44 is an explanatory diagram of an RXY tag.

FIG. 45 is an explanatory diagram of a CMT tag.

FIG. 46 is an explanatory diagram of a QCS tag.

FIG. 47 is a graph showing an exemplary distribution for lossy reconstruction.

FIG. 48 is a diagram showing typical coefficients.

FIG. 49 is a flowchart of a tail information analysis process.

FIG. 50 is a diagram for explaining the MSE alignment method.

FIG. 51 is a diagram for explaining a pyramid alignment method.

FIG. 52 illustrates a typical relationship between memory storage coefficients and alignment.

FIG. 53 is a diagram illustrating an example of a codeword.

[Fig. 54] Fig. 54 is a diagram for describing a method for syntax analysis of coefficients by the Huffman coding method.

FIG. 55 is a diagram illustrating an intermediate format of a 2D memory when performing second-level wavelet decomposition using a unit buffer.

FIG. 56 is a diagram showing an intermediate type of a 2D memory when performing a third-level wavelet decomposition using a unit buffer.

[Explanation of symbols]

 Reference Signs List 101 input image data 102 lossless wavelet transform block 103 embedding ordered quantization block 104 Gray coding block 105 horizontal context model block 106 entropy coder 110 method selection mechanism 111 multi-component processing mechanism 1001 header 1002 coding unit 1003 LL Coefficient 1004 First bit plane 1005 Second bit plane 1006 Last bit plane 1501 Multiplier 1502 Adder 1503 Multiplier 1504 Multiplier 1505 Adder 1601 Line access buffer 1602 Buffer 1901 Header 1902 LL coefficient 1903 Entropy coded data 2101 Compressed No original image 2102 Compressor 2103 Lossless compressed bit stream with markers 2 04 Parser 2106 Channel or storage 2107 Decompressor 2108 Decompressed image 2401 Lossless compressed data with marker 2402 Parser 2403 Communication path 2404 Decompressor 2405 Display module 2500 Code stream 2501 Decompression including quantization 2502 Image processing or distortion model 2503 Lossless decompression 2504 Image processing or distortion model 2505 MSE or HVS difference model 2506 Alignment adjustment 2801 Coefficient 2802 Sign bit 2803 Tail-on information 2804 Context bit 2805 Coefficient bit

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] July 2, 1997

[Procedure amendment 1]

[Document name to be amended] Drawing

[Correction target item name] All figures

[Correction method] Change

[Correction contents]

FIG.

FIG. 2

FIG. 3

FIG. 11

FIG.

FIG. 16

FIG. 41

FIG. 4

FIG. 5

FIG. 6

FIG. 13

FIG. 21

FIG.

FIG. 35

FIG. 38

FIG. 7

FIG. 8

FIG. 9

FIG. 14

FIG. 28

FIG. 29

FIG. 10

FIG.

FIG. 24

FIG. 26

FIG.

FIG. 36

FIG. 37

FIG. 39

FIG. 40

FIG. 42

FIG. 43

FIG. 44

FIG.

FIG.

FIG.

FIG. 23

FIG. 45

FIG. 46

FIG.

FIG. 27

FIG. 53

FIG. 25

FIG. 31

FIG. 33

FIG. 32

FIG. 34

FIG. 47

FIG. 48

FIG. 49

FIG. 50

FIG. 51

FIG. 54

FIG. 52

FIG. 55

FIG. 56

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Ahmed Zandy United States 94025 Menlo Park Sand Hill Road 2882 Ricoh Corporation, California (72) Inventor Martin Borrick United States 94025 Menlo Park Sand Hill Road 2882 Ricoh Corporation ( 72) Inventor Michael J. Gormish 94025 Menlo Park Sand Hill Road, California, USA 2882 Ricoh Corporation

Claims (43)

