CN1801946A

CN1801946A - Motion image encoding apparatus, motion image decoding apparatus, motion image encoding method, motion image decoding method, motion image encoding program, and motion image decoding program

Info

Publication number: CN1801946A
Application number: CN 200610005713
Authority: CN
Inventors: 铃木芳典; 文仲丞; 小林充
Original assignee: NTT Docomo Inc
Current assignee: NTT Docomo Inc
Priority date: 2005-01-07
Filing date: 2006-01-06
Publication date: 2006-07-12
Anticipated expiration: 2026-01-06
Also published as: CN100471280C

Abstract

A motion image decoding apparatus (200) for generating a prediction signal in blocks is provided with a low-resolution block decoder (151) for generating a low-resolution block with a smaller number of pixels than that of a prediction block by decoding encoded data. The motion image decoding apparatus (200) is further provided with an enhanced block generator (154) for enhancing a low-resolution block generated by a decoded low-resolution block to a block with the same number of pixels as that of the prediction block using a decoded image. Furthermore, the motion image decoding apparatus (200) is provided with a block divider (162) for generating plural small blocks by dividing an enhanced block based on a predetermined division rule and a small block predictor (152) for generating a predicted small block of a small block using a decoded image and the plural small blocks.

Description

Moving image encoding and code translator, method, program

Technical field

The present invention relates to the piece is the moving image decoding technique and the moving image encoding technology of unit generation forecast piece.

Background technology

In the past, the existing coded system as with the piece being the generation of the unit prediction signal (prediction piece) of carrying out the forecasting object piece for example had the technology of recording and narrating in the open patent 2002-354486 of Japan communique.This technology is exactly to carry out the technology that entropy coding reduces the tediously long degree of time orientation by prediction piece that generates to the motion vector of expression interframe movement, according to this motion vector and the error block between the forecasting object piece.At this, so-called forecasting object piece is the following block that carries out the incoming frame of prediction processing.

In order to improve the time prediction performance of the image-region that comprises careful athletic performance, need the information (below, be called " movable information ") of many motion vectors etc., in above-mentioned mode in the past, because need carry out the coding of movable information, being that the symbol weight of so-called additional information increases headed by the movable information.Thus, become the main cause that is difficult to improve the treatment effeciency in encoding process, the decoding processing.

Thereby problem of the present invention is exactly to increase by the symbol weight of restraining said additional information, improves the treatment effeciency in encoding process and decoding processing.

Summary of the invention

Dynamic image encoding device of the present invention is to be the dynamic image encoding device that unit predicts the generation of piece with the piece, possesses: the holding member (memory) that keeps the image after the decoding; Is the image segmentation of the coded object of having imported the 1st partition member of a plurality of forecasting object pieces; The conversion of above-mentioned forecasting object piece and after being encoded to the low resolution piece that pixel count also lacks than the pixel count of this forecasting object piece,, generate the low generation parts of differentiating piece of decoding by to the deciphering of the coded data of this low resolution piece; The low resolution of the above-mentioned decoding that has generated piece is amplified to the amplifier unit of the piece of the pixel count identical with the pixel count of above-mentioned forecasting object; By according to the rules cut apart the rule to cut apart the 2nd partition member that generates a plurality of fritters through the piece of above-mentioned amplification; The prediction parts that use image after the above-mentioned decoding and the above-mentioned fritter that has generated to generate the prediction piece of this fritter; By merging the above-mentioned a plurality of prediction fritters that generated and the merging parts of generation forecast piece according to the above-mentioned rule of cutting apart.

Dynamic image decoding device of the present invention is to be the dynamic image decoding device that unit predicts the generation of piece with the piece, possesses: the holding member (for example memory) that keeps the image after the decoding; The coded data of having imported is deciphered, generated the decoding unit of the pixel count low resolution piece also lower than the pixel count of prediction piece; The above-mentioned low resolution piece that has generated is amplified to amplifier unit with the piece of the pixel count same pixel number of above-mentioned prediction piece; Cut apart the partition member that this piece that has been exaggerated generates a plurality of fritters by the rule of cutting apart according to the rules, the prediction parts that use image after the above-mentioned decoding and above-mentioned divided fritter to generate the prediction fritter of this fritter; Merge these a plurality of prediction fritters that generated and the merging parts of generation forecast piece according to the above-mentioned rule of cutting apart.

In order between dynamic image encoding device and dynamic image decoding device, to carry out high-precision predictive coding, hope is in dynamic image encoding device, pixel count is carried out compressed encoding than the low resolution piece that the pixel count of predicting piece also lacks, coded data is sent to dynamic image decoding device.If adopt the present invention in this, because the low pixel count of differentiating piece also lack than the pixel count of prediction piece, so the symbol weight of hanging down compressed its additional information of coded data of resolution piece is also lacked than in the past movable information.Thus, the symbol weight that reduces additional information increases, its result, and the treatment effeciency in encoding process and decoding processing improves.

Particularly in dynamic image decoding device, at first, cut apart and amplify piece and obtain fritter, predict the generation of fritter according to the image after fritter and the decoding.Thus, can obtain to carry out a plurality of prediction fritters of the high-precision prediction in the careful zone of moving image.Then, by merging them, the generation forecast piece.Thereby, when above-mentioned treatment effeciency improves, can utilize the high-precision prediction of the few low resolution piece of symbol weight.

Wish that above-mentioned dynamic image encoding device further possesses: be the 3rd partition member of coded object piece to the image segmentation of above-mentioned coded object; Be that unit subtracts calculation through the above-mentioned coded object piece of cutting apart and above-mentioned prediction piece and the subtracting section spare of generated error piece with the pixel; The addressable part that the above-mentioned error block that has generated is encoded; Coded data through above-mentioned coding is deciphered, thereby generated the decoding unit of deciphering error block; Be that unit carries out adding of above-mentioned prediction piece and above-mentioned decoding error block and generates decode block with the pixel; And this decode block is outputed to the output block of above-mentioned holding member.

In addition, in dynamic image encoding device, amplifier unit uses the pixel through the low pixel of differentiating piece of the decoding of above-mentioned generation and the decode block adjacent with the top of forecasting object piece and the left side, generates the pixel count pixel count the same piece the same with above-mentioned forecasting object piece.

In addition, wish in dynamic image encoding device, generate parts by constituting with the lower part: transformation rule according to the rules is transformed to above-mentioned forecasting object piece the parts of the low resolution forecasting object piece that pixel count also lacks than this forecasting object piece; The low resolution prediction piece that hangs down the generation of differentiating the prediction piece generates parts; With the pixel is that unit subtracts the parts that calculation generates low resolution error piece to above-mentioned low resolution forecasting object piece and above-mentioned low resolution prediction piece; After low resolution error piece has carried out coding to this, by the coded data of above-mentioned low resolution error piece being deciphered the parts that generate the low resolution error piece of decoding; To be unit with the pixel to the low resolution error piece of above-mentioned decoding and above-mentioned low resolution prediction piece add restores the parts of above-mentioned low resolution piece, and the formation that above-mentioned low resolution prediction piece generates parts comprises: generate and the 2nd parts of predicting piece of the same pixel count of above-mentioned prediction piece; From the 2nd prediction piece, generate the parts of above-mentioned low resolution prediction piece according to above-mentioned transformation rule, the formation of above-mentioned generating apparatus further comprises: all differential pixel values in above-mentioned low resolution error piece are under zero the situation, the parts that above-mentioned the 2nd prediction piece is exported as above-mentioned prediction piece.

Wish that above-mentioned dynamic image decoding device further possesses: above-mentioned coded data is deciphered the generation parts of generated error piece; To be unit with the pixel to above-mentioned prediction piece and above-mentioned error block add generates decode block, then this decode block is outputed to the output block of above-mentioned holding member.

In addition, wish in dynamic image decoding device that amplifier unit uses the pixel of the piece after the pixel of the low resolution piece of above-mentioned generation and the decoding adjacent with the top of prediction piece and the left side, generate and above-mentioned prediction piece has the piece of pixel count.

In addition, wish in dynamic image decoding device that the formation of decoding unit comprises: thus the above-mentioned coded data of having imported is deciphered the parts that hang down the resolution error block decoding; Generate the low low resolution prediction piece of differentiating the prediction piece and generate parts; To be unit with the pixel to above-mentioned low resolution error piece and above-mentioned low resolution prediction piece add generates the parts of above-mentioned low resolution piece, and the formation that above-mentioned low resolution prediction piece generates parts comprises: generate and above-mentioned prediction piece has the parts that the 2nd of same pixel count is predicted piece; Rule according to the rules generates the parts of above-mentioned low resolution prediction piece from the 2nd prediction piece, it is under zero the situation that the formation of above-mentioned decoding unit further comprises whole differential pixel value in the above-mentioned low resolution error piece, the parts that above-mentioned the 2nd prediction piece is exported as above-mentioned prediction piece.

And then, moving image encoding decoding technique of the present invention is not only device, also can adopt the form of method or program, illustrate that below dynamic image encoding method, motion picture decoding method, dynamic image encoding and moving image translator that its feature is described have and dynamic image encoding device, and the identical effect of above-mentioned effect of dynamic image decoding device.

That is, it is the dynamic image encoding method of unit generation forecast piece with the piece that dynamic image encoding method of the present invention is to use the image after the decoding, comprises: be the image segmentation of the coded object of having imported the 1st segmentation procedure of a plurality of forecasting object pieces; The conversion of above-mentioned forecasting object piece and be encoded to have the low resolution piece that the pixel count than this forecasting object piece also lacks after, by this low coded data of differentiating piece is deciphered, generate the low generation step of differentiating piece of decoding; Use the image after the above-mentioned decoding, the low piece of differentiating of the above-mentioned decoding that has generated is enlarged into the amplification procedure that has the piece of same pixel number with above-mentioned forecasting object piece; Cut apart the above-mentioned piece that has amplified by the rule of cutting apart according to the rules, generate the 2nd segmentation procedure of a plurality of fritters; Use above-mentioned image and the above-mentioned fritter that has generated deciphered, generate the prediction steps of the prediction piece of this fritter; By merging the above-mentioned a plurality of prediction fritters that generated, the combining step of generation forecast piece according to the above-mentioned rule of cutting apart.

Motion picture decoding method of the present invention, being to use and deciphering image is the motion picture decoding method that unit predicts the generation of piece with the piece, comprise: the coded data of having imported is deciphered, generated decoding step with the low resolution piece that lacks than prediction piece pixel count; Use the above-mentioned image of having deciphered, the above-mentioned low resolution piece that has generated is enlarged into the amplification piece of the piece identical with above-mentioned prediction piece pixel count; Cut apart piece after this amplification by the rule of cutting apart according to the rules, generate the segmentation procedure of a plurality of fritters; Use above-mentioned image and the fritter after above-mentioned cutting apart deciphered, generate the prediction steps of the prediction fritter of this fritter; By merging the above-mentioned a plurality of prediction fritters that generated, the combining step of generation forecast piece according to the above-mentioned rule of cutting apart.

In addition, dynamic image encoding of the present invention makes that to possess that maintenance deciphered image and had with the piece be that the computer of holding member of dynamic image encoding of function of unit generation forecast piece is carried out following steps: is the image segmentation of the coded object of having imported the 1st dividing processing of a plurality of forecasting object pieces; The conversion of above-mentioned forecasting object piece and after being encoded to the low resolution piece that the pixel count than this forecasting object piece also lacks,, generate the low generation of differentiating piece of decoding and handle by this low coded data of differentiating piece is deciphered; Use the above-mentioned image of having deciphered, the low resolution of the above-mentioned decoding that has generated piece is enlarged into the processing and amplifying of the piece identical with above-mentioned forecasting object piece pixel count; Cut apart piece by the rule of cutting apart according to the rules, generate the 2nd dividing processing of a plurality of fritters through above-mentioned amplification; Use and above-mentionedly deciphered image and, generated the prediction processing of the prediction fritter of this fritter through the fritter of above-mentioned generation; By merging the above-mentioned a plurality of prediction fritters that generated according to the above-mentioned rule of cutting apart, the merging of generation forecast piece is handled.

Moving image translator of the present invention makes to be possessed maintenance and has deciphered image, with having with the piece is that the computer of holding member of dynamic image encoding of function of unit generation forecast piece is carried out following steps: the coded data of having imported is deciphered, generated the decoding processing of the low resolution piece that pixel count also lacks than the pixel count of prediction piece; Use the above-mentioned image of having deciphered, the low resolution piece through above-mentioned generation is enlarged into the processing and amplifying of the pixel count piece identical with above-mentioned prediction piece pixel count; Cut apart the piece that this has amplified by the rule of cutting apart according to the rules, generate dividing processing the light fritter; Use above-mentioned image and the fritter after above-mentioned cutting apart deciphered, generate the prediction processing of the prediction fritter of this fritter; By merging these a plurality of prediction fritters that generated according to the above-mentioned rule of cutting apart, the merging of generation forecast piece is handled.

To achieve these goals, dynamic image decoding device of the present invention is the piece with regulation, and thereby to be unit decipher the dynamic image decoding device of generation forecast piece to the coded data that becomes the decoding object, it is characterized in that possessing: the holding member that keeps having deciphered image; By the coded data of having imported being carried out entropy decoding, generate compare with the pixel count of above-mentioned decode block also less and and this decode block low resolution piece of on time, space, being in same position carried out the entropy decoding unit of the low resolution block message of coding; The low resolution block message that generates with above-mentioned entropy decoding unit is deciphered,, generated the low resolution block decoding parts of above-mentioned low resolution piece with reference to the above-mentioned image of having deciphered; Cut apart the low resolution piece that generates by above-mentioned low resolution block decoding parts by the rule of cutting apart according to the rules, generate the low resolution piece partition member of a plurality of low resolution blocks; Generate the low resolution piece partition member of a plurality of low resolution blocks; According to the image of decoding that is maintained in the above-mentioned holding member, with low resolution block, generate more than this pixel count that hangs down the resolution block and and hang down the resolution block is in the prediction fritter of same position on time, space prediction fritter generation parts with being somebody's turn to do by above-mentioned low resolution piece partition member generation; Merge the piece combining step of generation forecast piece by basis and the above-mentioned merging rule of cutting apart the corresponding regulation of rule by the prediction fritter of above-mentioned prediction fritter generation parts generation;

In addition, to achieve these goals, dynamic image encoding device of the present invention is with the block unit of regulation the image that becomes coded object to be encoded and the dynamic image encoding device of generation forecast piece, it is characterized in that possessing: the holding member that keeps having deciphered image; Is the image segmentation of having imported the code block segmentation parts of a plurality of coded object pieces; The pixel count of the coded object piece after being cut apart by above-mentioned code block segmentation parts by contrast is also few, and and this coded object piece is in the low resolution piece of same position and generates the low block message of differentiating on time, space, this low block message of differentiating is deciphered, with reference to the above-mentioned image of having deciphered, generate the low low piece of differentiating of decoding of differentiating piece of decoding and generate parts; Cut apart by the low low piece of differentiating of decoding of differentiating the generation of piece generation parts of above-mentioned decoding by the rule of cutting apart according to the rules, generate the low resolution piece partition member of a plurality of low resolution blocks thus; According to the image of decoding that is maintained in the above-mentioned holding member, with the low resolution block that generates with above-mentioned low resolution piece partition member, generation is more than the pixel count of this low resolution block, and differentiates block is in the prediction fritter of same position on time, space prediction fritter generation parts with hanging down; The prediction fritter that is generated the parts generation by above-mentioned prediction fritter is merged with the above-mentioned merging rule of cutting apart the corresponding regulation of rule by basis, the piece of generation forecast piece merges parts thus.

