JP2002543688A - How to copy media files - Google Patents

Abstract

The present invention copies a media file from a server to a client over a network and plays the file on the client, thereby providing an extra band that is not required to play the media file. A method is taught in which width is used to improve the copying of media files on a client. This leads to an entirely new browsing function, (i) viewing the sequence of video frames at a reduced speed (eg, slow motion view) or quality, or (ii) re-viewing the sequence of video frames. Or (iii) the typical viewing operation of pausing the viewing of a single video frame results in a video frame or frame sequence quality that is substantially better than normal playback speed. This is particularly useful when the user browses in these ways, as the user will probably be interested in even the most detailed parts.

Description

DETAILED DESCRIPTION OF THE INVENTION

TECHNICAL FIELD The present invention copies continuous media files, such as video and audio files, and encodes the files stored on the server to improve devices such as clients and other servers. A method comprising provided browsing and downloaded features. The network may include the Internet.

BACKGROUND OF THE INVENTION All types of media files, such as video and audio files, are
In recent years, there has been considerable interest in improving the technology of delivering to PC clients over low-bandwidth networks, driven not least by the ever-increasing use of video on eb pages. One useful set of features for video browsing is used in the Indeo product from Intel. In this system, one encoded video file is composed by the author and played on the client at many different data rates and quality levels. Each of these speeds and quality levels is adapted to the bandwidth of different networks or the platform's ability to play. The use of the "progressive quality" option in indio breaks down the video download by up to six phases. In the first phase, the lowest quality frames are delivered, and in subsequent phases, higher quality frames are delivered.
"Quality" in Indio has three levels: "Good" and "Bett".
er) "and" Best ". Therefore, the quality of the video is temporally divided into three discrete stages: (i) because of the" Good "quality. When the download of the encoded data is completed, (ii) “Better”
The stage when the download of data encoded for "best" quality is completed, and (iii) at the top level when the download of data encoded for "best" quality is completed. It is built when it reaches.

IEEE Transactions on Circuits and Systems in Video Technology, Special Issue on Image and Video Processing in Growing Interactive Multimedia Services,
September 1998 issue (IEEE Trans. Circuits & Systems for Video Technology, S
pecial Issue on Image and Video Processing for Emerging Interactive Mult
imedia Services, Sept. 1998), Beong-Jo Kim, Zixiang Xiong, and Dixian Xiong.
and William Pearlman), “Very Low Bit-Rate Embedded Coding with 3D Set Partitioning in Hierarchical Trees.
D Set Partitioning in Hierarchical Trees).
It discloses the application of the SPIHT compression method to wavelet-based transformation. This transform produces a coded bitstream of many spatial resolutions, with progressive quality ordering within a given spatial resolution.

SUMMARY OF THE INVENTION According to the present invention, a media file is copied to a device over a network,
A method of making a copy of the media file playable, so that extra bandwidth not needed to play the media file is used to improve the copy of the media file on the device. Be provided. Playback and enhancement (enhancement) may be controlled by respective sets of playback parameters and enhancement parameters. In one embodiment, these sets of parameters are also known as playback profiles and enhancement profiles.

[0005] The extra bandwidth can be achieved by (i) playing the media file at a reduced speed or quality, or (ii) playing a portion of the media file, or (
iii) The media file may be made available as a result of pausing playback. Therefore, the present invention contemplates a completely new browsing function, in which (i) slow (eg, slow motion view)
Alternatively, view a sequence of video frames in low quality (eg, fast scan at low resolution), or (ii) view the sequence of video frames again, or (iii) pause and replay a single video. The typical browsing operation of viewing frames results in extra bandwidth being available, which can be used more intelligently to improve the copying of media files on the device, for example, by Substantially better video frame or frame sequence quality is obtained (immediately or as a result) than playback speed. This is particularly useful when the user browses in these ways over a network, since the user will probably be concerned with even the finest details.
The invention has particular application to flowing continuous media files.

The term “network” should be interpreted broadly to encompass any kind of data connection between two or more devices. The network typically includes the Internet. A "file" can be any consistent dataset, so a "media file" is a consistent dataset representing one or more media samples, such as video frames or audio levels. . The term "client" is used herein to mean any device that receives data.

The enhancement is stored in the server, or the profile transmitted to the server by the device before or during the copy, or the value stored in the server and transmitted to the server by the device before or during the copy. It may be executed according to a combination with the set value. Enhancement generally occurs as a result of transmission of refinement data over a network.

[0008] The data required for reproduction is stored in the server, transmitted to the server before or during the copy, or transmitted to the server by the device before and during the copy of the value stored in the server and during the copy. The selection may be made according to a profile that is a combination with the transmitted value.

In one embodiment, the copy of the media file on the client acts as a temporary cache, in which all or part of the media file is deleted during or at the end of playback. A local cache or local buffer may be used to store data that is transmitted using extra bandwidth and that is decoded only when needed to improve the copy of the media file on the device. good.

In one embodiment, a media file encodes a video, and image data corresponding to each video frame or sequence of frames is generated using a wavelet transform and deformed using SPIHT compression, resulting. The bitstream includes several discrete bitstream layers, each of which allows image data to be displayed on a display at a different spatial resolution.

In another embodiment, the apparatus further comprises: (i) analyzing the media data; (ii) recompressing the media data to a new format;
) The media data is stored on paper, film, magnetic tape, disk, CD-ROM
Alternatively, the data is provided for purposes other than playback, including transferring to another medium, such as another digital storage medium.

Typically, the media file encodes video, audio data, or composite video and audio data.

In one aspect, a media file is provided that is copied using any of the methods defined above.

[0014] In another aspect, a client may improve the copy of a media file copied from a server to the client by a network by providing an extra band that is not required to play the media file at the client. A computer program is provided that is enabled by using width. In a related aspect, there is provided a client programmed with such a computer program.

[0015] In yet another aspect, the server improves the copy of the media file copied from the server to the client over the network, with additional bandwidth not required to play the media file at the client. Is provided. In a related aspect, there is provided a server programmed with such a computer program.

Detailed Description A. Key concepts Block-based motion-compensated coding schemes Discrete cosine transform, motion detection and compensation to compress video by removing spatial and temporal redundancy from the source. There are several example coding schemes. Of these, the most familiar is MPEG (i.e., MPEG-1 or MPEG-2). MPEG has three types of compressed images:
I-frame, P-frame, and B-frame are used. An I-frame is made up of square blocks (macroblocks) of 16x16 pixels, represented as a set of spatial frequencies obtained from a discrete cosine transform (DCT), with lower spatial frequencies generally being higher than higher spatial frequencies. Are also expressed with much better precision. Temporal redundancy is removed using P (prediction) and B (bidirectional) frames. P
-A particular macroblock in a frame is generated at the encoder by searching a previously encoded I-frame (or P-frame) for the macroblock that best matches the one under consideration. When one is found, a vector representing the offset between the two is calculated. This motion vector is
All with errors between the predicted block and the actual block, P-
It needs to be sent to reconstruct the frame. The third type is called a B (bidirectional) frame. B-frames are past and future I-,
And P-frames, which provide a smoothing effect, increase the compression factor and reduce noise.