[Claims] 1. A coder of a wavelet system for compressing image data by a reversible embedded wavelet, a coder of a binary system for compressing image data by a binary encoding scheme, and a selection control connected to select a wavelet system or a binary system. Data compression system consisting of parts. 2. The data compression system according to claim 1, wherein the wavelet type coder includes a reversible wavelet transform unit, an embedding ordering quantizer connected to the reversible wavelet transform unit, and an embedding ordering quantizer. A data compression system comprising a connected context model. 3. The data compression system according to claim 1, wherein the wavelet type coder further comprises an entropy coder. 4. The data compression system according to claim 1, wherein the wavelet method performs Gray coding. 5. The data compression system according to claim 1, wherein the wavelet method and the binary method share one encoder. 6. The data compression system according to claim 1, further comprising an entropy coder. 7. The data compression system according to claim 6, wherein the entropy coder comprises a finite state machine coder. 8. The data compression system according to claim 7, wherein the finite state machine coder comprises a look-up table. 9. The data compression system according to claim 6, wherein the entropy coder comprises a Q coder. 10. The data compression system according to claim 6, wherein the entropy coder comprises a QM coder. 11. The data compression system according to claim 6, wherein the entropy coder comprises a parallel coder. 12. A reversible wavelet transform unit, an embedding ordering quantizer connected to the reversible wavelet transform unit, a context model connected to the embedding ordering quantizer, an embedding binary encoding mechanism, and the context model An entropy coder connected to the embedded binary encoding mechanism, wherein the reversible wavelet transform, the embedding ordered quantizer and the context model are operable to compress image data with the reversible embedded wavelet; A data compression system operable to compress image data according to a binary encoding scheme, the binary encoding mechanism further comprising a selection control unit connected to select a wavelet type or a binary type. 13. The data compression system according to claim 12, wherein the binary system performs Gray coding. 14. The data compression system according to claim 12, wherein the entropy coder comprises a finite state machine coder. 15. The data compression system according to claim 14, wherein the finite state machine coder comprises a look-up table. 16. The data compression system according to claim 12, wherein the entropy coder comprises a Q coder. 17. The data compression system according to claim 12, wherein the entropy coder comprises a QM coder. 18. The data compression system according to claim 12, wherein the entropy coder comprises a parallel coder. 19. The data compression system according to claim 17, wherein the forward transform comprises a reversible wavelet. 20. A histogram compression mechanism, a reversible wavelet transform unit connected to the histogram compression mechanism, an embedding ordered quantizer connected to the reversible wavelet transform unit, and a context modeling connected to the embedding ordered quantizer. A system comprising a mechanism and a coder connected to the context modeling mechanism. 21. The system according to claim 20, wherein
A data compression system wherein the histogram compression mechanism creates a boolean histogram. 22. The data compression system according to claim 20, wherein the histogram compression mechanism maps an integer value to all possible pixel values in the image data. 23. The data compression system of claim 20, further comprising a notification mechanism connected to notify a decoder of a mapping used in the histogram compression mechanism. 24. The data compression system according to claim 23, wherein the mapping is notified by a header included in the compressed data received by the decoder. 25. The data compression system of claim 23, wherein one bit in the header, when set, tells the decoder that another histogram is used for the current tile. A data compression system characterized by indicating. 26. The data compression system of claim 23, wherein the decoder is notified by sending a number of bits equal to the dynamic range of the value, wherein each bit in the number of bits is a dynamic range. A data compression system characterized in that it is set when the corresponding value in is used. 27. A memory for storing a code stream having a header with at least one marker, at least one output device, connected to the memory, and connected to receive device characteristics from the at least one output device. A data compression system comprising a parser, wherein the parser is operable to perform device-dependent quantization. 28. The data compression system according to claim 27, wherein the code stream comprises lossless compressed data. 29. The data compression system of claim 27, wherein at least one marker indicates a number, subsampling, and alignment of components used for each tile in the codestream. . 30. The data compression system according to claim 27, wherein the code stream includes a main header, and a local header precedes each tile in the code stream. 31. The data compression system of claim 30, wherein the main header applies to all tiles in the codestream, and each local header applies only to the associated tile. 32. The data compression system according to claim 31, wherein at least one of the local headers has priority over the main header. 33. The data compression system according to claim 27, wherein the parser uses a marker in the code stream to quantize the code stream. 34. The data compression system according to claim 33, wherein at least one of the markers indicates frequency information. 35. The data compression system according to claim 27, further comprising a compression device for generating a code stream. 36. The data compression system according to claim 27, wherein the parser comprises a quantization selection device. 37. The data compression system according to claim 36, wherein the quantization selection device performs transformation and quantization of the set of images by discarding bit planes of various coefficients. 38. The data compression system of claim 27, wherein one of the tags indicates a level of importance in the data in each tile. 39. The data compression system according to claim 27, wherein the tag indicates a significance level locator signal, and the parser aborts according to the signal. 40. The data compression system of claim 27, wherein the tag indicates a number of importance levels to be stored. 41. The data compression system according to claim 27, wherein the tag indicates the number of bytes to be stored. 42. The data compression system of claim 27, wherein the tag includes an indication in each tile that associates an importance level with a byte count. 43. The data compression system of claim 33, wherein at least one marker indicates a number of bytes of a significance level in each tile.