In order between dynamic image encoding device and dynamic image decoding device, to carry out high-precision predictive coding, hope in dynamic image encoding device, for compressed encoding the low resolution block message of the low resolution piece that also lacks than coded object piece prime number carry out entropy coding and send to dynamic image decoding device.Because low resolution piece pixel count also lacks than the pixel count of coded object piece, and hangs down the resolution block message and also lack than the information of forecasting in the past that needs in the prediction of coded object piece, thus can avoid the symbol weight of additional information to increase, thus improve treatment effeciency.

In addition, in dynamic image decoding device, wish, decipher the low piece of differentiating the block message from low the resolution, the low resolution piece that obtains is divided into a plurality of low resolution blocks, hang down resolution block and deciphered image according to each, generate more than the low resolution of the pixel count block pixel count, and and should lowly differentiate block is in same position on time, space prediction fritter, generation forecast piece by merging this prediction fritter.In this case, obtain the low block of differentiating by further cutting apart the low piece of differentiating, differentiate block and deciphered image and the generation forecast fritter according to low, obtain carrying out the prediction fritter of the high-precision prediction in the careful zone of moving image thus, can the generation forecast piece by merging this prediction fritter.That is,,, can improve treatment effeciency simultaneously Yi Bian carry out high-precision prediction on one side can use the few low resolution block message of symbol weight.

Wish in dynamic image decoding device of the present invention: above-mentioned entropy code translator is deciphered above-mentioned coded data by entropy, further generate presentation code object piece that is included in this coded data and the error block information of predicting the error of piece, above-mentioned dynamic image decoding device further possesses: by the error block information that is generated by above-mentioned entropy code translator is deciphered the error block decoding unit of generated error piece; According to merging the prediction piece of parts generation and the error block that generates by above-mentioned error block decoding unit, the decode block generation parts of generation decode block by above-mentioned.

Wish that dynamic image encoding device of the present invention further possesses: by being that unit asks the coded object piece of being cut apart by above-mentioned coded object piece partition member with the pixel, with the error that merges the prediction piece of parts generation by above-mentioned, the error block that generates the error block of this error of expression generates parts; By the error block that is generated the parts generation by above-mentioned error block is encoded and the error block addressable part of generated error block message; By the error block information that is generated by above-mentioned error block addressable part is deciphered, generate the error block decoding unit of decoding error block; According to by the prediction piece of above-mentioned merging parts generation and the decoding error block that generates by above-mentioned error block decoding unit, carry out the decode block generation parts that decode block generates.

In addition, wish that dynamic image encoding device of the present invention further possesses: to generating the low resolution block message of parts generation and the entropy coding parts that carry out entropy coding by the error block information that above-mentioned error block addressable part generates by the low piece of differentiating of above-mentioned decoding.

And the present invention can also catch as motion picture decoding method, dynamic image encoding method, moving image translator, dynamic image encoding, can be described below.They play and above-mentioned dynamic image decoding device, the same effect of dynamic image encoding device.

Motion picture decoding method of the present invention, be to have deciphered the holding member of image possessing maintenance, and thereby to be unit with the piece of regulation decipher motion picture decoding method in the dynamic image decoding device of generation forecast piece to the coded data that becomes the decoding object, it is characterized in that, comprise: by the coded data of the decoding object imported being carried out entropy decoding, generation is also lacked for the pixel count of pixel count than above-mentioned decode block, and and this decode block in the time, the low resolution piece that is in same position on the space has carried out the entropy decoding step of the low resolution block message of coding; The low resolution block message that generates in above-mentioned entropy decoding step is deciphered, generated the low resolution block decoding step of above-mentioned low resolution piece with reference to the above-mentioned image of having deciphered; The rule of cutting apart by regulation is segmented in the low resolution piece that generates in the above-mentioned low resolution block decoding step, generates the low resolution piece segmentation procedure of a plurality of low resolution blocks; According to the low resolution block of deciphering image, in above-mentioned low resolution piece segmentation procedure, generating that is maintained in the above-mentioned maintenance step, generate more than this low pixel count of differentiating block of pixel count, and and should lowly differentiate block is in the prediction fritter of same position on time, space prediction fritter generation step; Merge the piece combining step of generation forecast piece with the above-mentioned merging rule of cutting apart the corresponding regulation of rule to generating the prediction fritter that generates in the step by basis at above-mentioned prediction fritter.

Motion picture decoding method of the present invention, be to have deciphered the holding member of image possessing maintenance, thereby and be unit to encode dynamic image encoding method in the dynamic image encoding device of generation forecast piece of the image that becomes coded object with the piece of regulation, it is characterized in that, comprising: be the image segmentation of the coded object of having imported the code block segmentation step of a plurality of coded object pieces; By also lacking for the pixel count of pixel count than the coded object piece of in above-mentioned code block segmentation step, cutting apart, thereby and and this coded object piece low resolution piece of on time, space, being in same position encode and generate the low block message of differentiating, this low block message of differentiating is deciphered, deciphered image and generate the low low piece of differentiating of decoding of differentiating piece of decoding and generate step with reference to above-mentioned; The rule of cutting apart by according to the rules is segmented in the low piece of differentiating of above-mentioned decoding and generates the low piece of differentiating of the decoding that generates in the step, generates the low resolution piece segmentation procedure of a plurality of low resolution blocks; According to the low resolution block of deciphering image, in above-mentioned low resolution piece segmentation procedure, generating that is maintained in the above-mentioned holding member, generate pixel count and be somebody's turn to do more than the low resolution block, and differentiated block is in the prediction fritter of same position on time, space prediction fritter generation step with hanging down; Merge the piece combining step of generation forecast fritter with the above-mentioned merging rule of cutting apart the corresponding regulation of rule to generating the fritter in advance that generates in the step by basis at above-mentioned prediction fritter.

Moving image translator of the present invention, it is characterized in that, it makes to be arranged on possesses the holding member that image has been deciphered in maintenance, and being unit with the piece of regulation deciphers and computer on the dynamic image decoding device of generation forecast piece has following function the coded data that becomes the decoding object: carry out entropy decoding by the coded data to the decoding object imported, generate and to have carried out pixel count and also lack than the pixel count of above-mentioned decode block, and with this decode block in the time, be in the entropy decoding unit of low resolution block message of coding of the low resolution piece of same position on the space; The low resolution block message that is generated by above-mentioned entropy decoding unit is deciphered,, generated the low resolution block decoding parts of above-mentioned low resolution piece with reference to the above-mentioned image of having deciphered; By the rule of cutting apart according to the rules the low resolution piece that is generated by above-mentioned low resolution block decoding parts is cut apart, generated the low resolution piece partition member of a plurality of low resolution blocks; According to be maintained on the above-mentioned holding member decipher image, by the low resolution block that above-mentioned low resolution piece partition member generates, generate pixel count this low differentiate more than block and and should lowly differentiate block is in the prediction fritter of same position on time, space prediction fritter generation parts; By generating the prediction fritter that parts generate according to merging by above-mentioned prediction fritter with the above-mentioned merging rule of cutting apart the corresponding regulation of rule, the piece of generation forecast piece merges parts.

Dynamic image encoding of the present invention, it is characterized in that, it makes to be arranged on possesses the holding member that image has been deciphered in maintenance, thereby and have following function with the piece of the regulation computer that to be unit encode on the dynamic image encoding device of generation forecast piece to the image that becomes the decoding object: be the image segmentation of the coded object of having imported the code block segmentation parts of a plurality of coded object pieces; By also lacking than the coded object piece of cutting apart by above-mentioned coded object piece segmenting device for pixel count, and the low resolution piece that is in same position with this coded object piece on time, space is encoded, generate the low block message of differentiating, this low block message of differentiating is deciphered, with reference to the above-mentioned image of having deciphered, generate the low low piece of differentiating of decoding of differentiating piece of decoding and generate parts; Cut apart by the low low piece of differentiating of decoding of differentiating the generation of piece generation parts of above-mentioned decoding by the rule of cutting apart according to the rules, generate the low resolution piece partition member of a plurality of low resolution blocks; According to the low resolution block of deciphering image, generating by above-mentioned low resolution piece partition member that is maintained in the above-mentioned holding member, generate more than this low pixel count of differentiating block of pixel count, and and should lowly differentiate block is in the prediction fritter of same position on time, space prediction fritter generation step; By generating the fritter in advance that parts generate according to merging by above-mentioned prediction fritter with the above-mentioned merging rule of cutting apart the corresponding regulation of rule, the piece of generation forecast piece merges parts.

If employing the present invention is restrained because the symbol weight of additional information increases, so can improve the treatment effeciency in encoding process and decoding are handled.

Description of drawings

Fig. 1 is the figure of the structure of the dynamic image encoding device in the expression execution mode 1.

Fig. 2 is the figure of the structure of the dynamic image decoding device in the expression execution mode 1.

Fig. 3 is the figure that is used for the illustrated block coupling.

Fig. 4 is used for the figure that illustrated block is dwindled processing.

Fig. 5 is the 1st figure that is used for the illustrated block processing and amplifying.

Fig. 6 is the 2nd figure that is used for the illustrated block processing and amplifying.

Fig. 7 is used to illustrate 4 * 4 figure of cutting apart that amplify piece.

Fig. 8 is used to illustrate 2 * 2 figure of cutting apart that amplify piece.

Fig. 9 is the figure that is used to illustrate 4 * 4 fritter motion inference process.

Figure 10 is the figure that is used to illustrate 2 * 2 fritter motion inference process.

Figure 11 is the flow chart that is used for the illustrated block encoding process.

Figure 12 is used to illustrate the low flow chart that block encoding/decoding is handled of differentiating.

Figure 13 is the flow chart that is used to illustrate fritter motion inference process.

Figure 14 is used for the flow chart that illustrated block decoding is handled.

Figure 15 is used to illustrate the low flow chart that block decoding is handled of differentiating.

Figure 16 is that the explanation storage is used for realizing that by computer system moving image encoding is handled or the figure of the data storage medium of the program that moving image decoding is handled.

Figure 17 is the figure of structure of the dynamic image encoding of expression execution mode 1.

Figure 18 is the figure of structure of the moving image translator of expression execution mode 1.

Figure 19 is used to illustrate the figure that follows the piece coupling that piece cuts apart.

Figure 20 is the figure that is used to illustrate the 4th deformation form that relates to the piece size.

Figure 21 is the figure that is used for illustrating the piece processing and amplifying of the 5th deformation form.

Figure 22 is the figure of the structure of the dynamic image encoding device in expression the 5th deformation form.

Figure 23 is the figure of the structure of the dynamic image decoding device in expression the 5th deformation form.

Figure 24 is used for illustrating that the piece of the 6th deformation form dwindles the figure of processing.

Figure 25 is the figure of the structure of the dynamic image encoding device in expression the 8th deformation form.

Figure 26 is the figure of the structure of the dynamic image decoding device in expression the 8th deformation form.

Figure 27 is the figure that is used for illustrating the piece processing and amplifying of the 8th deformation form.

Figure 28 is the figure of the dynamic image decoding device in the explanation execution mode 2.

Figure 29 is the figure of the dynamic image encoding device in the explanation execution mode 2.

Figure 30 is the figure of the block decoding handling process in the expression execution mode 2.

Figure 31 is the figure of the low resolution block decoding handling process in the expression execution mode 2.

Figure 32 is the figure of the fritter prediction processing flow chart in the expression execution mode 2.

Figure 33 is the figure of the block encoding handling process in the expression execution mode 2.

Figure 34 is the figure that is illustrated in the low resolution block encoding/decoding handling process in the execution mode 2.

Figure 35 is the figure that is used to illustrate the block prediction processing.

Figure 36 is the figure that is used for the illustrated block pixel arrangement.

Figure 37 is the structure chart of the moving image translator of execution mode 2.

Figure 38 is the structure chart of the dynamic image encoding of execution mode 2.

Embodiment

[execution mode 1]

Below, with reference to description of drawings embodiments of the present invention 1.Fig. 1 is the figure of the structure of the expression dynamic image encoding device (below, abbreviate " code device 100 " as) that can realize predictive coding of the present invention.Equally, Fig. 2 is the figure of the structure of the expression dynamic image decoding device of realizing predictive interpretation involved in the present invention (below, be called " code translator 200 ").The inscape that these each devices have connects in the mode that can carry out the signal input and output mutually via bus.In addition, in two figure, mark prosign having on the inscape of same function.

The formation of code device at first, is described with reference to Fig. 1.Code device 100 is the coded object pieces (forecasting object piece) 202 that incoming frame 201 are divided into 16 * 16 pixels with code block segmentation unit 251, then, predict/apparatus for encoding cutting apart a plurality of coded object pieces (forecasting object piece) 202 that obtain according to the order of raster scan.

At this, so-called forecasting object piece is by incoming frame 201 is carried out piece is cut apart becomes next forecasting object in the piece that obtains piece.So-called coded object piece is by incoming frame 201 is carried out piece is cut apart becomes next coded object in the piece that obtains piece.In the present embodiment, because the object piece of the object piece of prediction processing and encoding process is arranged to same size, the forecasting object piece is consistent with the coded object piece.In addition, so-called prediction piece is the piece that generates from the pixel of past decoding by prediction processing.

Coded object piece (forecasting object piece) 202 at first is imported into the low piece generation unit 252 of differentiating of decoding.Decoding is low differentiates piece generation unit 252 and generates the low piece 110 of differentiating of decoding that pixel counts also lack than the pixel count of forecasting object piece 202, and the low block message 102 of differentiating.Infer in the unit 211 in motion, according to raster scan order input forecasting object piece 202.Motion is inferred unit 211 at the forecasting object piece of importing 202, and execution block matching treatment between a plurality of candidates prediction pieces 203 of extracting out in the decoded frame in the past from be stored in frame memory 153, detects 1 low piece prediction motion vector 104 of differentiating.