Since P-frames are calculated from previous P-frames, errors also accumulate. Therefore, it is necessary to insert I-frames periodically. A typical MPEG frame sequence might look like this:

IBBPBBPBBIBBPBBPBB I .......... From the point of view of the present invention, MPEG defines three characteristics of encoded media files that are factors in the design of a system as described herein. Is illustrated.

The first property is that the initial sample structure is preserved during the encoding process, ie the identity of the individual frames is not lost in the encoded bit stream. This means that temporal properties can be manipulated by adding or removing frames in the compressed domain. Second, a temporal window is defined (MPEG group or GOP of images) in which the temporal redundancy is exploited to perform compression. Third, a complex set of dependencies is defined by the encoding.
That is, in MPEG, a P-frame requires an I-frame and a B-frame requires I and P frames for decoding.

H. 261 and H.E. There are other examples of block-based coding schemes that utilize motion detection and compensation, including H.263.

Coding scheme only within a frame in block units Other schemes using block-based coding without motion compensation, notably JPEG and DV (defined by SMPTE 314M-1999), etc. . In both JPEG and DV, the basic scheme is to transform the pixels of the block into frequency components using DCT, quantize that component by multiplying the current variable length code by a set of weighting factors, The result is to generate an encoded bitstream. However, DV introduces the concept of feed-forward to optimize the compression process prior to the compression applied. To do this, the DCT transformed image is examined and classified into regions of low, medium and high spatial detail. Using this information, different tables of quantization factors are selected for the frequency response of the human audio-visual system, along with the target to match which frequency coefficient is represented and its fidelity according to the region. Used.

The second feature of DV is the use of adaptive block / field processing. This means that a DCT block with 64 entries is a pixel of an 8 × 8 area in a non-interlaced frame (8-8 DCT), or
Two 4 × 8 areas (2-4-8 DC) in the first and second fields of the frame
T) can be expressed. The choice between the two is made by motion detection. The former method is used if there is almost no motion occurring between fields, and the latter method is used if motion is detected. The selection made here is based on a block unit.

Similarly, for MPEG coding, three observations are made regarding the nature of the coded bitstream. As mentioned earlier, through encoding, the sample structure is preserved, and then a temporal window is defined which in this case represents the two fields of the frame. Third, a set of dependencies is defined. For example, whenever a 2-4-8 DCT block is generated, dependencies exist between fields in the frame.

Subband Coding Scheme A more recent alternative to block-based coding schemes using transforms such as DCT is subband coding. This processes the complete image (rather than its small blocks) into a set of frequency / space limited bands. This is often referred to as a scale. This example is a wavelet transform.

The wavelet transform has only recently matured as a tool for image analysis and compression. For example, references include IEEE Transactions on Pattern Analysis and IEEE Transactions on Machine Intelligence.
Machine Intelligence, Vol. 11, No. 7, pp. 674-692, Morat, Stephan G., July 1989. (Mallat, Stephane G.), "A Theory for Multiresolution S
ignal Decomposition: The Wavelet Representation), which describes the Fast Wavelet Transform (FWT). The FWT creates a factorial of 2 hierarchical structure or subband, in which each step is The spatial sampling frequency-the "fineness" of the represented detail-is reduced by a factor of two in x and y.This procedure reduces the image samples and most of the energy to a small number of high-amplitude coefficients in the subband. And the remaining coefficients are approximately zero or smaller, providing a significant opportunity for compression.

Each subband describes an image by a particular combination of spatial / frequency components. This is illustrated in FIG. 2A. Here, the wavelet filter is applied twice to the image. After the first application, at level 0, the four subbands result in each subband being one quarter of the original scale. After the second application, four new, eighth-scale sub-bands are created. To fully reconstruct the image, the procedure will be performed in the reverse order using an inverse wavelet filter. The figure also illustrates the scheme used to name the subbands. That is, "L" and "H" denote low-pass and high-pass wavelet filters, which are applied in x and y to generate subbands.
Applying an H filter to x and y gives the “HH” subband, “L” to x,
Applying “H” to y results in “LH”, and so on.

The subbands do not always have to be completely reconstructed into the original image. They can be combined in different ways to produce the final image, satisfying individual demanded sets. Three examples are shown in FIG. In FIG. 2B, the level 1 LL subband is used as a 1/16 scaled version of the original. In FIG. 2C, the four sub-bands at level 0 are transformed back to one-quarter of the original.
Restructure the edition. In FIG. 2D, a third example, the LL and LH subbands at level 1 are used together with the LH subband at level 0 to reconstruct the original resolution image, but with horizontal features at all scales. Be emphasized.

Image Tree-Based Compression A tree-based data structure is used by the compression scheme to take advantage of spatial redundancy in the image. A recent development of such a scheme is the SPIHT algorithm. IEEE Transactions on Video Technology Circuits and Systems, Special Issue on Image and Video Processing in Growing Interactive Multimedia Services, September 1998 (IEEE Trans. Circuits & Systems for Video Technology,
Special Issue on Image and Video Processing for Emerging Interactive Mu
ltimedia Services, Sept. 1998), Beong-Jo Kim, Zixiang Xion, and Zixiang Xion.
g and William Pearlman) “Very Low Bit-Rate Embedded Coding with Partitioning of 3D Sets in Hierarchical Trees”
3D Set Partitioning in Hierarchical Trees) ”.

[0029] The SPIHT algorithm is an effective compression step used in conjunction with the wavelet transform. This is because a high level of data reduction can be obtained by efficiently utilizing the decorrelation of the input image provided by the conversion. For the purposes of the present invention, an important feature of the SPIHT algorithm is the Significance
) Is the ability to analyze the image converted. It is the bit importance (sign
efficiency, i.e., efficiently locate and partially process all samples for the position where the most significant bit in the sample is set. Since this corresponds to the size of the sample, the sample is efficiently processed in terms of its energy or contribution to the object to be reconstructed. The critical layer is created by selecting the bit position and outputting the value (1 or 0) of that bit position for every sample for which the bit position is defined.

Similarly, for MPEG coding, a set of dependencies between related image components is defined by a scheme such as wavelet / SPIHT compression. In this case, the significant layer is decoded only when the next most important layer is also decoded. Similarly, subbands (eg, LH in FIG. 1)
In order to decode 0), the parent subband (LH1 in FIG. 1) must also be decoded.

3D Schemes Wavelet and SPIHT compression schemes are extended to the third (time) dimension in a very straightforward manner. In this case, a frame sequence is obtained and processed as three-dimensional block data. The wavelet filter is applied along the time axis as many times along the vertical and horizontal directions, resulting in transformed blocks having a set of spatial and temporal cubes, each of which is a subband .
The 3D-SPIHT algorithm represents these subbands by eight trees (as opposed to four trees in 2D), thus producing a bitstream.

As will be appreciated by those skilled in the art, other compression schemes can also be used with the present invention.

B. Overview Image Encoding In one embodiment, the method of the present invention encodes image data such that the resulting data for a frame sequence is divided into two or more slices using an encoding process. Is required. Different slices will encode different parts of the image sequence at different resolutions and qualities. At a particular display resolution, quality, and frame rate, it must be possible to decode a subset of the complete data set in order to reconstruct the frame sequence.

In a preferred example, the division of the bitstream generated using wavelet transform and SPIHT compression into slices is described later in this specification. By decoding one or more slices of this data, it is possible to reconstruct a frame or sequence of frames at a particular display resolution, quality, and frame rate. In general, the quality of the resulting frame or frame sequence depends on the total amount of data to be decoded. The selection of the slice used to reconstruct the frame or sequence of frames is based on the desired resolution, quality, and frame rate. Also, by adding more slices, the quality, resolution,
And it is possible to improve the frame rate. This general process is called “improvement (
(In the wavelet-based example given later, the term "refinement" is used for this process.) Typically, an enhancement involves the existence of a new slice. This can occur by adding to the current slice or by adding the first slice from a new layer, possibly, but rarely, by adding all new layers (all slices for that layer). ) May be involved.

C. System Description A system capable of performing the method of the present invention has an encoder, a network, a server, and N clients, as shown in FIG.

The encoder compresses the incoming media, layer it, and before transmitting it as a bit stream over a digital communication network,
Add layer labeling information and store it on the server as described in the next section.

In FIG. 1, there are N clients, each engaged in a session with a media server, during which media content and control information are transmitted. For each client, a control channel to the server is maintained so that the client can request the server for media to be transmitted. The types of requests made include (but are not limited to): • Seek to a specific point in a media file. • Send media clips with a specific quality. • Improve media quality at the client in a specified way by sending special information.

These requirements support various useful application-level features at the client, such as video preview using fast forward and rewind, forward and backward in one frame steps, and frame freeze. For example, in a media browsing application, a low-quality version of the media is sent to the client with little delay, allowing the user to quickly start a task, for example, quickly scanning a file and locating items of interest. To Initially, the user may not be as concerned with the absolute quality of the media as much as the responsiveness of the system, as he demands to move quickly from one point to another. Upon finding the item of interest, the application can determine that this particular section of the media should be rendered with better quality and issue an appropriate request to the server. The server computes which extra layer should be delivered to improve the media information in the requested manner and sends only that layer. Note that extra processing of the media file is not required during procedures other than locating and transmitting an appropriate set of data. Also, quality improvements can occur in many ways. This includes: Increase the temporal resolution of the media (ie, for video, allow more individual frames to be analyzed). -Increasing the spatial resolution of the video. -Increasing the audio sampling frequency.・ To reduce distortion of media.

At the heart of the operation of this scheme is the conversion of media files in any encoding format, including those described in the previous section, into a layered format with properties that support the operations described above. is there. This is the purpose of layer labeling as described in the next section.

D. Description of Layer Labeling Overview A layered bit stream divides media into packets (here called chunks or slices) in a manner determined by the coding type, and labels (Labels). It is built from media encoding by uniquely describing its contribution to the reconstructed file in addition to the chunk. Static (stat) tools are needed to build and configure distributed tools and to understand the format of many layered bitstreams in a distributed environment.
ic) Signal information (a constant over the life of the stream) is made available as a reference to a globally available file or as data transmitted once to a bitstream. Provision is made for the dynamic signaling information carried in the bitstream with respect to configuration information that must change within the stream.

To work, a labeling scheme must do three things. -To enable a part of a media file to be efficiently and controllably selected to obtain a particular quality of service to meet media quality, bit rate or delay requirements. • Efficiently support the location and transmission of media parts needed to reconstruct a particular item to a specified quality. • Encapsulating basic media coding to hide complex details, while providing a uniform, media coding independent, layered view file.

The initial request is handled through the concept of a filter that specifies which parts of the file should be selected for transmission. The second requirement requires in-band signaling information to describe the layered structure in the bitstream. The third requirement is addressed through an indirect mechanism that maps specific coding details into a uniform layered view. This indirect mechanism is named Codebook. A codebook is defined for each new coding scheme or variant of an existing scheme with new parameters.

Layer Labeling Format This section defines the format of the layer labels used in the present invention and converts the media file into a layered stream format. Streams are stored and transmitted as contiguous groups of chunks. Each chunk is
It has a chunk header that defines the length and type of that chunk. Some chunk types are reserved for carrying data (named slices), while others are used for signaling.

Special configuration information about these streams, such as encoding formats and slice dependencies, is associated with the data streams. The static portion of this information is stored externally or in-band using chunk signaling. Dynamic information is always stored in-band.

Chunk Header All chunks are guided by a chunk header, the format of which is shown in FIG. All chunks, regardless of type, are introduced by one of these headers. The result is a data stream with a fixed structure, while leaving room for future expansion.

Type The 2-bit type field is used to distinguish between four independent sets of chunk types. (The issue of independence is important because there is no way to specify a dependency between two labels belonging to different types.) The following types of values are currently defined: Data Chunk (data chunks) (slices) Signaling chunks.

Length A 22-bit long field gives the chunk length in bytes including the header. Therefore, chunks without payload information (ie only headers)
Is 4 in length.

Label The label field of 8 bits is a layer label for the chunk. The interpretation of the label depends on the type field (see above), so the header format allows four sets of 256 labels.

Terminology Data chunks (termed slices as equivalents) are used to encapsulate a stream of multimedia data. The label attached to the slice is used to indicate the nature of the slice contents and the dependency of the stream on other slices. The signaling chunk is used in the stream to pass special information about the data, such as details about how the data was encoded or filtered.

Slice The way slice labels are assigned depends on the exact nature of the multimedia data being encapsulated. Although there is a mutually agreed upon set of basic schemes among all tools, new allocation schemes can be added to tailor the system to accommodate different situations. The only restrictions are: • All slices in the data stream must be labeled using the same allocation scheme. • Slices must be uniquely labeled with respect to the data they contain. • Label assignments must be made in a slice-dependent order.