At this, mate with reference to Fig. 3 illustrated block.So-called piece coupling is to detect the technology of the motion vector of forecasting object piece.In the piece coupling, setting seek scope 522 on the decoded in the past decoded frame 520 in unit 211 is inferred in motion, retrieves in seek scope.Below, the difference evaluation of estimate of detection and forecasting object piece 202 is the position of the piece 105 of minimum.Then, from decoded frame and forecasting object piece 202 be in the locational imaginary piece 521 of the same space, the amount of movement till detected 105 extremely as motion vector 104.And, can be useful in as the difference evaluation of estimate pixel corresponding between the piece of the piece of detected object and coded object difference absolute value and, perhaps the scaled value of the symbol weight of motion vector the value on this difference absolute value etc. that adds.

Motion compensation units 133 generates with low and differentiates the piece prediction and predict piece 106 with the corresponding low resolution of motion vector 104.Specifically,, in the decoded frame 520 in the past from be stored in frame memory 153, obtain the pixel corresponding, and it is predicted piece 106 as low the resolution with the piece of 8 * 8 pixels based on the low piece predicted motion vector 104 of differentiating.Fig. 4 represents the corresponding relation of the low resolution piece 11 of the piece 10 of 16 * 16 pixel sizes and 8 * 8 pixel sizes.As shown in Figure 4, the low resolution piece 11 of 8 * 8 pixel sizes is to extract the piece of pixel out every 1 row in length and breadth from the piece of 16 * 16 pixel sizes.

At this, generate the low piece of differentiating by the pixel extraction and have the effect of raising in the estimated performance of fritter prediction processing described later.That is, because low each pixel of differentiating piece is keeping radio-frequency component, so code device 100 when carrying out motion detection, is caught careful feature easily.Its result, estimated performance improves.

For coded object piece (forecasting object piece) 202 also is with dwindling processing unit 134, according to the rule transformation shown in Figure 4 low resolution coded object piece (the low forecasting object piece of differentiating) 205 that to be pixel count also lack than the pixel count of coded object piece 202.213 pairs of each pixels in low resolution coded object piece (the low forecasting object piece of differentiating) 205 and low resolution prediction piece 106 of subtrator subtract calculation, generate low resolution error piece 206.

Converter unit 214 is implemented 4 * 4 discrete cosine transforms (DCT:Discrete CosineTransform) to these pieces of 4 * 4 after low resolution error piece 206 being divided into 44 * 4 piece.Then, generate and 16 conversion coefficients 207 that low resolution error piece is relevant respectively.215 pairs of conversion coefficients 207 of quantifying unit quantize, and generate the quantization transform coefficient 107 of low resolution error piece.The quantization transform coefficient 107 of merge cells 216 arrangement motion vectors 104 and low resolution error piece outputs to entropy coding unit 255 as the low block message 102 of differentiating.

136 pairs of quantization transform coefficients of inverse quantization unit 107 carry out inverse quantization, generate the decoding quantization transform coefficient 108 of low resolution error piece.Inverse transformation unit 137 pairs of decoding quantization transform coefficient 108 carries out anti-dct transform, generates the low resolution error piece 109 of decoding.Each pixel that adder unit 217 adds and differentiates in prediction piece 106 and the low resolution error piece 109 of decoding low generates the low piece 110 of differentiating of decoding.

Amplify piece generation unit 154 the low piece 110 of differentiating of decoding is enlarged into amplification piece 127 shown in Figure 5.Promptly, amplify in the decoding pixel of the present frame of piece generation unit 154 from be stored in frame memory 153, and become the top of the piece position of now forecast object, and the decoding pixel of the adjacent piece in the left side obtains as border decoding pixel group 123, as shown in Figure 5, be configured in decoding low differentiate piece 110 around.

Then, amplify the interpolate value that the pixel of representing with the black square is calculated in the interpolation computing shown in Figure 6 of piece generation unit 154 usefulness, generate and amplify piece 127.By this processing, the decoding of 8 * 8 block sizes is low differentiates that piece 110 is exaggerated and the amplification piece 127 of the prediction piece onesize (16 * 16) by prediction processing generation described later.And, be under the situation of end block of present frame when the piece of forecasting object, because adjacent piece does not exist, can't obtain border decoding pixel group 123 so amplify piece generation unit 154.In this case, replace it with the low boundary pixel of differentiating piece 110 of decoding.

Piece cutting unit 162 is shown in the piece 127a of Fig. 7, amplifying the fritter 112a that piece 127 is divided into 16 4 * 4.And then piece cutting unit 162 is shown in the piece 127b of Fig. 8, amplifying the fritter 112b that piece 127 is divided into 64 2 * 2.As 1 group, order is input to fritter predicting unit 152 to these divided each fritters 14 * 4 fritter 112a and 42 * 2 fritter 112b spatially being in same position with it.

Fritter predicting unit 152 is by constituting with the lower part: the fritter motion vector that mates the detection of the motion vector 115 that carries out fritter 112 by piece is inferred unit 163; The fritter motion vector 115 that use is detected generates the fritter motion compensation units 164 of the prediction fritter 116 of fritter 112.In the present embodiment, fritter predicting unit 152 uses these functions to generate the prediction fritter 116b of each 2 * 2 fritter 112b.

More particularly:

(1) the fritter motion infers that the motion vector of unit 163 execution 4 * 4 fritters detects processing.Below, in 16 4 * 4 fritter 112a, the function of the piece 112a1 of the upper end that special concern is shown in Figure 9 explanation fritter predicting unit 152, but for other 15 4 * 4 fritters, owing to also be the piece position difference on the space, thereby handle with same order.Fig. 9 is 4 * 4 fritters that are used for illustrating after being cut apart by 16, is positioned at the figure of detection method of motion vector 115a1 of 4 * 4 fritter 112a1 of left upper end.If the fritter motion infers that unit 163 is input to the fritter motion to 4 * 4 fritter 112a and infers unit 163, then the frame 520 (being kept in the frame memory 153) of decoding is gone up and is set seek scope 181a1 in the past.Then, retrieve in seek scope according to the rule that gives, the difference evaluation of estimate (summation of the difference absolute value of the pixel in fritter) of detection and 4 * 4 fritter 112a1 is the position of 4 * 4 fritter 116a1 of minimum, and it is predicted fritter 116a1 as 4 * 4.Then, fritter motion is inferred unit 163 spatially being in 4 * 4 117a1 on the decoded frame 520 of same position with 4 * 4 fritter 112a1, and the amount of exercise between 4 * 4 prediction fritter 116a1 that are detected is as the motion vector 115a1 of 4 * 4 fritter 112a1.

And seek scope 181a1 is on being arranged to predict with the position of motion vector 104 as the center with the low resolution piece behind the coding that uses in the generation of low resolution prediction piece as the size of predesignating.This is to consider the motion vector of 4 * 4 detected fritters near the low high cause of possibility of differentiating the piece prediction with motion vector 104, thereby has the effect that improves retrieval precision and can set little seek scope.Thereby, for the central point of the seek scope of 16 4 * 4 fritters that belong to same coded object piece too, be set at that expression is low differentiates the piece prediction with on the position of motion vector 104.

(2) the fritter motion infers that unit 163 detects the motion vector that belongs to same group 42 * 2 fritters with 4 * 4 fritter 112a1 that detect motion vector in (1).Below, in 42 * 2 fritters, the piece 112b1 of the upper end that special concern is shown in Figure 10 also describes.Figure 10 is 2 * 2 fritters that are used for illustrating after being cut apart by 64, is positioned at the figure of detection method of motion vector 115b1 of 2 * 2 fritter 112b1 of left upper end.The fritter motion is inferred unit 163 when 2 * 2 fritter 112b1 being input to fritter motion deduction unit 163, and Yi Ma frame 520 (being kept on the frame memory 153) is gone up and set seek scope 181b1 in the past.Then, rule is according to the rules retrieved in seek scope, and the difference evaluation of estimate (summation of the difference absolute value of the pixel in fritter) of detection and 2 * 2 fritter 112b1 is the position of 2 * 2 fritter 116b1 of minimum, and it is predicted fritter 116b1 as 2 * 2.Then, fritter motion infers that 163 of unit and 2 * 2 predict that fritter 112b1 spatially are in 2 * 2 117b1 on the decoded frame 520 of same position, and the amount of movement between 2 * 2 fritter 116b1 that are detected, as the motion vector 115b1 of 2 * 2 fritter 112b1.And then, for 32 * 2 remaining fritters, also use the same method and detect the motion vector of 2 * 2 fritters.

And seek scope 181b1 is as the size predesignated, and the motion vector 115a1 that is set in to belong to same group 4 * 4 fritter 112a1 with 2 * 2 fritter 112b1 is on the position at center.This is because consider that the motion vector of 2 * 2 detected fritters spatially is in same position, promptly, this motion vector is near belonging to the near high cause of possibility of motion vector of same group 4 * 4 fritters, can set for a short time effect to seek scope so have.Thereby, fritter motion infer unit 163 for belong to same group 42 * 2 fritters seek scope central point too, be set at on the represented position of the motion vector 115a1 of 4 * 4 fritters that spatially are in same position.

The motion vector 115b of 42 * 2 fritters that (3) are detected is imported into fritter motion compensation units 164.Fritter motion compensation units 164 obtains the prediction fritter 116b corresponding with the motion vector 115b of 2 * 2 fritters from decoded frame 520 (being kept at the frame memory 153).The processing of the predicting unit 152 of such fritter is carried out because of the data of only using code translator 200 to generate from coded data, so even also can access same treatment effect in code translator 200.That is, code translator 200 also can detect the motion vector same with code device 100.Thereby the motion vector of each 2 * 2 fritter does not need to encode.

Piece merge cells 155 generates the prediction piece 118 of 16 * 16 pixels thus by each of 64 2 * 2 prediction fritters is merged based on the original position of cutting apart before rule side by side becomes to cut apart in piece cutting unit 162.The difference value that subtrator 253 calculates each pixel in coded object piece 202 and prediction piece 118 generates error block 208.

254 pairs of error block 208 of error block coding unit generated error block message 119 of encoding outputs to entropy coding unit 255.As shown in Figure 1, error block coding unit 254 converter unit 218 and quantifying unit 219 as its inscape.Converter unit 218 carries out discrete cosine transform (DCT) with 4 * 4 units to error block 208, for each of 16 4 * 4, generates 16 conversion coefficients 209.219 pairs of each conversion coefficients 209 of quantifying unit quantize, generated error block message (for example, quantization transform coefficient) 119.

156 pairs of error block information 119 of error block decoding unit are deciphered and are generated decoding error block 121.As shown in Figure 1, error block decoding unit 156 inverse quantization unit 140 and inverse transformation unit 141 as its inscape.Inverse quantization unit 140 by carrying out the inverse quantization of 16 quantization transform coefficients 119, generates decoding quantization transform coefficient 120 for each of 16 4 * 4.Inverse transformation unit 141 generates decoding error block 121 by each decoding quantization transform coefficient of 4 * 4 120 is carried out anti-dct transform.

Each pixel in 157 pairs of predictions of adder unit piece 118 and the decoding error block 121 adds, and generates decode block 122.The decode block 122 that generates is stored in the frame memory 153, and its result has deciphered image and has been updated.Entropy coding unit 255 is by carrying out entropy coding and be multiplexed as coded data 101 low block message 102 and the error block information 119 differentiated.

Below, the action of code device 100 is described.Explanation constitutes each step of dynamic image encoding method of the present invention in the lump.Figure 11 is the flow chart of encoding process that is used to illustrate the object piece of predictive coding.And in the block encoding of Figure 11 was handled, expression block decoding described later was handled and its content and result are the symbols of shared step in the figure bracket.

At first, cut apart incoming frame 201 with code block segmentation unit 251 after, then code device 100 obtains coded object piece 202.In S1, the low piece generation unit 252 of differentiating of decoding is carried out low resolution block encoding shown in Figure 12/decoding processing.

In the S101 of Figure 12, import next coded object piece (forecasting object piece).In S102, motion infer unit 211 according to the forecasting object piece, be stored in the decoded frame of the past decoding in the frame memory, detect the low piece prediction motion vector of differentiating.And the detection method of motion vector is because narrate its explanation of Therefore, omited in the low formation of differentiating piece generation unit 252 of decoding illustrates.

In S103, motion compensation units 133 obtains from the decoded frame of frame memory and the low piece prediction corresponding low resolution prediction piece of motion vector of differentiating.In S104, dwindle the above-mentioned transformation rule of processing unit 134 based on reference Fig. 4 explanation, the forecasting object piece is transformed to the low forecasting object piece of differentiating.In S105, subtrator 213 is differentiated the forecasting object piece and is hanged down the calculation that subtracts of carrying out each pixel between the resolution prediction piece low, generates low resolution error piece.

In S106, converter unit 214 is after carrying out dct transform to low resolution error piece, and quantifying unit 215 is carried out the quantification of DCT coefficient.Its result generates the quantization transform coefficient that hangs down the resolution error piece.And the quantization accuracy setting in S106 must be also higher than the quantified precision in the S8 of Figure 11 described later.This is because the cause of such effect is arranged, and promptly improves the low image quality of differentiating piece by code device 100, can improve 2 * 2 little prediction piece accuracy of detection among S3, the S4.

In the S107 of Figure 12, the quantization transform coefficient of 137 pairs of low resolution error pieces of inverse quantization unit 136 and inverse transformation unit is deciphered (inverse quantization/inverse transformation).Its result generates the low resolution error piece of decoding.In S108, adder unit 217 carries out adding of each pixel low the resolution between prediction piece and the low resolution error piece of decoding, generates the low piece of differentiating of decoding.Then, merge cells 216 merges by the quantization transform coefficient to motion vector and low resolution error piece and generates the low block message (S109) of differentiating.

Return Figure 11, in S2, amplify piece generation unit 154, the low resolution of decoding piece is enlarged into and predicts the amplification piece 127 of piece onesize (16 * 16 pixel) as Fig. 5 and shown in Figure 6.For the generation method of amplifying piece, because in the function declaration that amplifies piece generation unit 154, be illustrated its explanation of Therefore, omited with reference to Fig. 5, Fig. 6.

In S3, as Fig. 7, shown in Figure 8, piece cutting unit 162 is divided into 4 * 4 fritters and 2 * 2 fritters amplifying piece.Each fritter after cutting apart as one group, is input to 14 * 4 fritter and 42 * 2 fritters spatially being in same position with it the fritter motion and infers unit 163.

In S4, the fritter motion infers that unit 163 detects the motion vector of 42 * 2 fritters with above-mentioned group of unit.Fritter motion inference process is carried out respectively 14 * 4 fritter and 42 * 2 fritters.Earlier 4 * 4 fritters are carried out processing according to this order, thereafter, 42 * 2 fritters execution are handled.

At this, fritter motion inference process is described with reference to Figure 13.At first, import fritter in S41, evaluation of estimate E min is set to max-int (initial value) (S42).Then, in S43, judge the size of fritter.Carry out and this result of determination corresponding treatment S 441 or S442 thereafter.That is, on the decoded frame 520 (being kept on the frame memory) in the past, set the size of seek scope and the center (retrieval center point) of seek scope.

Then, begin candidate according to the prediction fritter of spiral ordered retrieval in the seek scope of decoded frame 520 from this central point, its result, specific above-mentioned evaluation of estimate E min is minimum prediction fritter.Specifically, the prediction fritter 113 (S45) of next candidate is obtained in fritter motion deduction unit 163 from the decoded frame of frame memory 153, calculate evaluation of estimate (S46).Evaluation of estimate E is the summation of the absolute value of the pixel difference value between fritter and prediction fritter.

In S47, judge whether evaluation of estimate E is also littler than evaluation of estimate E min.When evaluation of estimate E than the little situation of evaluation of estimate E min under (S47:YES), when evaluation of estimate E min was updated to evaluation of estimate E, the motion vector of fritter was updated to from the position of fritter the amount of movement (S48) of position of the present candidate piece of prediction fritter.On the other hand, when evaluation of estimate E under evaluation of estimate E min or above situation (S47:NO), omit the processing of S48, transfer to S49.

In S49, judge whether the retrieval of the prediction fritter of the whole candidates in seek scope finishes.After, till the retrieval that is judged as the prediction fritter of the whole candidates in seek scope in S49 finishes, repeat a series of processing of S45～S49.Then, be judged as the moment that above-mentioned retrieval finishes, fritter motion inference process finishes.

Return Figure 11, in S5, fritter motion compensation units 164 obtains 2 * 2 prediction fritters 116 corresponding with the motion vector of detected 42 * 2 fritters in S3 and S4 respectively from the decoded frame of frame memory 153.Each of S3～S5 handled 16 groups fritter carried out.

In S6, piece merge cells 155 predicts that fritters are arranged in the position before cutting apart, and merge into the prediction piece of 16 * 16 pixels thus to 64 2 * 2 by based on the rule of using of cutting apart in S3.In S7, subtrator 253 carries out the calculation that subtracts of each pixel, generated error piece between coded object piece and prediction piece.In S8,254 pairs of error block of error block coding unit are encoded (transform/quantization), generated error block message (for example, quantization transform coefficient).

In S9,156 pairs of above-mentioned error block information of error block decoding unit are deciphered (inverse quantization/inverse transformation), generate the decoding error block.In S10, adder unit 157 carries out each pixel and adds between prediction piece and decoding error block, generate decode block.If the decode block that is generated is stored in the present decoded frame of frame memory 153 (S11), then 255 pairs of entropy coding unit are low differentiates block messages and error block information is carried out entropy coding (S12), and the block encoding processing finishes.

Below, the structure of code translator shown in Figure 2 200 is described.Code translator 200 is that the prediction processing of execution block unit and the decoding of coded data 101 are handled, and with the device of block unit according to raster scan order playback decode block 122.In the present embodiment, prediction piece that generates by prediction processing and decode block big or small identical all is arranged to 16 * 16 pixels.

After the input of the coded data 101 that entropy decoding unit 150 needs,, coded data 101 is separated into low block message 102 and the error block information 119 differentiated by entropy decoding in the generation that is received in as the decode block of deciphering object.Hang down resolution block decoding unit 151 and pass through to decipher low resolution error information 102, the generation pixel count hangs down resolution piece 110 than the decoding of 8 * 8 pixels that the pixel count of predicting piece also lacks.

Separative element 131 is separated into low piece prediction motion vector 104 and the low resolution error block message (for example, hanging down the quantization transform coefficient of resolution error piece) 107 differentiated hanging down resolution block message 102.In the decoded frame 520 of motion compensation units 133 from be stored in frame memory 153, obtain with low and differentiate the piece prediction with the corresponding low resolution prediction piece 106 of motion vector 104.

The quantization transform coefficient 107 of 136 pairs of low resolution error pieces of inverse quantization unit carries out inverse quantization to be handled, and generates the decoding quantization transform coefficient 108 of low resolution error piece.Inverse transformation unit 137 carries out anti-dct transform by decoding is quantized variation coefficient 108, generates the low resolution error piece 109 of decoding.Adder unit 135 calculates the addition calculated value of differentiating each pixel in prediction piece 106 and the low resolution error piece 109 of decoding low, generates the low piece 110 of differentiating of decoding.

And each function that amplification piece generation unit 154, piece cutting unit 162, fritter predicting unit 152 and piece merge cells 155 have is identical with the explanation of code device 100 with result.Thereby its explanation and the diagram that is used to illustrate are omitted.

In addition, 156 pairs of error block information of error block decoding unit (for example, quantization transform coefficient) 119 decipher and generate decoding error block 121.Error block decoding unit 156 has inverse quantization unit 140 and inverse transformation unit 141.140 pairs of quantization transform coefficients of inverse quantization unit 119 carry out inverse quantization, generate decoding quantization transform coefficient 120.Inverse transformation unit 141 pairs of decoding quantization transform coefficient 120 carries out anti-dct transform, generates decoding error block 121.

Adder unit 157 calculates for each pixel of prediction piece 118 and deciphers the addition calculated value that error block 121 spatially is in pixel He this pixel of consistent location, generates decode block 122 thus.Decode block 122 is upgraded the image of decoding that is stored in the frame memory 153.

Then, the action of code translator 200 is described, explanation constitutes each step of motion picture decoding method of the present invention in the lump.And, in the block decoding of Figure 14 is handled, represent the symbol of the step that block encoding processing shown in Figure 11 and its content and result are shared in the drawings in the bracket.

After the decoding that begins next piece is handled, then at first in T1, the coded data of 1 amount that needs is input to entropy decoding unit 150 (T1) in the decoding of this piece.In T2, the coded data of 150 pairs of inputs of entropy decoding unit is carried out entropy decoding, obtains low block message and the error block information differentiated.In these information, the low block message of differentiating handles 404 by the low block decoding of differentiating, and is imported into the low block decoding unit 151 of differentiating, and is interpreted as the low piece of differentiating of decoding.

At this, handle with reference to the low resolution block decoding of Figure 15 explanation in T3.In T31, the low block message of differentiating is input to separative element 131.Separative element 131 is separated into motion vector and low resolution error block message to the low block message of differentiating in T32.In T33, motion compensation units 133 obtains with low and differentiates the piece prediction and predict piece 106 with the corresponding low resolution piece of motion vector 104 from the decoded frame 520 (being kept at the frame memory 153) of past decoding.

In T34, the quantization transform coefficient 107 of the low resolution error block message of 136 pairs of conducts of inverse quantization unit carries out inverse quantization to be handled, and generates decoding quantization transform coefficient 108.And then 137 pairs of these decoding quantization transform coefficients 108 in inverse transformation unit carry out anti-dct transform, generate the low resolution error piece 109 of decoding.In T35,135 pairs of low each pixels of differentiating prediction piece 106 and the low resolution error piece 109 of decoding of adder unit add, and generate the low piece 110 of differentiating of decoding.

Return Figure 14,, amplify the amplification piece 127 (T4) that piece generation unit 154 is enlarged into the low resolution of decoding piece 110 and predict piece onesize (16 * 16 pixel) as Fig. 5 and shown in Figure 6.As the generation method of amplifying piece, because in the function declaration that amplifies piece generation unit 154, be illustrated its explanation of Therefore, omited with reference to Fig. 5, Fig. 6.

In T5～T7, piece cutting unit 162 and fritter predicting unit 152 are divided into 64 2 * 2 fritters amplifying piece, generate 64 2 * 2 prediction fritters.And, each of the T5 of Figure 14, T6 handle because with each of S3, S4 shown in Figure 11 handle identical respectively, so omit its detailed description.In addition, the processing of T7 is because identical with the processing of the S5 of Figure 11, thereby its explanation omission.

In T8, piece merge cells 155 predicts that fritters are arranged in the original position before cutting apart to 64 2 * 2 that generate based on the rule of cutting apart in T5 in T6 and T7.Thus, above-mentioned prediction fritter is merged into the prediction piece of 16 * 16 pixels.

In T9,156 pairs of error block information of error block decoding unit (for example quantization transform coefficient) are deciphered (inverse quantization/inverse transformation), generate the decoding error block.In T10, each pixel of adder unit 157 pairs of predictions piece and decoding error block adds, and generates decode block.In T11, decode block is kept in the frame memory 153, its result upgrades and has deciphered image.

Figure 16 (a) illustration stores the dynamic image encoding that the image encoding that is used for realizing present embodiment is handled, and the physical format of the floppy disk of the moving image translator that is used to realize that same image decoding is handled.In addition, front and the outward appearance and the cross-section structure of the above-mentioned floppy disk of Figure 16 (b) expression.Floppy disk FD is built in the shell F, and inwardly week forms plurality of tracks Tr with concentric circles from periphery on the surface of this dish.These magnetic tracks are divided into 16 fan-shaped Se according to angle direction.On the zone of being cut apart on the floppy disk FD, said procedure as data record.

When to floppy disk FD record said procedure, shown in Figure 16 (c), undertaken by floppy disk FDD from computer system Cs.When constructing code device 100 or code translator 200 with the program in the floppy disk FD in computer system Cs, floppy disk FDD reads said procedure from floppy disk FD, be transferred among the computer system Cs.

Figure 17 is the figure of the structure of expression dynamic image encoding 261 of the present invention.Dynamic image encoding 261 is recorded on the recording medium 260.Recording medium 260 for example is above-mentioned floppy disk, but also can be CD and IC-card, CD-ROM, DVD, perhaps semiconductor memory.In computer system Cs, be not limited to PC (Personal Computer: personal computer), also comprise and possess CPU, utilize software to differentiate the DVD player of information processing and control, set-top box, mobile phone etc.

Below, the structure of dynamic image encoding of the present invention and moving image translator is described.Dynamic image encoding 261 is stored in the program storage area 260a that is formed on the recording medium 260 as shown in figure 17.The primary module 261a that dynamic image encoding 261 is handled the Comprehensive Control moving image encoding; The coded object piece is cut apart module 261b; The low piece generation module 261c that differentiates of decoding; The low piece amplification module 261d that differentiates; Bulk is cut apart module 261e; Fritter prediction module 261f; Piece merges module 261g to have as the unit.By carrying out the code block segmentation unit 251 of function that these modules realize and above-mentioned dynamic image encoding device 100; The low piece generation unit 252 of differentiating of decoding; Bulk generation unit 154; Piece cutting unit 162; Fritter predicting unit 152; Each function of piece merge cells 155 is identical respectively.

Moving image translator 161 is stored in the program storage area 160a that is formed on the recording medium 160 as shown in figure 18.The primary module 161a that moving image translator 161 is handled the decoding of master control moving image; Entropy decoding module 161b; The low block decoding module 161c that differentiates; The low piece amplification module 161d that differentiates; Amplify piece and cut apart module 161e; Fritter prediction module 161f; Piece merges module 161g to have as the unit.By carrying out the entropy decoding unit 150 of function that these each modules realize and above-mentioned dynamic image decoding device 200; The low block decoding unit 151 of differentiating; Amplify piece generation unit 154; Piece cutting unit 162; Fritter predicting unit 152; Each function of piece merge cells 155 is identical respectively.

And, the formation of dynamic image encoding 261 can be its part or all transmit via the transmission medium of communication line etc., receive and write down (comprising installation) with other equipment.For moving image translator 161 too, it constitutes also can be that its part or all transmission medium via communication line etc. transmits, and is received and record (comprising installation) by other equipment.

And the form of recording and narrating in the above-mentioned execution mode is a preferred example of moving image encoding/decoding technique of the present invention, and the present invention is not limited to such form.

(the 1st deformation form)

This form relates to the motion vector detecting method of fritter.In the above-described embodiment, be with in 4 * 4 fritters for ± 32, in 2 * 2 fritters be ± 1 seek scope during as the operation vector detection of fritter.But also can be different seek scopes.In addition,, be not only integer value, can also be arranged to real number value for the precision of the motion vector composition of fritter.1/2 pixel, 1/4 pixel etc. are arranged in real number value, as vectorial composition the computational methods of the predict pixel under the situation of real number value, for example can be useful in H.264 the method for recording and narrating in (" Text of ISO/IEC 14496-10Advanced Video Coding 3rd Edition ", September 2004).

In addition, in the above-described embodiment, the central point of the seek scope when detecting for the motion vector at fritter is made as the prediction of low resolution piece and uses motion vector in 4 * 4 fritters, be made as the motion vector of 4 * 4 fritters that spatially are in same position in 2 * 2 fritters.But,, then be not limited thereto if can well-determinedly be worth in coding side and decoding side.For example, also can be set at the central point of seek scope 0 vector, by method selected predicted vector from the motion vector of adjacent isles of median prediction etc.

And then, though the evaluation of estimate during the motion vector that detects fritter as the absolute value of pixel difference value and, be not limited to this, so long as can well-determined evaluation method get final product in coding side and decoding side.For example, also can be to the quadratic sum of pixel difference value, perhaps with the weighted absolute value and the distortion such as (quadratic sums) of the corresponding pixel difference value of location of pixels.Though the decoded frame of forecasting object is defined as 1 frame, also can be a plurality of decoded frame as object, generation forecast fritter from the different decoded frame of each fritter.At this moment, if as evaluation of estimate use the pixel difference value absolute value and etc., in the coding side with decipher the identical evaluation of estimate of side, then do not need the selection information of decoded frame is encoded, be high efficiency.

(the 2nd deformation form)

This form relates to Forecasting Methodology.As Forecasting Methodology, limit and illustrated prediction, even but in the prediction of the direction in space of having deciphered of deciphering image resolution in having used the identical frame of coded object piece or decode block, also can fit and use the same method at time orientation.That is, also can use intra-prediction to amplify the low piece of differentiating of above-mentioned decoding.More particularly, the low low resolution piece of differentiating piece of the decoding that 200 pairs of code translators have been cut apart is again cut apart, the adjacent pixel of having deciphered in using same frame, for example from H.264 (" Text of ISO/IEC 14496-10Advanced Video Coding 3rd Edition ", September 2004.) a plurality of intra-prediction mode in determined best pattern, the situation of the piece of the high-resolution signal that prediction is suitable with the low resolution of this decoding piece etc. is also contained among the present invention.In this case, because do not need to transmit the predictive mode that respectively dwindles block expressly, can be so play with a spot of supplementary effect of the excellence of forecasting object piece accurately.In addition, make up the prediction of above-mentioned time orientation and the prediction of direction in space and also can carry out high-precision prediction.For example, carry out the low processing and amplifying of differentiating piece of the decoding of adopting intra-predictions at amplification piece generation unit 154, this amplification piece is divided into fritter, the method for the processing of execution fritter predicting unit 152 is also effective.