The first limitation is self-evident, but considers other advantages further. The slice label is taken as an indication of the data content of the slice and is used by the system to separate or merge the slices to obtain different quality data streams. The slice label must uniquely identify the actual data content so that the stream decoder may be stateless. As a result, slices that are too long to be encapsulated as a single chunk are represented in such a way that the decoder can determine which is the base data and which is the continuous information, Must be split.

Slice labeling must also reflect the hierarchy of dependencies, with all slices having numerically smaller label values than those that depend on them. This means that when the slices of the stream are encoded in dependency order, the label value is dependent on the dependent unit ("context" below).
(See ""), which not only simplifies the job of merging slices from previously split data streams, but also allows for implicit boundaries between contexts (if If the slice label is not numerically larger than the previous one, it must belong to the new context).

Context A group of slices related by dependencies (either spatially or temporally) is called a context. As described above, the label values for slices in the context must be adjusted so that they are in numerically increasing order. This usually allows for implicit detection, without the need for signaling chunks where the boundaries between contexts are clear. Based on the label allocation scheme described above, so that slices always precede those that depend on them,
The slices in the context are also adjusted in a dependent order.

Number of available slice labels (and limitations on unique slice labeling)
Therefore, the maximum number of slices with context will be 256.
The hierarchical structure of the dependency on the slice label is defined in the codebook (see below), and the duration of the data stream is fixed. It is not essential that slice labels are continuously assigned, and it is not essential that a specific slice label be present in all contexts.

As well as grouping with slices that are related by dependencies, context plays a very important role in defining the boundaries where editing is performed. There is no need for explicit signaling chunks to determine a valid edit point.

A context is assumed to be separate from its neighbors in temporal space and independent of them. In other words, when context is extracted from the data stream, each should be decodable as a standalone entity. One exception to this is illustrated by encoding schemes such as MPEG that have a temporal dependency between adjacent screen groups. This is handled by associating a temporal unit (see sample below) with "overlaps" with each context that can be marked as dependent on the previous context.

The value of the overlap stored in the media header (see below) defines how dependencies between these contexts are handled. That is, an overlap value n indicates that the first n samples of the context are dependent on the previous context. The normal overlap state is that the context is self-contained in time, so its default overlap value is zero. If it is non-zero, the slice dependency hierarchy must reflect the inter-context properties of the coding scheme, including additional "phantom" dependencies if necessary. Is accompanied.

Sample A multimedia data stream is usually represented as a sequence of discrete units, each of which has a well-defined temporal position in the stream. This is especially true for video data, which is typically modeled as a set of separate frames (eg, for example, MPE
Using G or the like, the frame is replaced in the encoding process,
Even if they may have been grouped). The data stream preserves the natural temporal units of the encoding scheme and has each discrete unit named a sample. Whether a context contains a single sample or a group of samples depends on the coding scheme used, but the particular technique usually follows a very strict repeating pattern. Thus, by default, each context contains a fixed number of samples and has a context interval (samples per context repetition count) defined in the system header (see below).

The distinction between context and sample is (3-D wavelet / SPIH
It is important when considering the temporal dependence of multimedia data (such as the T scheme) and the ability to perform temporal reductions (such as playing every fourth frame of a video sequence). The context is a temporal coding unit (MPE
It contains the minimum number of samples that make up the GOP in G and the GOF in wavelet / SPIHT), with the spatio-temporal dependencies handled in exactly the same way. Within the context, each slice is given a temporal priority (see below) that allows for a high degree of temporal shortening, but by itself does not imply any kind of temporal dependency. Absent.

System Header (System Header) The system header is used to define attributes of a stream that are relevant only in the system layer. Basically, this means the parameters required by the server and provides the level of service required for them. For example, a default context interval is needed to configure the mapping between context and sample.

Media Header The media header is used to define the attributes of the stream that are relevant only in the media layer. This implies parameters that are needed by the stream decoder to make the data stream meaningful, but not required by the server. Examples that fall into this category are the horizontal and vertical resolutions of video coding, overlapping dependencies between contexts, and references to the codebook used for coding.

[0062] Rather than combining information with that of the codebook, the reason that there is a separate media header chunk is that the codebook has the property of being comprehensive and independent of the raw nature of the original media. This is because there is a tendency. The media header satisfies these details for a particular recording, thus greatly reducing the number of codebooks required by the system as a whole.

Code Book A code book is used to define slice dependencies and quality relationships within a context, and is needed to construct filters on the data stream. Regardless of whether the dependency is spatial or temporal,
The relationship between individual slices is represented as a simple hierarchical structure. That is, each slice type (ie, a slice with a particular label) has a predetermined set of other slice types on which it depends. Although the nature of the hierarchy clearly depends on the labeling scheme used, the codebook technique allows for great flexibility and does not violate the basic rules of slice labeling. (See above).

In a codebook environment, “dependency” has the following meaning. Because one slice depends on another slice, it must be impossible to decode the slice without it. (For example, this is true for MPEG video where an I-frame is required before the first P-frame is decoded.)
When it comes to producing acceptable quality, this dependency relationship says nothing about the liking of slices.

As well as defining the hierarchy of dependencies, it is the job of the codebook to store the mapping between individual slice values and the “quality” of the decoded data stream containing the slices. As needed, compared to the strict slice dependence described above, the quality of the stream is ambiguous and a subjective factor. That is, in fact there is probably more than one quality dimension. For example, 3-D wavelet / SP
IHT encoded video is considered to have four quality axes: scale (image size), fidelity (image quality), color, and temporal ambiguity.

Most codebooks have a series of tables (one for each slice label) containing a set of dependency data, temporal priority, and quality parameters, one for each quality axis. I have. The number and type of quality axes defined depends on the nature of the multimedia data and the coding scheme used. To simplify the problem,
Three important restrictions apply to the quality axis.

The quality axis is assigned a name (or quality tag) with overall importance. Codebooks employing the same quality tag must use the same allocation scheme when it comes to quality parameters for the relevant axis.

The quality parameters are always specified for the highest quality (unfiltered) data stream. For example, in the case of video, this allows a common codebook to be used regardless of image size.

The quality parameters themselves, if practical, must be independent of the actual coding scheme used. For example, the scale parameter for video data is
If the aspect ratio does not change, it may represent the width or height of the scaled image compared to the original.

The codebook header contains a list of quality tags, for which quality parameters will be found in the following table. Combined with the limitations outlined above, this allows the creation of filters to be done in a manner that is completely independent of the type and encoding of the multimedia data. Using a filter created with a missing or unspecified quality axis always results in slices not being filtered with respect to that axis.