(the 3rd deformation form)

This form relates to the coding of error block.To the encoding process of the error block between prediction piece and the coded object piece not necessarily.That is, can error of omission block encoding unit 254 (Fig. 1) and error block decoding unit 156 (Fig. 1 and Fig. 2).In addition, also can determine whether coding at each coded object piece or decoding object piece.For error block, and the concrete coding/decoding method of low resolution error piece, be not limited to above-mentioned method.And then in the above-described embodiment, being the coded object piece is set to identically with the forecasting object block size, but also can realize Forecasting Methodology of the present invention with different sizes.In addition, the coding of error signal is not limited to block unit, also can carry out with picture unit.But, in this case, should prepare the decoding pixel of differentiating the block boundary that uses in the processing and amplifying of piece low, and need before carry out the low processing and amplifying of differentiating piece, carry out the generation processing of adjacent prediction piece and the decoding of predictive error signal and handle.

(the 4th deformation form)

This form relates to block size.In the above-described embodiment, the low motion vector of differentiating piece prediction usefulness being set respectively is 1 to coded object piece and decoding object piece.But, as shown in figure 19, the coded object piece is divided into a plurality of, for being the method that unit implements the piece coupling with the block, Forecasting Methodology of the present invention also can be suitable for.In this case, be when unit determines the size of fritter with the coded object piece, (104c 104d) encodes for 104a, 104b to the size information of block and a plurality of motion vector.

And then in the example of Figure 19, the central point of the seek scope when the motion vector of 4 * 4 fritters detects can be set in from 4 low motion vectors of piece prediction usefulness of differentiating to the position that the motion vector that spatially is in same position is represented.

And, in Figure 19, be that dividing method is arranged to 8 * 8 units, but be not limited to this, can utilize H.264 such dividing method shown in (" Text of ISO/IEC 14496-10 AdvancedVideo Coding 3rd Edition ", September 2004.) adaptively.In addition, for the size of the size of coded object piece and decode block, be not limited to 16 * 16.That is, even different block sizes such as 32 * 32,8 * 8,4 * 4 also can be suitable for the present invention by the size that correspondingly changes fritter according to its size.

The size of the fritter in fritter predicting unit 152 (S3 of Fig. 6 and S4) also is not limited to 4 * 4 and 2 * 2.For example, as shown in figure 20 5 * 5 fritter 112c and 3 * 3 fritter 112d are such, by making vectorial fritter overlapped, also can amplify block sizes.Little block size is because exist because of making the estimated performance downward trend under the situation that does not have feature in piece, so by adopting this form, estimate that the estimated performance of local shape improves.And the method that block sizes is switched adaptively with 5 * 5 and 4 * 4 and 3 * 3 and 2 * 2 size is also effective.The same with it, the block size when carrying out transform/quantization is also unqualified.That is, code device 100 and code translator 200 are that unit carries out transform/quantization to error block with 8 * 8 and 16 * 16 for example.

(the 5th deformation form)

This form relates to the low amplification method of differentiating piece of decoding.Be made as in the above-described embodiment: amplify piece generation unit 154 from the decoding pixel of the piece adjacent with the left side, obtain the border decoding pixel group 123 of reference in order to amplify decoding low resolution piece with the top of prediction piece.But, as shown in figure 21, also can hang down and obtain decoding pixel group 123 in border the decoding pixel of differentiating piece from the decoding of adjacent piece, generate and amplify piece.

Figure 22, Figure 23 represent to be used to realize the code device 300 and the code translator 400 of above-mentioned deformation form respectively.Code device 300, code translator 400 and code device 100, code translator 200 are different in the following areas.

(i) the low piece 110 of differentiating of decoding temporarily is kept in the frame memory 153,

(ii) merging into the low piece of differentiating of decoding as border decoding pixel with the low decoding pixel of differentiating piece of the spatially corresponding decoding of adjacent block,

(iii) the low resolution piece 124 (with reference to Figure 21) that comprises border decoding pixel as unit, be input to and amplify piece generation unit 154,

By formation code device 300 like this and code translator 400, can change border decoding pixel.

And the feature of each device illustrated in figures 1 and 2, Figure 22 and each feature of installing shown in Figure 23 can make up aptly according to condition.For example, each device the low resolution piece of adjacent piece with the identical situation of the piece of motion vector and present forecasting object under, the low decoding pixel of differentiating piece of decoding is used as border decoding pixel.In different situations, the decoding pixel of decoded frame is used as border decoding pixel.In addition, also can switch aptly with block unit in the coded system of a plurality of Forecasting Methodologies, as condition, pixel is deciphered on the change border the Forecasting Methodology of the adjacent block of present forecasting object piece.

And then, also can not use border decoding pixel, and as handling, only from the low pixel of differentiating in the prediction piece, generate and amplify piece at frame one end.In addition, in the above-described embodiment, be to extract decoding pixel in border the image out, but also can use the decoding pixel of the decoded frame in the past that spatially is in same position from deciphering of present frame.Decoding pixel in border is made as the piece adjacent pixels with forecasting object, but if the pixel of having deciphered then is not limited to this, can use the decoding pixel of leaving block boundary, perhaps uses the pixel that generates from a plurality of decoding pixels etc.

(the 6th deformation form)

This form relates to dwindles method.In the above-described embodiment, as shown in Figure 4,, be not limited to the method though, adopted on vertical/transverse direction and carried out rejecting at interval the method for handling with 1 between-line spacing as the processing of dwindling of piece.For example, the low pixel count of piece of differentiating can be set pixel count for than decode block (coded object piece) as long as pre-determine between code device 100 and code translator 200, the also few arbitrary shape of pixel count.For example, for the piece of 1 pixel that each pixel of low resolution prediction piece 106 shown in Figure 4 is staggered on direction in length and breadth and as the low resolution of Figure 24 predict the piece 1061 piece of pixel count being arranged to half, also can be suitable for the present invention.

In addition, in the above-described embodiment, be only to realize dwindling of piece, but also can reject processing implementing the laggard between-line spacing of low-pass filtering treatment to the pixel of piece with rejecting at interval to handle.But,,, also need to implement same low-pass filtering treatment for becoming the candidate prediction piece 113 of inferring unit 163 inputs to the fritter motion in order to improve the motion retrieval precision.

(the 7th deformation form)

This form relates to the low piece prediction of differentiating.In the above-described embodiment, each device is directly to obtain low the resolution to predict piece from frame memory 153, but also can obtain from the decoded frame 520 in the past being stored in frame memory 153 in order to generate the low prediction piece 1061 needed data 15 of differentiating., in motion compensation units 133 inside, generate the prediction piece 105 of shown in Figure 3 16 * 16 pixel sizes, also can be transformed to prediction piece 105 the low resolution prediction piece 106 of 8 * 8 pixel sizes thereafter.

In addition, in the above-described embodiment, if being input to motion to the forecasting object piece of 16 * 16 pixels, each device infers unit 211, but the low piece of differentiating is predicted the described method of present embodiment that is not limited to, for example, also can be the low resolution piece of the transformation rule of the forecasting object piece having been implemented Fig. 4 as above-mentioned input.In this case, with motion vector the low prediction piece of differentiating is diminished with the low predicated error of differentiating coded object piece (the low forecasting object piece of differentiating), have the effect of the symbol weight of the low resolution error piece of reduction by the prediction of selected low resolution piece.

(the 8th deformation form)

This form relates to the coding of many predictive modes.In the above-described embodiment, only the Forecasting Methodology of coded object piece (decode block) as Forecasting Methodology of the present invention.But also can be arranged to coded object piece (decode block) is unit, selects as recording and narrating in the background technology the block predication in the past that a plurality of motion vectors are encoded and relate to one of Forecasting Methodology of the present invention aptly.

And then, in above-mentioned Forecasting Methodology, establish with all pixels in 16 * 16 of the Forecasting Methodology of the present invention predictions.But, also can cut apart 16 * 16, be that unit switches Forecasting Methodology of the present invention aptly and used the inter-frame prediction method of low resolution piece prediction with motion vector with the block.For example, Figure 25 is illustrated in 16 * 16 coded object pieces (prediction piece) is divided into 48 * 8, is that unit carries out under the situation of adaptive prediction with 8 * 8, can carry out the code device 500 of this enforcement.Figure 26 represents same code translator 600.Figure 27 is illustrated in low piece and the amplification piece differentiated of decoding in this form.

The code device 500 of Figure 25 is different in the following areas with code device 100 shown in Figure 1.The 1st, amplifying piece generation unit 1541 is that unit carries out the low processing and amplifying of differentiating piece of decoding with 8 * 8.That is, amplifying piece generation unit 1541 is that unit expands to 8 * 8 extension blocks to the low piece of differentiating of decoding with 4 * 4, outputs to piece cutting unit 162.Amplify piece generation unit 1541 as shown in figure 27, the low piece 110 of differentiating of 8 * 8 decodings is divided into 44 * 4 low piece 110a, 110b, 110c, the 110d of differentiating.Then, be that unit merges border decoding pixel group 128a～128d with 4 * 4, be enlarged into 8 * 8 and amplify piece 1271.8 * 8 amplification pieces that are exaggerated like that, are split into 4 * 4 fritters and 2 * 2 fritters at piece cutting unit 165 shown in piece 1271a and piece 1271b.Then, be imported into fritter predicting unit 152.

And, because in border decoding pixel group 128a～128d, use the decoding pixel, so in this adaptive prediction mode, need be the encoding process that unit carries out predicated error with 8 * 8.But in existing coded system, for common 8 * 8 being the encoding process that unit carries out predicated error, a handoff predictions is handled and the executable unit of error coding processing, also can be corresponding with above-mentioned adaptive prediction mode.

The 2nd, identifying unit 256 is the selection that unit carries out Forecasting Methodology and prediction piece with 8 * 8.Use low 16 * 16 prediction pieces 105 that the piece prediction is read from frame memory 153 with motion vector 104 of differentiating to identifying unit 256 input forecasting object pieces 202 and motion compensation units 1331.Thereafter, these pieces are split into the low resolution prediction piece that 48 * 8 coded object pieces and 4 have cut apart behind 16 * 16 prediction pieces respectively and select to predict pieces with 8 * 8.

On the other hand, 8 * 8 prediction pieces 118 that generate little block predication of the present invention to identifying unit 256 inputs from piece merge cells 155.256 pairs 2 kinds of identifying units prediction piece, with 8 * 8 be that unit calculates and the pixel difference absolute value of coded object piece and, selective value are little 8 * 8 predict pieces, merge into the prediction piece 126 of 16 * 16 pixels.Then, the selection information 125 of the prediction mode of selecting is outputed to merge cells 2161.

The 3rd, merge cells 2161 merges selection information 125 and the low block message 1021 of differentiating.Select to have used under the situation of low resolution piece prediction with the prediction mode of motion vector 104 when selection information 125, in the low resolution of the decoding piece of 8 * 8 block sizes, do not need the information in 4 * 4 corresponding zones.Therefore, under these circumstances, because the quantization transform coefficient of 4 * 4 corresponding low resolution error pieces all is zero, so do not merge with the low block message 1021 of differentiating.

The code translator 600 of Figure 26 is different in the following areas with code translator 200 shown in Figure 1.The 1st, amplifying piece generation unit 1541 is that unit carries out the low processing and amplifying of differentiating piece of decoding with 8 * 8.This method is because as with Figure 27 explanation, so detailed in decoding one side, also need be carried out the decoding of 8 * 8 corresponding blocks of prediction and be handled, and decipher 8 * 8 decode block in the moment that has generated 8 * 8 prediction pieces.

The 2nd, switch 159 switches 2 kinds of prediction mode according to from the low block message 1021 isolated selection information 125 of differentiating.But, select to have used when selection information 125 under the situation of prediction mode of motion vector 104, differentiate in the piece in that 8 * 8 decodings are low, in the low information that does not comprise 4 * 4 corresponding zones in the block message 1021 of differentiating.That is, the quantization parameter of low resolution error piece all is zero.Therefore,, do not carry out corresponding 4 * 4 low decodings of differentiating piece and handle for this zone, and fritter prediction processing of the present invention.

And, in the 8th distortion example, also can be useful in the piece amplification method that above-mentioned the 5th distortion illustrates in the example, the Forecasting Methodology according to the adjacent block of forecasting object piece can adapt to ground handoff block amplification method.

[execution mode 2]

Below, with reference to the description of drawings embodiments of the present invention.

[apparatus structure]

Figure 28 represents to realize the structure of the dynamic image decoding device (hereinafter referred to as " code translator ") of predictive interpretation of the present invention, and Figure 29 represents to realize the structure of the dynamic image encoding device (hereinafter referred to as " code device ") of predictive coding of the present invention.And, suppose that the processing section that is marked with duplicate numbers in Figure 28 and Figure 29 acceptance of the bid has same function.

[structure of code device]

The code device of Figure 29 is incoming frame 201 to be divided into 8 * 8 coded object piece 202 in code block segmentation unit 251, according to raster scan order a plurality of coded object pieces 202 that obtain in cutting apart is carried out apparatus for encoding.The same with execution mode 1, even because also be in the present embodiment the block size of the coded object piece of the forecasting object piece of forecasting object and coded object is arranged to identical, so the forecasting object piece is consistent with the coded object piece.In addition, so-called prediction piece is the piece that generates by prediction processing from the pixel of past decoding.And the forecasting object piece is different with the block size of coded object piece also passable.Coded object piece 202 at first is transfused to the low piece generation unit 252 of differentiating of decoding.

Decoding is low, and to differentiate piece generation unit 252 are the processing units that generate " the low piece 110 of differentiating of decoding that pixel count also lacks than the pixel count of coded object piece (forecasting object piece) 202 " and " the low block message 102 of differentiating ".At this, the low action of differentiating piece generation unit 252 of decoding is described.Infer in the unit 211 in motion, according to raster scan order input coding object piece (forecasting object piece) 202, motion infers that unit 211 at the coded object piece of having imported (forecasting object piece) 202 be maintained between a plurality of candidates prediction pieces 203 in the frame memory 153 and implement the piece matching treatment, detects 1 motion vector 104.The detected motion vector 104 of motion vector prediction/coding unit 212 input calculates each is become differential motion vector 103 after score value deducts the value of each composition of predicted vector.Predicted motion vector by the left side of current block, just go up, adjacent 3 motion vector of upper right (under frame situation in addition, being upper left) predicts that according to median (at every kind of composition) calculates as object.In the decoded frame of having deciphered 520 of motion compensation units 133 from be maintained at frame memory 153, obtain the piece 105 corresponding with motion vector 104.Piece 105 is such as shown in figure 36 by dwindling processing unit 134, is transformed to the low resolution prediction piece 106 of having extracted pixel in length and breadth every 1 row out.And then coded object piece (forecasting object piece) 202 too, is transformed to the low resolution coded object piece (the low forecasting object piece of differentiating) 205 that pixel count also lacks than coded object piece (forecasting object piece) 202 pixels with dwindling processing unit 134.Low each pixel of differentiating coded object pieces (the low forecasting object piece of differentiating) 205 of 213 pairs of subtrators are calculated and low prediction the piece 106 spatially pixel of position consistency and the difference value of this each pixel differentiated, and generate low resolution error piece 206.214 pairs of low resolution error pieces 206 of converter unit carry out 4 * 4 discrete cosine transforms (DCT), generate the conversion coefficient 207 of 16 low resolution error pieces.215 pairs of conversion coefficients 207 of quantifying unit quantize, and generate the quantization transform coefficient 107 of low resolution error piece.Merge cells 216 is concluded the quantization transform coefficient 107 of differential motion vector 103 and low resolution error piece, outputs to entropy coding unit 255 as the low block message 102 of differentiating.136 pairs of quantization transform coefficients of inverse quantization unit 107 carry out inverse quantization, generate the decoding quantization transform coefficient 108 of low resolution error piece.Inverse transformation unit 137 pairs of decoding quantization transform coefficient 108 carries out anti-dct transform, generates the low resolution error piece 109 of decoding.Low each pixel of differentiating prediction pieces 106 of 217 pairs of adder units are carried out and deciphers the calculating of the addition calculated value of the pixel of position consistency and this pixel spatially of low resolution error piece 109, generate the low resolution of decoding piece 110.

Below, the low piece cutting unit 166 of differentiating is the same 2 * 2 the low resolution block 112 of being divided into of the low piece of differentiating piece 110 and Figure 36 of decoding 106.Each is low differentiates block 112 and is imported into fritter predicting unit 158 (be equivalent to Fig. 1, Fig. 2 152).

Fritter predicting unit 158 is to use the image of decoding that is maintained in the frame memory 153, to low each of block 112 differentiated, carries out the processing unit of the generation of 4 * 4 prediction fritter 116.The processing of this fritter predicting unit 158 is because only implement the image of having deciphered, so also can implement same processing for code translator as input.At this, the action of fritter predicting unit 158 is described.4 * 4 (being called " candidate prediction fritter ") 113 reduced processing units 134 that are imported into fritter predicting unit 158 are transformed to the few candidate of pixel count and dwindle prediction fritter 114, then are output to amplify motion deduction unit 138.Amplify motion infer 138 pairs of unit shown in Figure 35 2 * 2 low differentiate blocks 112 (112a～112d), with spatially corresponding 4 pixels (candidate is dwindled prediction fritter 114) of low 4 pixels differentiating block between carry out the piece coupling, detect the motion vector 115 that hangs down the resolution block (115a～115d).The difference evaluation of estimate of this moment is arranged to difference absolute value and (detailed content is narrated afterwards with Figure 32) of 4 pixels of block " low differentiate " and " in 16 pixels of 4 * 4 (candidate prediction fritters 113) of retrieval and low spatially corresponding 4 pixels (candidate is dwindled prediction fritter 114) of 4 pixels of differentiating block ".Only such decipher image as the difference evaluation of estimate of input value because of also calculating, so in code translator, also can detect the motion vector the same with code device with code translator.The prediction fritter obtain unit 139 from the decoded frame 520 of frame memory 153, obtain with the low motion vector 115 of differentiating block (115a～115d) corresponding 4 * 4 predict fritters 116 (116a～116d).

Below, piece merge cells 155 merges 4 prediction fritter 116a～116d like that shown in Figure 35 left side, generate 8 * 8 prediction fritter 118 (Figure 29).Subtrator 253 calculates for each pixel of coded object piece 202 and prediction the fritter 118 spatially pixel of position consistency and the difference value of this each pixel, generated error piece 208.

Error block coding unit 254 is to the error block 208 generated error block message 119 of encoding, and then outputs to the processing unit of entropy coding unit 255.At this, the action of specification error block encoding unit 254.218 pairs of error block of converter unit 208 are carried out 8 * 8 discrete cosine transforms (DCT), generate 64 conversion coefficients 209.219 pairs of conversion coefficients 209 of quantifying unit quantize, generated error block message (quantization transform coefficient) 119.

Error block decoding unit 156 is processing units of error block information 119 being deciphered and generated decoding error block 121.At this, the action of specification error block decoding unit 156.140 pairs of quantization transform coefficients of inverse quantization unit 119 carry out inverse quantization, generate decoding quantization transform coefficient 120.Inverse transformation unit 141 pairs of decoding quantization transform coefficient 120 carries out anti-dct transform, generates decoding error block 121.

Then, each pixel of 157 pairs of adder units prediction piece 118 is calculated and decoding the error block 121 spatially pixel of position consistency and the addition calculated value of this pixel, generates decode block 122.The decode block 122 that generates is remained in the current decoded frame 610 of frame memory 153.At last, 255 pairs of low resolution block messages 102 in entropy coding unit and error block information 119 are carried out entropy coding, are multiplexed on the coded data 101.

[structure of code translator]

The code translator of Figure 28 is that coded data 101 is deciphered, with the device of raster scan order playback 8 * 8 decode block 122.The coded data 101 that is equivalent to 18 * 8 decode block at first is imported into entropy decoding unit 150.Entropy decoding unit 150 generates low block message 102 and the error block information 119 differentiated by entropy decoding.

The low block decoding unit 151 of differentiating is that the low block message 102 of differentiating is deciphered, and generates the low processing unit of differentiating piece 110 of decoding that pixel count also lacks than the pixel count of decode block.At this, low action of differentiating block decoding unit 151 is described.Separative element 131 is separated into differential motion vector 103 and low resolution error block message (quantization transform coefficient of low resolution error piece) 107 to the low block message 102 of differentiating.Motion vector prediction/decoding unit 132 comes calculation of motion vectors 104 by the value of each composition of the predicted motion vector that adds on the value of each composition of differential motion vector 103.Predicted motion vector for example by the left side of current block, just go up, upper right (outside frame situation under be upper left) 3 adjacent motion vectors calculate according to median prediction (at each composition) as object.In the decoded frame of having deciphered 520 of motion compensation units 133 from be maintained at frame memory 153, obtain the piece 105 corresponding with motion vector 104.Piece 105 is transformed to the low resolution prediction piece 106 of having extracted pixel in length and breadth every 1 row out as shown in figure 36 like that by dwindling processing unit 134.The quantization transform coefficient 107 of 136 pairs of low resolution error pieces of inverse quantization unit carries out inverse quantization, generates the decoding quantization transform coefficient 108 of low resolution error piece.137 pairs of decoding of inverse quantization unit quantization transform coefficient 108 carries out anti-dct transform, generates the low resolution error piece 109 of decoding.Low each pixel of differentiating prediction pieces 106 of 135 pairs of adder units are calculated and the low resolution error piece 109 of the decoding spatially pixel of position consistency and the addition calculated value of this pixel, generate the low piece 110 of differentiating of decoding.

Then, the piece 106 of low resolution piece cutting unit 166 and Figure 36 is the same, the low resolution of decoding piece 110 is divided into 2 * 2 low resolution block 112.Each low block 112 of differentiating is imported into fritter predicting unit 158.

Fritter predicting unit 158 (be equivalent to Fig. 1, Fig. 2 152) is to use the image of decoding that is maintained in the frame memory 153, and each is lowly differentiated the processing unit that block 112 generates 4 * 4 prediction fritter 116.At this, the action of fritter predicting unit 158 is described.Amplify motion and infer that unit 138 is at shown in Figure 35 2 * 2 low resolution block 112 (112a～112d), and with spatially corresponding 4 pixels (candidate is dwindled prediction fritter 114) of low 4 pixels differentiating block between implement the piece coupling, detect the motion vector 115 (115a～115d) of low resolution block.The difference evaluation of estimate of this moment is arranged in absolute value and (the using Figure 32 to narrate afterwards in detail) of the difference of 4 pixels of block " low differentiate " and " in 16 pixels of 4 * 4 that detect (candidate prediction fritters 113) and low spatially corresponding 4 pixels (candidate is dwindled prediction fritter 114) of 4 pixels of differentiating block ".Only such decipher image as the difference evaluation of estimate of input value because of in code translator, also calculating, so in code translator, also can detect the motion vector the same with code device.The prediction fritter obtain unit 139 from the decoded frame 520 of frame memory 153, obtain with the low motion vector 115 of differentiating block (115a～115d) corresponding 4 * 4 predict fritters 116 (116a～116d).

Then, piece merge cells 155 merges 4 prediction fritter 116a～116d like that shown in the left side of Figure 35, generate 8 * 8 prediction piece 118 (Figure 28).

Error block decoding unit 156 is that error block information (quantization transform coefficient) 119 is deciphered the processing unit that generates decoding error block 121.At this, the action of specification error block decoding unit 156.140 pairs of quantization transform coefficients of inverse quantization unit 119 carry out inverse quantization, generate decoding quantization transform coefficient 120.Inverse transformation unit 141 pairs of decoding quantization transform coefficient 120 carries out anti-dct transform, generates decoding error block 121.

Then, each pixel of 157 pairs of adder units prediction piece 118 are calculated and decoding the error block 121 spatially pixel of position consistency and the addition calculated value of this each pixel, generate decode block 122.Decode block 122 is maintained in the current decoded frame 610 of frame memory 153.

[explanation of handling process]

Figure 30 represents to realize the moving image decoding handling process of predictive interpretation of the present invention, and Figure 33 represents to realize the moving image encoding handling process of predictive coding of the present invention.And being added the processing that same number is arranged in Figure 30 and Figure 33 is the processing of deriving same result.

[moving image decoding handling process]

Realize the moving image decoding handling process of predictive coding of the present invention with explanation among Figure 30.Handle (handling 401) if begin the decoding of next decode block, then at first,, the coded data of needed 1 amount in the decoding of next decode block is input to the entropy decoding unit 150 of Figure 28 by handling 402.The coded data that is transfused to is implemented entropys decoding by entropy decoding unit 150 in processing 403, obtain low block message and the error block information differentiated.Wherein low resolution block message is being handled in 404, handles with the low low block decoding of differentiating of block decoding unit 151 enforcements of differentiating, and generates and deciphers the low piece of differentiating.

At this, use Figure 31 to illustrate and handle 404 low resolution block decoding handling process.Handle (handling 421) if begin the low block decoding of differentiating, then at first, separative element 131 is obtained the low block message (handling 422) of differentiating, and this low block message of differentiating is separated into differential motion vector and low resolution error block message (handling 423).Handling in 424, motion vector prediction/decoding unit 132 calculates (decoding) motion vector 104 by the value of each composition of the predicted motion vector that adds on the value of each composition of differential motion vector 103.Predicted motion vector for example by the left side of current block, just go up, the 3 adjacent motion vector of upper right (being upper left under the situation outside the frame) calculates according to median prediction (at each composition) as object.Carrying out motion vector prediction in the decoding of the motion vector that obtains for afterwards decode block handles and is held.

Handle in 425 at next, in the decoded frame of having deciphered 520 of motion compensation units 133 from be maintained at frame memory 153, obtain the piece 105 corresponding with motion vector 104.Handle in 426 at next, acquired 105 is transformed at the low resolution prediction piece 106 of having extracted pixel in length and breadth every 1 row out as shown in figure 36 like that by dwindling processing unit 134.

Handle in 427 at next, the quantization transform coefficient 107 of 136 pairs of low resolution error pieces of inverse quantization unit carries out inverse quantization, generates the decoding quantization transform coefficient 108 of low resolution error piece.Then, inverse transformation unit 137 pairs of decoding quantization transform coefficient 108 carries out anti-dct transform, generates the low resolution error piece 109 of decoding.Then, handling in 428,135 pairs of low each pixels of differentiating prediction pieces 106 of adder unit are calculated and the low resolution error piece 109 of the decoding spatially pixel of position consistency and the addition calculated value of this pixel, generate the low piece 110 of differentiating of decoding, finish the low block decoding of differentiating and handle (handling 429).

Return Figure 30 explanation.In processing 405, low resolution piece cutting unit 166 is divided into the low resolution of decoding piece 110 2 * 2 low resolution block 112.Handling in 406,158 pairs 4 low blocks 112 of differentiating of fritter predicting unit are implemented the fritter prediction processing.

At this, fritter prediction processing flow process is described with Figure 32.If beginning fritter prediction processing (handling 430) then at first, handling in 431, as initial value, is set at max-int to evaluation of estimate E min, and is set in 0 vector of cutting apart motion vector.Then, handling in 432, the low block 112 of differentiating is being input to fritter predicting unit 158.Then in the amplification campaign inference process of processing 434, infer the low motion vector of differentiating block of unit 138 detections with amplifying motion.

Amplify in the motion inference process at this, (0 vector) extracts candidate prediction fritter out according to the spiral order from its center in the seek scope (on longitudinal and transverse direction being ± 16 pixels) of the decoded frame of frame memory 153, and detecting evaluation of estimate E min is minimum prediction tile position.More particularly, handling in 4341, obtain the prediction fritter of next candidate from the decoded frame of frame memory 153, handling in 4342, acquired candidate prediction fritter for example hangs down the dimensions (pixel count) of differentiating the prediction piece with piece 106 the same being transformed to of Figure 36.Handling in 4343, calculate evaluation of estimate E (" low 4 pixels differentiating block " and " in 4 * 4 16 pixels of retrieval; and low 4 spatially corresponding pixels of 4 pixels of differentiating block " difference absolute value with), handling in 4344, judge whether evaluation of estimate E is littler than evaluation of estimate E min.At this, when evaluation of estimate E than the little situation of evaluation of estimate E min under, handling in 4345, evaluation of estimate E min is updated to evaluation of estimate E, cut apart motion vector and be updated to the cut apart motion vector corresponding with evaluation of estimate E.After, till finishing, the retrieval of the whole candidate prediction fritter in seek scope repeats to handle 4341～processing 4345.At last, handling in 435, the prediction fritter obtain unit 139 by from the decoded frame of frame memory 153, obtain with evaluation of estimate E for the minimum corresponding prediction fritter of motion vector of cutting apart, finish fritter prediction processing (handling 436).Generate 4 prediction fritters by above fritter prediction processing.

Return Figure 30 explanation.4 prediction fritters that generate in processing 406 are merged into the prediction piece by piece merge cells 155 in processing 407.Handle in 408 at next, 156 pairs of error block information of error block decoding unit (quantization transform coefficient) are deciphered (inverse quantization/inverse transformation), generate the decoding error block.Handle in 409 at next, each pixel is predicted adding of piece and decoding error block with adder unit 157, generate decode block.At last, handling in 410, decode block is remained in the current decoded frame of frame memory 153, the block decoding processing finishes (handling 411).

[moving image encoding handling process]

Below, use Figure 33 that the encoding process flow process of the coded object piece of realizing predictive coding of the present invention is described.At first, cut apart incoming frame 201, obtain coded object piece 202 by the code block segmentation unit 251 of Figure 29.If the encoding process (handling 441) of beginning coded object piece 202, then at first, in processing 442, the low piece generation unit 252 of differentiating of decoding is carried out the low resolution block encoding of Figure 34/decoding processing.