Temporal Priority It is often necessary to shorten the data stream, reduce bandwidth requirements, or play the data stream at a faster than normal rate. Given a different data encoding scheme, it is not always wise to operate with a naive subsampling scheme such as "one sample every four". There are two reasons for this.

The inter-slice dependence of relevant samples may prohibit simple shortening.
This is true, for example, for MPEG video. In this case, there is a temporal dependency such as a complicated spider web between the I, P, and B frames.

When attempting to maintain a cache of sliced data and ensure efficient use of available bandwidth, repeated (or sophisticated) sub-sampling requests can be carefully coordinated to reduce data overlap. Maximizing is essential.

Thus, each slice of the context is assigned a temporal priority that is used to control the shortening. For example, in MPEG video, temporal priorities of 0, 1, and 2 are assigned to slices belonging to I, P, and B frames, respectively. It is these temporal priorities that are used when performing temporal filtering below the level of the context.

Signaling Chunks Signaling chunks are used to identify and annotate data chunks in a data stream. Whether they are stored together by the server or generated on the fly at delivery time is a detail at the implementation level that is outside the scope of this document. The signaling chunks have three distinct categories: static, dynamic,
And padding.

Static chunks Static chunks define parameters that do not change over the life of the data stream. This includes defining basic default values that apply to the entire stream, such as the identity of the codebook employed to generate the data slice. Because these chunks are static, i.e., it is reasonable that the information is transmitted out of band, or by reference, as in the case of reusable codebook tables, they are: It need not be included as part of the data stream. However, if they are transmitted or stored in bands, they must be unique, prior to the first data slice, and stored in numerically increasing order of the Opcodes. Must.

Dynamic Chunks Dynamic chunks define parameters that change depending on their position in the data stream. This includes filtering information because the filters used to generate the stream can change when the slice is delivered. It also contains all variable context and sample information, such as indicating "seek" operations and processing contexts with non-constant context intervals. By nature, dynamic chunks carry location information in the data stream and must always be transmitted in bands. If it is present, dynamic chunks are only valid at context boundaries.

Padding chunks Since padding chunks have no semantic effect on the data stream,
It can be placed on any chunk boundary.

Static Signaling Chunks System Header Type: Static Opcode: 0x00 Description: Information specific to the system layer only (media server), not required to filter or decode the data stream Is the context spacing Format: name / value pair Status: required Position: Outside the band, or inside the band before the first data slice.

Media header type: Static Op code: 0x01 Description: Information specific to the media layer only (decoder), necessary for operation of the media server. Parameters are original media parameters context dependency overlap code book reference Format: name / value pairs Status: required Location: out of band or first data slice In the previous band.

Code book type: Static Op code: 0x02 Contents: Specific information for generating slice filter only, unnecessary for operation of media server or decoder Quality tag list (slice dependency information) Slice quality information Slice temporal priority information Status: Mandatory Location: Out of band, by reference, or in band before first data slice.

Metadata Type: Static Opcode: 0x04 Content: Information related to the data stream but whose storage, filtering, or transmission is not required. Examples are: Data, Time, and Location Data History Copyright Notice Format: Name / Value Pairs Status: Optional Location: Outside the band or in the band before the first data slice.

User Data (1,2,3,4) Type: Static Op Code: 0x05, 0x06, 0x07, 0x08 Contents: Private information added to the data stream by the coding generator. Not interpreted in any way by the system Format: opaque byte array Location: Outside the band or in the band before the first data slice.

Dynamic Signaling Chunks Baseline Type: Dynamic Opcode: 0x80 Content: Information about the current position of the stream in relation to context and sample sequence number, usually within the data stream. Used to indicate “seek” operation. It has two parameters, any of which can be omitted if the value can be inferred from the stream position. Context sequence Sample sequence Unless otherwise specified at the beginning of the data stream, both the context and the sample sequence number are zero. Begins with Format: 64-bit little-endian Binary Status: Optional Location: In band at context boundary.

Context Start Type: Dynamic Opcode: 0x81 Content: Context interval for the immediately following context Format: 32-bit little-endian binary Status: Optional. Required only when context interval is different from default Position: In band at context boundary.

Context End Type: Dynamic Op Code: 0x82 Contents: None Format: n / a Status: Optional. Only needed when context boundaries cannot be determined by a simple comparison of slice type values. Position: Within the band at the context boundary.

Filter definition type: Dynamic Op code: 0x83 Description: Used to encode stream filter information for the following data slice. It takes the form of a (possibly compressed) pair of bit sequences that indicate which data is filtered from the subsequent stream.

The slice mask indicates which of the 256 possible slice types has been filtered from the data stream. It does not imply that the remaining slice types are found in the stream, but does ensure that existing slices comply with the dependency criteria specified in the codebook not.

[0089] A context mask is used to indicate which context has been filtered out of the data stream. Each context has a unique position in the sequence and the mask is a modulo 240 (numerical characteristic of 2, 3, 4, 5, 6, 8, 10, 12, 15, 16, 20, 24 and a common multiple of 30) Refers to the value at the position of the number selected to be). Unlike slice masks, context masks can be used to determine which context exists and which context does not. However, in the case of dependencies between contexts, there is no guarantee that relevant information actually exists. Status: Optional Location: In band at context boundary.

Padding Signaling Chunk Padding Type: Dynamic Op Code: 0xff Description: Opaque byte array. Padding chunks may be necessary if a fixed bit rate stream is encoded or if specific byte alignment is required. It is also used to relabel slices at fixed locations for data filtering experiments. Format: Opaque byte array Status: Optional Location: In band.

Example of Layer Labeling, Filtering, Merging, and Codebook FIG. 4 gives an overview of some typical transformations experienced by media files such as layered, filtered, and encoded. I have. In particular, the figure shows two possible styles of filtering: filtering at the slice level (within the dependency units) and filtering at the context level (passing through all dependency units). Filtering at the slice level is used to change the media color depth, spatial resolution, fidelity, or temporal resolution. Context-level filtering has a coarser "grain" and is designed primarily to support media time savings.

Wavelet Example FIG. 5 shows a more detailed example of labeling wavelet / SPIHT encoded data. A "2" depth wavelet decomposition with four importance levels results in a context of 28 slices. A portion of one possible codebook for this encoding is shown in FIG. 6, and an example showing how the codebook is used to generate the filter is shown in FIG.

In FIG. 6, the header of the codebook is a codebook ID (ID) for self-identification.
CodebookID) field, a quality tag (which specifies the quality axis available for this encoding)
QualityTags) field, a QualityDivisions field that specifies how these axes are labeled using just the text name, and an OriginalSize that specifies the spatial resolution of the original media.
Includes fields and provides a baseline for scale calculations.