Below, with Figure 34 the low block encoding/decoding processing of differentiating is described.If begin the low block encoding/decoding processing (handling 451) of differentiating, then at first, to handle in 452, input coded object piece thereafter handling in 453, detects motion vector from the decoded frame of coded object piece and frame memory.And, for the detection method of the motion vector here, so in the low formation explanation of differentiating piece generation unit 252 of the decoding of Figure 29, be illustrated omission.Handle in 454 at next, from the decoded frame of frame memory, obtain the piece corresponding with motion vector.In following processing 455, after predicted motion vector generates, between motion vector and predicted motion vector, each composition is subtracted calculation and handle, generate differential motion vector.The differential motion vector of motion vector and generation is to keep in order to carry out motion vector prediction in the coded object piece afterwards.And, for the generation method of the predicted motion vector here, so because in the formation explanation of operation vector forecasting/coding unit 212 of Figure 29, be illustrated omission.In following processing 456, coded object piece (forecasting object piece) for example piece 106 with Figure 36 is the same, be transformed to the low coded object piece (the low forecasting object piece of differentiating) of differentiating, and then handling in 457, the piece corresponding with motion vector is the same with above-mentioned coded object piece (forecasting object piece), is transformed to the low prediction piece of differentiating.

Handling in 458, differentiating coded object piece (the low forecasting object piece of differentiating) and low the resolution between the prediction piece subtracts calculation to each pixel, generating low resolution error piece low.Handle in 459 at next, hang down the coding (transform/quantization) of resolution error piece, generate the quantization transform coefficient of low resolution error piece.In following processing 427, carry out the decoding (inverse quantization/inverse transformation) of quantization transform coefficient, generate the low resolution error piece of decoding.In following processing 428, low the resolution between prediction piece and the low resolution error piece of decoding each pixel is added, generate the low piece of differentiating of decoding.At last, handling in 460, carry out the merging of differential motion vector and quantization transform coefficient, the low block message of differentiating of generation hangs down and differentiates block encoding/decoding processing end (handling 461).

Return Fig. 3 explanation.Handling in 405, the low piece cutting unit 166 of differentiating is divided into the low block of differentiating to the low piece of differentiating of the decoding that generates.Handling in 406, the low resolution block that 158 pairs of fritter predicting unit are divided into carries out the fritter prediction processing, the generation forecast fritter.And at this, relevant fritter prediction processing (Figure 32) is because be illustrated its explanation of Therefore, omited.

Handling in 407, piece merge cells 155 is merged into the prediction piece to the prediction fritter that generates.Handle in 443 at next, subtrator 253 is implemented subtracting of each pixel and is calculated processing, generated error piece between coded object piece and prediction piece.Handle in 444 at next, error block coding unit 254 carries out error block coding (transform/quantization), generated error block message (quantization transform coefficient).In following processing 408, error block decoding unit 156 carries out the decoding (inverse quantization/inverse transformation) of error block information (quantization transform coefficient), generates the decoding error block.Handle in 409 at next, adder unit 157 is implemented the processing that adds of each pixel between prediction piece and decoding error block, generate decode block.Then, handling in 410, in the decode block that generates being remained on the current decoded frame of frame memory 153 after, handling in 445, entropy coding unit 255 hang down the entropy coding of differentiating block message and error block information, block encoding processing then finishes (processing 446).

[the formation example of computer system]

Figure 16 is used to illustrate use to store with the image encoding of above-mentioned execution mode and handles relevant dynamic image encoding, handles the storage medium of the floppy disk etc. of relevant moving image translator, the figure when implementing by computer system with image decoding.

Outward appearance, cross-section structure and floppy disk that Figure 16 (b) expression is seen from the front of floppy disk, Figure 16 (a) expression is as the example of the physical format of the floppy disk of recording medium main body.Floppy disk FD is built in the shell F, inwardly week forms plurality of tracks Tr with concentric circles from periphery on the surface of this dish, and each magnetic track is split into 16 sector Se on angle direction.Thereby, in storing the floppy disk of said procedure, on the zone of distributing on the above-mentioned floppy disk FD, writing down data as said procedure.

In addition, Figure 16 (c) expression is used for carrying out the formation of the record reproducing of said procedure on floppy disk FD.Under situation about said procedure being recorded on the floppy disk FD, from computer system Cs, the data as said procedure are write via floppy disk.In addition, when constructing in computer system with the program in the floppy disk under the situation of above-mentioned code device or code translator,, be transferred to computer system with floppy disk read routine from floppy disk.

And, in the above description, use floppy disk to be illustrated, even but use CD can carry out too as data carrier.In addition, recording medium is not limited to this, so long as the medium that IC-card, ROM box etc. can logging programs just can similarly be implemented.In addition as computer, comprise the DVD player that possesses CPU and adopt software to handle and control, set-top box, mobile phone etc.

Below, the formation of account for motion image decoding program and dynamic image encoding.

As shown in figure 37, moving image translator 161 is stored in the program storage area 160a that is formed on the recording medium 160.The formation of moving image translator 161 possesses: the primary module 161a that the decoding of Comprehensive Control moving image is handled; Be used to carry out the entropy decoding module 161b of action of the entropy decoding unit 150 of Figure 28; Be used to carry out the low resolution block decoding module 161c of action of the low resolution block decoding unit 151 of Figure 28; The low resolution piece of action that is used to carry out the low resolution piece cutting unit 166 of Figure 28 is cut apart module 161h; Be used to carry out the fritter prediction module 161i of action of the fritter predicting unit 158 of Figure 28; The piece of action that is used to carry out the piece merge cells 155 of Figure 28 merges module 161g.

As shown in figure 38, dynamic image encoding 261 is stored in the program storage area 260a that is formed on the recording medium 260.The formation of dynamic image encoding 261 possesses: the primary module 261a that the Comprehensive Control moving image encoding is handled; Be used to carry out the code block segmentation module 261b of action of the code block segmentation unit 251 of Figure 29; Be used to carry out the low low piece generation module 261c that differentiates of decoding that differentiates the action of piece generation unit 252 of decoding of Figure 29; The low resolution piece of action that is used to carry out the low resolution piece cutting unit 166 of Figure 29 is cut apart module 261h; Be used to carry out the fritter prediction module 261i of action of the fritter predicting unit 158 of Figure 29; The piece of action that is used to carry out the piece merge cells 155 of Figure 29 merges module 261g.

And, the formation of above-mentioned moving image translator 161 can also be its part or all transmit via the transmission medium of communication line etc., receive and write down (comprising installation) with other equipment.Equally, the formation of dynamic image encoding 261 also can be its part or all transmit via the transmission medium of communication line etc., receive and write down (comprising installation) with other equipment.

[explanations of various deformation forms]

Embodiments of the present invention more than have been described, but also can have carried out following such distortion, all forms are included among the present invention.

(1) deformation form relevant with Forecasting Methodology

In the above description, as Forecasting Methodology the prediction that is defined in time orientation has been described, but has used the same method at direction in space for using coded object piece (forecasting object piece) or also can fitting with the prediction of deciphering image of having passed through decoding of decode block in same frame.Promptly, resolution piece based on above-mentioned decoding, in the adjacent pixel of the low resolution block of having cut apart the low resolution of this decoding piece again being used in the same frame of having deciphered, from for example, H.264 (" Textof ISO/IEC 14496-10 Advanced Video Coding 3rd Edition ", September 2004.) a plurality of intra-prediction mode in determined best pattern after, situation about predicting with the low piece of differentiating the suitable high-resolution signal of piece of this decoding etc. is also contained among the present invention.In this case, because do not need the property expressed ground to transmit each low predictive mode of differentiating block, so have with few supplementary good effect of forecasting object piece accurately just.In addition, the prediction by making up above-mentioned time orientation and the prediction of direction in space also can be carried out high-precision prediction.

(2) deformation form relevant with the coding of error block

Error block between prediction piece and the coded object piece is encoded not necessarily.That is, the processing of error block coding unit 254 (Figure 29) and error block decoding unit 156 (Figure 28 and Figure 29) can be omitted.In addition, can also determine having or not of coding to each coded object piece or decode block.In addition, the concrete coding/decoding method for error block and low resolution error piece is not limited to above-mentioned method.

(3) deformation form relevant with the coding of a plurality of predictive modes

In the above-described embodiment, only as Forecasting Methodology of the present invention, and, also can be suitable for Forecasting Methodology of the present invention to the Forecasting Methodology of the encoding process that is applicable to the coded object piece for the coding method of dwindling in the past of not carrying out the coded object piece.

(4) deformation form relevant with block size

In the above description, be that the coded object piece (forecasting object piece) and the size of decode block are arranged to 8 * 8, the low size of differentiating piece is arranged to 4 * 4, the low block of differentiating is arranged to 2 * 2, but predictive coding of the present invention is not limited to these sizes.For example, the size of coded object piece and decode block is arranged to and H.264 the same 16 * 16, the low piece of differentiating is arranged to 8 * 8, the low block of differentiating is arranged to 2 * 2 etc., it is also effective further to cut apart situation the present invention of cutting apart low resolution piece more meticulously than 4.Equally, the block size of enforcement transform/quantization is not defined yet.For example, as figure H.264, error block is divided into 4 * 4, it is good to be with 4 * 4 that unit carries out transform/quantization.

(5) deformation form relevant with dwindling method

In the above description, as shown in figure 36, be arranged in the method for rejecting 1 row on vertical/transverse direction at interval, but the processing of dwindling of piece is not limited to the method dwindling to handle.For example, if the low piece of differentiating pre-determines between code device and code translator, then can be arranged to pixel count than decode block (coded object piece, the forecasting object piece) also few shape arbitrarily, also comprise each pixel of the piece 106 of Figure 36 on direction in length and breadth, stagger 1 pixel piece (piece 1060 of Figure 36) and have the pixel of piece 106 of Figure 36 and the pixel both sides' of piece 1060 piece 1062 etc.In addition, can be useful in piece has been implemented after the low-pass filtering treatment to reduce the method for sampling (down sampling) and 4 neighbors are averaged and are transformed to the method etc. of 1 pixel on direction in length and breadth, be applicable to the whole bag of tricks that image dwindles.

(6) with the relevant deformation form of enforcement order that dwindles processing

Dwindling processing unit can omit.For example, the function of dwindling processing unit does not need to dwindle processing unit being contained in motion compensation units and amplifying motion and infer under the situation in the unit and can carry out not comprising under the situation of the processing that does not belong to the low pixel of differentiating piece in evaluation.In addition, differentiate dwindling processing and adopting the processing of dwindling of hanging down resolution prediction piece generation unit between code translator and code device, to predesignate of block decoding unit, then also can use diverse ways if so-called employing is hanged down.

(7) infer relevant deformation form with the amplification motion

Seek scope in the motion inference process, looked-up sequence and search starting point and do not limit and above-mentioned method.If seek scope is predesignated in code device and code translator, the scope arbitrarily that then can set.For looked-up sequence too, for example, having more than is to begin spiral retrieval from the center, also can be suitable for from upper left to the bottom right according to the method for raster scan order retrieval scope etc.For the retrieval starting point, be not 0 vector also, for example can also set seek scope to predicted vector as the center.In addition, also be not limited to the piece coupling that detects motion vector for motion deductions/motion compensation process, even carry out the Calculation Method of the predicted value of the interior pixel of piece with the kinematic parameter of affine parameter and perspective projection transformation parameter etc., the present invention also can be suitable for.In this case, do not need to transmit with the detected kinematic parameter of fritter predicting unit.The fritter prediction of the 1st embodiment is not limited to the piece coupling too.

(8) with separative element, the relevant deformation form of merge cells

The separative element 131 of Figure 28 and the merge cells 216 of Figure 29 are not necessarily.For example, in Figure 28, motion vector prediction/decoding unit 132 and inverse quantization unit 136 can directly receive data from entropy decoding unit 150.In Figure 29, can from motion vector prediction/coding unit 212 and quantifying unit 215, directly output to entropy coding unit 255 to data.

Claims

1. dynamic image encoding device is to be the dynamic image encoding device of unit generation forecast piece with the piece, it is characterized in that comprising:

Keep having deciphered the holding member of image;

Is the coded object image segmentation of having imported the 1st partition member of a plurality of forecasting object pieces;

The conversion of above-mentioned forecasting object piece and after being encoded to the low resolution piece that pixel count also lacks than the pixel count of this forecasting object piece,, generate the low generation parts of differentiating piece of decoding by deciphering this low coded data of differentiating piece;

The low piece of differentiating of the decoding of above-mentioned generation is enlarged into amplifier unit with the piece of above-mentioned forecasting object piece same pixel number;

Cut apart the above-mentioned piece that has been exaggerated by the rule of cutting apart, generate the 2nd partition member of a plurality of fritters based on regulation;

Use the above-mentioned fritter of having deciphered image and above-mentioned generation, generate the prediction fritter prediction parts of this fritter;

By merging the above-mentioned a plurality of prediction fritters that generated, the merging parts of generation forecast piece based on the above-mentioned rule of cutting apart.

2. dynamic image encoding device according to claim 1 is characterized in that also comprising:

Is above-mentioned coded object image segmentation the 3rd partition member of coded object piece;

With the pixel is the subtracting section spare that calculation generates error block that subtracts that unit carries out above-mentioned divided coded object piece and above-mentioned prediction piece;

The addressable part that the above-mentioned error block that generates is encoded;

Above-mentioned coded data of having encoded is deciphered, generated the decoding unit of decoding error block;

Be that unit carries out adding of above-mentioned prediction piece and above-mentioned decoding error block and generates decode block with the pixel, this decode block outputed to the output block of above-mentioned holding member.

3. dynamic image encoding device according to claim 1 is characterized in that:

Above-mentioned amplifier unit uses the pixel of the low pixel of differentiating piece of the above-mentioned decoding that generates and the decode block adjacent with the top of forecasting object piece and the left side, generates and the piece of the same pixel count of above-mentioned forecasting object piece.

4. dynamic image encoding device according to claim 1 is characterized in that:

The formation of above-mentioned generation parts comprises:

Above-mentioned forecasting object piece is transformed to the parts of the low resolution forecasting object piece that pixel count also lacks than the pixel count of this forecasting object piece based on the transformation rule of regulation;

Generate the low low resolution prediction piece of differentiating the prediction piece and generate parts;

With the pixel is that unit carries out above-mentioned low resolution forecasting object piece and subtracting of above-mentioned low resolution prediction piece calculated the parts that generate low resolution error piece;

Behind the coding that carries out this low resolution error piece, by the coded data of above-mentioned low resolution error piece being deciphered the parts that generate the low resolution error piece of decoding;

With the pixel is that unit carries out adding of low resolution error piece of above-mentioned decoding and above-mentioned low resolution prediction piece, restores the parts of above-mentioned low resolution piece,

The formation that above-mentioned low resolution prediction piece generates parts comprises:

Generate the parts of the 2nd prediction piece identical with above-mentioned prediction piece pixel count;

From the 2nd prediction piece, generate the parts that piece is predicted in above-mentioned low resolution based on above-mentioned transformation rule,

The formation of above-mentioned generation parts also comprises: under the whole differential pixel value in the above-mentioned low resolution error piece is zero situation, predict the parts that piece is exported as above-mentioned prediction piece to the above-mentioned the 2nd.