Each coding scheme may have its own codebook defined, but any codebook that shares a common QualityTags entry will have its quality using the very same QualityDivisions scheme. You must label the axis. Text names are used to map similar but not identical properties of different media encodings into a single quality scheme. This results in a significant capability of the system. That is, the application
Written to manipulate media using these names in a manner that is completely independent of the encoding format.

In that figure, assume that the file manipulation tool is trying to extract a medium fidelity, monochrome, half-scale 352 × 288 video encoding. To accomplish this, the required value of the quality parameter (QualityParams) for the label whose codebook has the largest number, ie, “scale = half, fideli
ty = medium ”(actually this is optimized using a pre-calculated index). When this label (13) is found, until the dependency graph is completely analyzed, All dependent labels are examined, and as this proceeds, a bit mask is constructed that represents the slice filter, ie, a "zero (0)" at bit position n filters the slice with label (n). Required by the stream, a "1" means that there is no slice, and the result of using this filter is shown in FIG.

The filtered slice is preceded by a filter definition signaling chunk that holds the slice mask used to perform the filtering. This value is used by downstream tools that need knowledge of the stream characteristics to, for example, initialize a decoder or set up memory areas for buffering. The slice mask is also used in determining the slice content of the resulting stream, as shown in FIG. 8, where the filtered bitstreams are merged. Here, slices 16, 17, 20, 21,
The refinement stream, including 24, and 25, is merged with the half-scale, low-resolution filtered stream of FIG. 7 to obtain a full-scale, low-resolution encoding. The slice mask for the combined stream is obtained using a simple bitwise OR of the two input slice masks.

Example of MPEG FIG. 9 shows an example of labeling MPEG encoded data, and FIG. 10 shows a part of a corresponding codebook. In this example, it is assumed that the file manipulation tool wants to generate a filter that operates on the slice in such a way as to generate a fast playback stream. In this encoding, the slice is MPEG
Since it represents a frame, the effect of the filter is to shorten the stream in time. From the codebook, the quality parameter (QualityParam
s) resolve the hierarchy of dependencies at playrate = fast,
Build a slice filter as described above. The result of applying the filter is shown in FIG.

Example of DV FIG. 12 shows an example of labeling DV encoded data, and FIG. 13 shows a part of a corresponding codebook. In this example, the DV encoded data is labeled according to the following scheme. That is, data representing both 2 × 4 × 8 and 8 × 8 DCT blocks from field 1 is assigned slice level 0, and data representing 8 × 8 DCT blocks from field 2 alone is assigned slice level 1. . The result is two layers: a base layer with half the vertical resolution resulting from field 1 and a refinement layer that produces a reconstructed image with full resolution. FIG. 14 illustrates the result of applying a filter to select only the label 0 slice, which produces a video stream with half the vertical resolution.

E. Client-Driven Image Delivery The method described herein allows a client to specify information about an image it wants to receive and to have the server deliver the appropriate slice of that image data at the appropriate time. Become. The client can specify (among other parameters) the image size, image quality, frame rate, and data rate. The server uses this information to determine which section of the image data should be sent to the client and at what time.

The client can also specify the playback speed at which the data will be distributed. The display time of each frame is used to deliver the data in a timely appropriate manner for the selected speed.

F. Playback Parameters When a client requests data delivery, it typically specifies a set of playback parameters that are specific to the client's needs and status. The playback parameter may specify an acceptable value for any quality parameter specified for the media. The playback parameters may be specified by the client, may be already stored in the server, or may be configured by a combination of the parameters of the client and the server.

In particular, the playback parameters may define one or more of the following media parameters for an image or sequence of images: (i) the spatial resolution of the image or images; (ii) the image or images. (Iii) the number of images that can be displayed; (iv) the selection of color components in one or more images; (v) a subset of available frames to be delivered; and (vi) one or more. Attention area in the frame of.

The playback parameters may also define one or more of the following media parameters for audio: (i) audio distortion; (ii) audio dynamic range; (iii) transmitted audio. Number of channels; and (iv) choice of mono, stereo or 4-channel audio.

For codebooks, some of these parameters may be encapsulated in a client-generated or server-generated filter for the media.

G. The server playback enhancement may be specified by the client, may be already stored in the server, or may be configured by a combination of parameters of the client and the server. The server uses the playback parameters to deliver appropriate data to the client. This process involves the following steps: 1) The server determines the entire set of transmitted slices; 2) The server delivers slices at the appropriate rate.

H. Determining a Slice Set for Playback The server selects slices that need to be sent based on some of the playback parameters. It starts with a complete set of slices, then discards the slices based on: Discard all slices that have encoded color components that are not needed. Discard all slices encoded only for frames outside the specified frame set. Discard all slices encoded only for frames not required for the selected frame rate. -Discard all slices that have a parent layer coded for at least the required quality and are coded for quality higher than the required quality. Discard all slices that have a parent layer encoded for at least the required scale and that have been encoded for scales higher than the required scale. Discard all slices already marked as available on the client. Discard all slices for which any quality parameter is outside the tolerance specified by the playback parameter for that parameter.

Other selection criteria may be applied at this point.

I. Distribution of playback slice The server distributes the selected slice at the data rate required for display at the specified playback rate. Using the timing information associated with the slice, the data is delivered according to a standard coordination mechanism that drains the time-dependent data.

J. Discarding Data Sections In order to display the resulting frames in a timely manner, data must be delivered at the requested rate. When this is not possible, the server discards some of the slices. This may occur because the data rate of the selected slice is faster than the maximum rate specified in the playback parameters, or the transmission rate may be higher or lower, regardless of whether the transmission channel is permanent or temporary. It occurs because you do not have enough ability. Data is discarded according to a list of criteria, including those included in a subset of playback parameters known as discard parameters. The most important criterion is the dependency between the data section and its parent. Parents in a data section are never discarded before the data itself. The remaining criteria applied and their order are specified by the client in the discard parameters. The following criteria apply: Slices required for a given frame rate or higher may be discarded.・ Slices encoded for a given scale or higher can be discarded.・ Slices encoded for a given quality or higher can be discarded.・ Slices encoded for specific areas can be discarded.・ Slices encoded for specific color components can be discarded. • Slices with quality parameters that exceed a given limit may be discarded.

[0110] Other selection criteria may be applied in this regard. Each criterion in the list may refer to any quality parameter that applies to the slice, or the same quality parameter may appear in multiple criteria. For example, the discard parameter first discards all data of a certain quality or higher, then discards the replacement frame, and then discards the second, all data of a lower quality or higher. Data is discarded until the requested data rate falls below the currently available capacity.

K. Changes to Playback Parameters The client may change the playback parameters at any time. Taking into account the slices already delivered to the client, the server recalculates a new set of requested slices. If the slice is currently being transmitted, the transmission may be completed if no more slices appear in the request list, in which case the transmission ends. If the transmission is completed, the data stream is marked accordingly and the client is notified of the incomplete slice.

L. Data storage by the client When the client successfully receives the data, it is stored for decoding and reuse. Later, the server can improve the stored data by sending another slice for the frame or sequence of frames. The client appends this data to the existing stored data, giving it compressed data that can be decoded,
Generate frames or frame sequences with higher quality, resolution, or frame rate. If it takes time to display a frame or a frame sequence, the client uses a combination of the stored data and the data currently distributed from the server to provide a compression format of the frame and the frame sequence. This data is then decoded and displayed appropriately.

M. Handling of Limited Channel Capacity A data channel has a limited capacity that depends on factors such as the overall capacity of the network, the utilization of the network, and the capacity allocated to the channel.
If the capacity requested by the playback profile is greater than the available capacity, the server reduces the amount of data sent using the criteria given by the discard parameter. If the capacity required by the quality profile is less than the capacity available on that channel, or if it is not possible to exactly match the available data rate for playback, surplus capacity )
Is said to be.

N. Excess Channel Capacity Excess capacity can be made available in many ways. For example, the client may specify the maximum quality setting in the playback profile while changing the requested frame rate depending on the user action. When the video is being played in slow motion, the required capacity for the video data may be reduced to create extra capacity. In extreme cases, the user may pause playback and display a still image. Once the still image is transferred with the required quality, no more data is needed and the entire channel capacity is available as surplus capacity. Similarly, the client may play a previously delivered frame or frame sequence. Some of the data required for the frame has already been stored by the client, and the capacity required to play back the data at the specified quality will not be so large. Playback parameters may also be selected to specify distributable quality without using all available channel capacity. The remaining capacity is the surplus capacity. Thus, some or all of the channel capacity may be surplus.

Use of Surplus Capacity When there is surplus capacity in a channel, it is used to improve the quality of a video frame by preloading some or all of the image data for the video frame. This may involve loading data for new frames that were not previously downloaded (ie, "complementing" previously downloaded frames), or for more slices for those frames May involve improving existing frames. This process is known as enhancement and is under the control of a set of enhancement parameters. Therefore, "enhancement" includes supplementing previously downloaded frames. The set of enhancement parameters is also known as an enhancement profile, which is used to select the order of frames to be pre-loaded and the amount of data to be loaded for each frame.

For example, the surplus capacity may be used to load a low resolution version over the entire sequence and allow subsequent high speed playback through the data. Alternatively, the surplus capacity may be used to load the keyframes of the sequence to refine the details of the data. Alternatively, it may be used to improve frame quality around the current playback position.

The data may be retained at the client for later reuse, or may be treated as temporary and discarded unless the client wants to do so immediately. Is also good.

In some cases, there may be problems delivering data from the server to the client. For example, this may be the case when the distribution is over a network using an unreliable transfer protocol. In this case, there may be gaps in the stored data due to data loss in the network. The surplus capacity is used to fill these gaps and provide a complete set of compressed data.

O. Server Enhancement The server uses the enhancement parameters to deliver the appropriate data to the client. This process involves the following steps: The server determines the entire set of transmitted slices. The server orders slices according to the enhancement parameters. The server uses all available surplus capacity to deliver slices.

Slice Selection for Enhancement The enhancement parameters are used to select slices for enhancement in the same way that playback parameters are used to select slices for playback.

Ordering Slices for Enhancement Once an enhancement slice has been selected, there is an additional step in which the slices are reordered to meet the enhancement parameter requirements.
This allows the client to specify the order in which the slices are delivered, and therefore, how the extra bandwidth is used to improve or supplement a local copy of the image.

The slices are ordered by assigning a score to each slice. This score is calculated based on the value for each available quality parameter and the importance assigned to the quality parameter in the enhancement parameter.

The following parameters are used to score the slices: Scale / resolution available in a given slice; Distortion present in a given slice; Encoding by a given slice A region of interest in a frame encoded by a given slice; a color component encoded by a given slice; some other quality parameter specified for that slice; Other ordering criteria that may be applied at the time.

The slices are ordered according to the score for each slice. Finally,
The slices are ordered to ensure that the parent of the slice always appears before the slice itself, and that each slice is decoded as soon as it arrives.

Slice Distribution for Enhancement Slices for enhancement are distributed using all the surplus capacity available on the communication channel. Unlike playback, there is no adjustment or inflow of media data. Thus, even when not needed for immediate playback of media, sufficient capacity of the communication channel is always used.

Extension to Other Users In the description above, the resulting video frames are displayed on the client computer. This technique is also used to provide media data for other uses. For example, frames may be used as input to an image analysis system to provide features such as shot change detection and object recognition. The frame may also be recompressed to the new format and stored on media such as paper, film, magnetic tape, disk, CD-ROM, or other digital storage media.

Extension to Other Data Types This technique can be applied to other data types such as audio, graphics, animation. This technique can support any data that is compressed into a format that allows for enhancement as described above.

P. No. Appendix 1 Some Definitions of Terms Used in the Environment of the Illustrated Embodiment Scale Function is analyzed into a set of components with different time / space / frequency content. These components are called scales, and the process of analysis that turns them into scales is called multiscale decomposition. This analysis is performed by a limited time waveform and zero integration and is called a wavelet. Highly localized in time or space,
Components with a poorly defined frequency spectrum are small scale, capturing fine details. The components that are diffused in space but have a precisely defined frequency spectrum are large scale and capture the overall trend.

Layer A layer is a conceptual part of a media file that initializes or refines a single homogeneous component of the media file, where the components of the file are spatial resolution, temporal resolution, sample fidelity, Data representing color or another quality axis.
Therefore, it is a data set for one level of quality in a complete file.

Slice A slice is a part of a media file that encodes data related to a layer in a single context. Therefore, each layer in a particular media file is itself divided into a series of slices, one for each context.

Base Layer (Base Layer) This layer is transmitted first to a client and initializes the display of a media object in the client. A refinement layer is added to this layer.

Refinement Layer A layer sent to the client following the base layer, which measures the quality of the display of media objects at the client according to some metric.
"Improve.

Significance Layer A layer in which all refinement information refers to a particular bit position for all coefficients that experience refinement.

Scale Layer A layer in which all refinement information is for a specific scale.

Region Layer A layer in which all refinement information is directed to a specific related area in a space or time-varying function.

Distortion Image distortion is measured by the ratio of the maximum signal to noise (PSNR),
NR = 10 log (255 ² / MSE), where MSE is the root mean square error of the image.

Quality Tags A set of axes that can be used to represent different aspects of perceived quality, for example, spatial resolution, temporal resolution, and fidelity of reconstructed pixels with respect to the original.