5. dynamic image decoding device is to be the moving picture decoding device of unit generation forecast piece with the piece, it is characterized in that also comprising:

Keep having deciphered the holding member of image;

The coded data of having imported is deciphered, generated pixel count than the decoding unit of predicting the low resolution piece that the piece pixel count also lacks;

The above-mentioned low resolution piece that has generated is enlarged into amplifier unit with the piece of the same pixel count of above-mentioned prediction piece;

By cutting apart the piece that this is exaggerated, generate the partition member of a plurality of fritters based on the rule of cutting apart of regulation;

Use above-mentioned image and the fritter after above-mentioned cutting apart deciphered, generate the prediction parts of the prediction piece of this fritter;

By merging these a plurality of prediction fritters that generated, the merging parts of generation forecast piece based on the above-mentioned rule of cutting apart.

6. dynamic image decoding device according to claim 5 is characterized in that also comprising:

Decipher above-mentioned coded data, the generation parts of generated error piece;

Be that add above-mentioned prediction piece and above-mentioned error block of unit generates decode block with the pixel, this decode block outputed to the output block of above-mentioned holding member.

7. dynamic image decoding device according to claim 5 is characterized in that:

Above-mentioned amplifier unit uses the pixel of the above-mentioned low resolution piece that has generated and the pixel of the decode block adjacent with the top of prediction piece and the left side, generates and the piece of the same pixel count of above-mentioned prediction piece.

8. dynamic image decoding device according to claim 5 is characterized in that:

The formation of above-mentioned decoding unit comprises:

The parts that the above-mentioned coded data of having imported is deciphered and low resolution error piece is deciphered;

Be that add above-mentioned low resolution error piece and above-mentioned low resolution prediction piece of unit generates the parts of above-mentioned low resolution piece with the pixel,

The parts of the 2nd prediction piece of generation and the same pixel count of above-mentioned prediction piece;

From the 2nd prediction piece, generate the parts that piece is predicted in above-mentioned low resolution based on predetermined rule,

The formation of above-mentioned decoding unit also comprises:

Under the whole differential pixel value in the above-mentioned low resolution error piece is zero situation, the parts that above-mentioned the 2nd prediction piece is exported as above-mentioned prediction piece.

9. dynamic image encoding method is to use and deciphers image and be the dynamic image encoding method of unit generation forecast piece with the piece, it is characterized in that comprising:

Is the image segmentation of the coded object of having imported the 1st segmentation procedure of a plurality of forecasting object pieces;

The conversion of above-mentioned forecasting object piece and after being encoded to the low resolution piece that pixel count also lacks than the pixel count of this forecasting object piece,, generate the low generation step of differentiating piece of decoding by this low coded data of differentiating piece is deciphered;

The low piece of differentiating of the above-mentioned decoding that has generated is enlarged into amplification procedure with the piece of the same pixel count of above-mentioned forecasting object piece;

By cutting apart the above-mentioned piece that has amplified, generate the 2nd segmentation procedure of a plurality of fritters based on the rule of cutting apart of regulation;

Use above-mentioned image and the above-mentioned fritter that has generated deciphered, generate the prediction steps of the prediction fritter of this fritter;

By merging the above-mentioned a plurality of prediction fritters that generated, the combining step of generation forecast piece based on the above-mentioned rule of cutting apart.

10. motion picture decoding method is to use and deciphers image and be the motion picture decoding method of unit generation forecast piece with the piece, it is characterized in that comprising:

The coded data of having imported is deciphered, generated pixel count than the decoding step of predicting the low resolution piece that the piece pixel count also lacks;

The above-mentioned low resolution piece that has generated is enlarged into amplification procedure with the piece of the same pixel count of above-mentioned prediction piece;

By cutting apart the piece that this has amplified, generate the segmentation procedure of a plurality of fritters based on the rule of cutting apart of regulation;

Use above-mentioned image and the above-mentioned divided fritter deciphered, generate the prediction steps of the prediction fritter of this fritter;

By based on above-mentioned a plurality of prediction fritters that rule merges above-mentioned generation, the combining step of generation forecast piece cut apart.

11. a dynamic image encoding is characterized in that:

Make that to possess that maintenance deciphered image and had with the piece be that the computer of holding member of dynamic image encoding of function of unit generation forecast piece is carried out following the processing:

Is the image segmentation of the coded object of having imported the 1st dividing processing of a plurality of forecasting object pieces;

The conversion of above-mentioned prediction piece and after being encoded to the low resolution piece that pixel count also lacks than the pixel count of this forecasting object piece,, generate the low generation of differentiating piece of decoding and handle by this low coded data of differentiating piece is deciphered;

The low piece of differentiating of the above-mentioned decoding that has generated is enlarged into processing and amplifying with the piece of the same pixel count of above-mentioned forecasting object piece;

By cutting apart the above-mentioned piece that is exaggerated, generate the 2nd dividing processing of a plurality of fritters based on the rule of cutting apart of regulation;

Use above-mentioned image and the above-mentioned fritter that has generated deciphered, generate the prediction processing of the prediction fritter of this fritter;

By cutting apart a plurality of prediction fritters that rule merges above-mentioned generation based on above-mentioned, the merging of generation forecast piece is handled.

12. a moving image translator is characterized in that:

Make that to possess that maintenance deciphered image and had with the piece be that the computer of holding member of moving image translator of function of unit generation forecast piece is carried out following the processing:

The coded data of having imported is deciphered, and the generation pixel count is handled than the decoding of the low resolution piece that the pixel count of predicting piece also lacks;

The above-mentioned low resolution piece that has generated is enlarged into processing and amplifying with the piece of the same pixel count of above-mentioned prediction piece;

By cutting apart the piece that this has amplified, generate the dividing processing of a plurality of fritters based on the rule of cutting apart of regulation;

Use above-mentioned image and the above-mentioned fritter of having cut apart deciphered, generate the prediction processing of the prediction fritter of this fritter;

By merging these a plurality of prediction fritters that generated based on the above-mentioned rule of cutting apart, the merging of generation forecast piece is handled.

13. a dynamic image decoding device, be piece with regulation to be unit decipher the coded data that becomes the decoding object, the dynamic image decoding device of generation forecast piece is characterized in that comprising:

Keep having deciphered the holding member of image;

By the coded data of having imported being carried out entropy decoding, generate pixel count and also lack, and encoded and on time/space, be in the entropy decoding unit of low resolution block message of the low resolution piece of same position with this decode block than the pixel count of above-mentioned decode block;

The low resolution block message that is generated by above-mentioned entropy decoding unit is deciphered,, generated the low resolution block decoding parts of above-mentioned low resolution piece with reference to the above-mentioned image of having deciphered;

Cut apart the low resolution piece that generates by above-mentioned low resolution block decoding parts by the rule of cutting apart, generate the low resolution piece partition member of a plurality of low resolution blocks based on regulation;

According to the image of decoding that is maintained on the above-mentioned holding member, with the low resolution block that generates by above-mentioned low resolution piece partition member, generate pixel count and be somebody's turn to do more than the low resolution block pixel count, and differentiated block is in the prediction fritter of same position on time/space prediction fritter generation parts with hanging down;

By generating the prediction fritter that parts generate based on merging by above-mentioned prediction fritter with the above-mentioned merging rule of cutting apart regular corresponding regulation, the piece of generation forecast piece merges parts.

14. dynamic image decoding device according to claim 13 is characterized in that:

Above-mentioned entropy decoding unit is by carrying out entropy decoding to above-mentioned coded data, further generate the error block information that is included in the error of presentation code object piece and prediction piece in this coded data,

Above-mentioned dynamic image decoding device also comprises:

By the error block information that is generated by above-mentioned entropy code translator is deciphered the error block decoding unit of generated error piece;

Based on merging the prediction piece of parts generation and the error block that generates by above-mentioned error block decoding unit, the decode block generation parts of generation decode block by above-mentioned.

15. a picture coding device, be with the regulation piece be that unit encodes to the image that becomes coded object, the dynamic image encoding device of generation forecast piece is characterized in that comprising:

Keep having deciphered the holding member of image;

Is the image segmentation of having imported the code block segmentation parts of a plurality of coded object pieces;

By the pixel count of pixel count than the coded object piece of being cut apart by above-mentioned code block segmentation parts also lacked, and the low resolution piece that is in same position with this coded object piece on time/space is encoded, generate the low block message of differentiating, this low block message of differentiating is deciphered, with reference to the above-mentioned image of having deciphered, generate the low low piece of differentiating of decoding of differentiating piece of decoding and generate parts;

Cut apart by the low low piece of differentiating of decoding of differentiating the generation of piece generation parts of above-mentioned decoding by the rule of cutting apart, generate the low resolution block partition member of a plurality of low resolution blocks based on regulation;

According to the image of decoding that is maintained in the above-mentioned holding member, with the low resolution block that generates by above-mentioned low resolution piece partition member, generate more than this low pixel count of differentiating block of pixel count, and and should lowly differentiate block is in the prediction fritter of same position on time/space prediction fritter generation parts;

16. dynamic image encoding device according to claim 15 is characterized in that also comprising:

By being that unit asks the encoding block cut apart by above-mentioned code block segmentation parts and with above-mentioned error that merges the prediction piece that parts generate, the error block that generates the error block of this error of expression generates parts with the pixel;

By the error block that is generated the parts generation by above-mentioned error block is encoded the error block addressable part of generated error block message;

By the error block information that is generated by above-mentioned error block addressable part is deciphered, generate the error block decoding unit of decoding error block;

Based on merging the prediction piece of parts generation and the decoding error block that generates by above-mentioned error block decoding unit, the decode block generation parts of generation decode block by above-mentioned.

17. dynamic image encoding device according to claim 16 is characterized in that also comprising:

To generating the low resolution block message of parts generation and carry out entropy coding, generate the entropy coding parts of coded data by the error block information that above-mentioned error block addressable part generates by the low piece of differentiating of above-mentioned decoding.

18. motion picture decoding method, be to have deciphered the holding member of image possessing maintenance, piece with regulation is that unit is deciphered the coded data that becomes the decoding object, and the motion picture decoding method in the dynamic image decoding device of generation forecast piece is characterized in that comprising:

By the coded data of the decoding object imported being carried out entropy decoding, generation is also lacked the pixel count of pixel count than above-mentioned decode block, and and this decode block low resolution piece of on time/space, being in same position carried out the entropy decoding step of the low resolution block message of coding;

The low resolution block message that generates in above-mentioned entropy decoding step is deciphered,, generated the low resolution block decoding step of above-mentioned low resolution piece with reference to the above-mentioned image of having deciphered;

Be segmented in the low resolution piece that generates in the above-mentioned low resolution block decoding step by the rule of cutting apart, generate the low resolution piece segmentation procedure of a plurality of low resolution blocks based on regulation;

According to the image of decoding that is maintained in the above-mentioned holding member, with the low resolution block that in above-mentioned low resolution piece segmentation procedure, generates, generate more than this low pixel count of differentiating block of pixel count, and and should lowly differentiate block is in the prediction fritter of same position on time/space prediction fritter generation step;

Generate the prediction fritter that generates in the step, the piece combining step of generation forecast piece based on being incorporated in above-mentioned prediction fritter with the above-mentioned merging rule of cutting apart regular corresponding regulation.

19. dynamic image encoding method, be to have deciphered the holding member of image possessing maintenance, piece with regulation is that unit encodes to the image that becomes coded object, and the dynamic image encoding method in the dynamic image encoding device of generation forecast piece is characterized in that comprising:

Is the image segmentation of the coded object of having imported the code block segmentation step of a plurality of coded object pieces;

Also lack than the coded object piece that in above-mentioned code block segmentation step, is partitioned into by the encoded pixels number, and and this coded object piece is in the low resolution piece of same position on time/space, generate the low block message of differentiating, this low block message of differentiating is deciphered, with reference to the above-mentioned image of having deciphered, generate the low low piece of differentiating of decoding of differentiating piece of decoding and generate step;

Be segmented in the low a plurality of low resolution piece that generates in the piece generation step of differentiating of above-mentioned decoding by the rule of cutting apart, generate the low resolution piece segmentation procedure of a plurality of low resolution blocks based on regulation;

According to the image of decoding that is maintained in the above-mentioned holding member, with the low resolution block that in above-mentioned low resolution piece segmentation procedure, generates, generate more than the low pixel count of differentiating block of pixel count, and and should lowly differentiate block is in the prediction fritter of same position on time/space prediction fritter generation step;

By generating the prediction fritter that generates in the step, the piece combining step of generation forecast piece based on being incorporated in above-mentioned prediction fritter with the above-mentioned merging rule of cutting apart regular corresponding regulation.

20. moving image translator, it makes to be arranged on possesses the holding member that image has been deciphered in maintenance, and to be unit with the piece of regulation decipher and computer in the dynamic image decoding device of generation forecast piece has following functions of components the coded data that becomes the decoding object:

By the coded data of the decoding object imported being carried out entropy decoding, generation is also lacked the pixel count of pixel count than above-mentioned decode block, and and this decode block low resolution piece of on time/space, being in same position carried out the entropy decoding unit of the low resolution block message of coding;

Cut apart the low resolution piece that generates by above-mentioned low resolution block decoding parts by the rule of cutting apart, generate the low resolution block partition member of a plurality of low resolution blocks based on regulation;

21. dynamic image encoding, it makes to be arranged on possesses the holding member that image has been deciphered in maintenance, thereby and have following functions of components with the piece of the regulation computer that to be unit encode in the dynamic image encoding device of generation forecast piece to the image that becomes coded object:

Is the image segmentation of the coded object of having imported the code block segmentation parts of a plurality of coded object pieces;

By the pixel count of pixel count than the coded object piece of being cut apart by the code block segmentation parts also lacked, and the low resolution piece that is in same position with this coded object piece on time/space is encoded, generate the low piece of differentiating, this low block message of differentiating is deciphered, with reference to the above-mentioned image of having deciphered, generate the low low piece of differentiating of decoding of differentiating piece of decoding and generate parts;

Cut apart by the low low piece of differentiating of decoding of differentiating the generation of piece generation parts of above-mentioned decoding by the rule of cutting apart, generate the low resolution piece partition member of a plurality of low resolution blocks based on regulation;

According to image after the decoding that is maintained in the above-mentioned holding member, with the low resolution block that generates by above-mentioned low resolution piece partition member, generate pixel count and be somebody's turn to do more than the low resolution block, and differentiated block is in the prediction fritter of same position on time/space prediction fritter generation parts with hanging down;