Quality Divisions The method of marking the axis of the Quality Tags, eg, the axis of spatial resolution may be marked as low, medium, high.

Quality Parameters (QualityParams) A set of classifiers used to describe the contribution of encoded media items to the perceived quality of the reconstruction. The quality parameter (QualityParams) classifier is (Qua
lityTags = QualityDivisions), for example, fidelity
= 6, or resolution = high.

Codebook A table with an extraction mechanism so that the operations of quality operations are defined on valid media files regardless of the original compression format. Codebook (
Codebook achieves format independence through the use of a QualityParams classification system.

Filter A structure of information that defines the division of a layered media file into two parts. That is, one part represents the media file with reduced quality, and the other part represents the improvement information required to reconstruct the original file. Incorporation into the simplest implementation is a bit mask, with "0" and "1" in n bit positions.
Specifies whether a data item with a particular label (n) is required or not required in a low quality output file.

Filter Mask A piece of information that is added to a filtered file and informs a downstream tool about the information content of that file (ie, filtered from that file). If the filter was implemented in the implementation as a simple bit mask, the filter mask is simply a copy of this filter.

[Brief description of the drawings]

 The present invention will be described with reference to the accompanying drawings, which show the following:

FIG. 1 is a diagram of a network used to execute a media file distribution method according to the present invention.

FIG. 2 is a diagram illustrating subbands obtained as a result of applying a wavelet transform to an image in a conventional multi-scale decomposition, together with an example of partial reconstruction according to the present invention.

FIG. 3 illustrates a format of a “chunk” data structure according to the present invention.

FIG. 4 is a diagram illustrating a typical path of one unit of media by a layering system according to the present invention.

FIG. 5 illustrates a labeling mechanism as applied to wavelet coding according to the present invention.

FIG. 6 is an example of a portion of a codebook for the example labeling of FIG. 5, in accordance with the present invention.

FIG. 7 is a diagram illustrating slice filtering as applied to the example of labeling of FIG. 5, according to the present invention.

FIG. 8 is a diagram illustrating slice merging as applied to the example of labeling of FIG. 5 according to the present invention.

FIG. 9 illustrates labeling as applied to MPEG coding according to the present invention.

FIG. 10 is an example of a portion of a codebook for the example labeling of FIG. 9 in accordance with the present invention.

FIG. 11 is a diagram illustrating slice filtering as applied to the example of labeling of FIG. 9 according to the present invention.

FIG. 12 illustrates a labeling mechanism as applied to DV coding according to the present invention.

13 is an example of a portion of a codebook for the example labeling of FIG. 12, in accordance with the present invention.

FIG. 14 illustrates slice filtering as applied to the example of labeling of FIG. 12 according to the present invention.

──────────────────────────────────────────────────続 き Continuation of the front page (71) Applicant Mount Pleasant House, 2 Mount Pleasant, Huntingdon Road, Cambridge CB30RN, United Kingdom (72) Inventors King, Tony, Richard 9 Cambridge, England UK Nunham, Marlow Road 28 (72) Inventor Grouert, Timothy, Holloyd Cambridge CB2 5E brew, Little Shelford, Fitlessford Road 44 F-term (reference) 5C052 AB02 CC01 CC11 DD04 DD06 5C053 FA15 GA11 GB17 GB21 HA33 KA01 LA15

Claims (23)

[Claims] 1. A method of copying a media file to a device over a network and making the copy of the media file playable, wherein extra bandwidth is not required on the device to play the media file. The method of claim 1, wherein the method is used to improve copying of the media file. 2. The extra bandwidth includes: (i) playing the media file at a reduced speed or quality; (ii) playing a portion of the media file; or (iii) playing the media file. The method of claim 1, wherein the method is available as a result of pausing playback of a file. 3. The method according to claim 1, wherein the improvement is performed according to a profile stored in the server. 4. The method according to claim 1, wherein the improvement is performed according to a profile sent by the device to a server before or during the copying. 5. The improvement according to claim 1, wherein the improvement is performed according to a profile that is a combination of a value stored on the server and a value transmitted to the server by the device before or during copying. Alternatively, the method according to 2. 6. The method according to claim 1, wherein the data required for playback is selected according to a profile stored in a server. 7. The method according to claim 1, wherein the data required for playback is selected according to a profile transmitted to a server before or during copying. . 8. The data required for playback is selected according to a profile which is a combination of a value stored on the server and a value transmitted to the server by the device before or during copying. A method according to any of claims 1 to 7, characterized in that: 9. The copy of the media file on the client acts as a temporary cache, wherein all or part of the media file is deleted during or at the end of playback. 9. The method according to any one of 1 to 8. 10. The media file encodes video, and image data corresponding to each video frame or sequence of frames is generated using a wavelet transform and transformed using SPIHT compression, and the resulting bit stream is 10. A method according to any of the preceding claims, comprising several discrete bitstream layers, each bitstream layer allowing to display image data on a display at a different spatial resolution. . 11. The apparatus further comprising: (i) analyzing the media data; (ii) recompressing the media data into a new format; (iii)
) The media data is stored on paper, film, magnetic tape, disk, CD-ROM
A method according to any of the preceding claims, wherein the data is provided for purposes other than the playback, including transferring to another medium, such as a digital storage medium. . 12. The method according to claim 1, wherein the media file encodes audio data. 13. The method according to claim 1, wherein the media file encodes composite video and audio data. 14. A local cache for storing data that is transmitted using the extra bandwidth and that is decoded only when needed to improve the copy of the media file on the device. The method according to any one of claims 1 to 13, wherein the method is used for: 15. A local buffer for storing data transmitted using the extra bandwidth and decoded only when needed to improve the copy of the media file on the device. 2. A method according to claim 1, wherein
15. The method according to any one of claims 14 to 14. 16. The extra capacity comprises: (i) loading a low quality version of the data of the entire sequence to enable high speed playback with the data; and (ii) loading certain key frames for the sequence. Remove the degradation of the data (scrub
16. The method according to any of the preceding claims, wherein (iii) is used for one or more of the enhancements of enhancing the quality of the frame surrounding the current playback position. the method of. 17. The method according to claim 1, wherein the device is a client in a client-server network. 18. The method according to claim 1, wherein the device is a server or an edge server. 19. A media file copied using the method according to claim 1. Description: 20. Using a client to improve or supplement a copy of a media file copied from a server to the client by a network, using extra bandwidth not required to play the media file at the client. Computer programs made possible by 21. A client programmed with the computer program according to claim 20. 22. A server that improves the copying of media files copied from said server to a client over a network by using extra bandwidth not required to play said media files at said client. Computer program made possible. 23. A server programmed with the computer program according to claim 22.