CN1893656A

CN1893656A - Device and method for predicting and encoding image and recording medium

Info

Publication number: CN1893656A
Application number: CN 200610095683
Authority: CN
Inventors: 文仲丞; 申省梅; 陈朝庆
Original assignee: Matsushita Electric Industrial Co Ltd
Current assignee: Sun Patent Trust Inc
Priority date: 1996-05-28
Filing date: 1997-05-28
Publication date: 2007-01-10
Anticipated expiration: 2017-05-28
Also published as: CN100539695C; CN1893655A; CN100521787C

Abstract

There is disclosed image predictive coding/decoding apparatus and method, of which the transform efficiency is remarkably improved in comparison with the prior art, and recording medium. According to the image predictive coding apparatus and method, when dividing inputted image data to be coded into image data of a plurality of small regions which are adjacent to each other and coding the image data of an objective small region to be processed among the image data of the plurality of divided small regions which are adjacent to each other, reconstructed image data of a reproduction small region adjacent to the image data of the objective small region to be processed is used as image data of an intra-frame prediction small region of the objective small region to be processed, the image data of the intra-frame prediction small region is used as image data of an optimum prediction small region and image data of a difference small region which are differences between the image data of the objective small region to be processed and the image data of the optimum prediction small region is generated.

Description

Image forecasting encoding/decoding apparatus and method and recording medium

The application is that application number is dividing an application of 200510067737.6 patent application.

Technical field

The present invention relates to image prediction encoding device and method, image prediction/decoding device and method and recording medium.Particularly relate to and for example store into DID on the recording medium such as CD or be used for carrying out image transmitted predictive coding apparatus and method and image prediction/decoding device and method by communication line as the image of still image or cardon elephant.In addition, also relate to the program recording medium and the program recording medium that writes down the step that comprises above-mentioned visual prediction decoding method that record comprises the step of above-mentioned image prediction/encoding method.

Background technology

In order to store effectively or the transmission of digital image, must carry out compressed encoding.As the method that is used for digital picture is carried out compressed encoding, except with the discrete cosine transform of JPEG (JointPhotographic Experts Group) and MPEG (Motion Picture ExpertsGroup) representative (below, be called dct transform) outside, also have sub-band coding, wavelet coding, waveform coding method such as (Off ラ Network Le) coding partly.In addition,, use the inter picture prediction of moving compensation, differential signal is carried out waveform coding in order to remove the signal of the redundancy between image.

In the MPEG mode, input picture is divided into a plurality of 16 * 16 macro block and handles.And then be 8 * 8 piece, and after being handled, 8 * 8 dct transform carries out quantification treatment with 1 macroblock partition.This just is called intraframe coding.

On the other hand, in moving detection method based on the piece coupling, from other adjacent in time frame, target macroblock is detected the predicted macroblock of error minimum, detected predicted macroblock is deducted from target macroblock, generate the difference macro block, carrying out carrying out quantification treatment behind 8 * 8 the dct transform.This is called interframe encode, predicted macroblock is called the prediction signal of time zone.

Common image, spatially similar zone is many, can utilize this character to carry out image approximate at area of space and handle.Identical with the prediction signal of time zone, also can from identical frame, ask prediction signal.This is called the spatial prediction signal.

Because 2 spatially adjacent pixel values are approaching, so the prediction signal of area of space is usually located at the position approaching with echo signal.On the other hand, in receiver side or regeneration side owing to there is not an original image, so, must use encode in the past and the signal of regenerating as prediction signal.Must be according to 2 factors prediction signal in span zone at high speed.This is must to decode and regenerate because be used for the signal of generation forecast signal.

Therefore, must be with the simple mode prediction signal in span zone accurately.In addition, the structure that requirement can be carried out high-speed computation in encoding apparatus and decoding apparatus.

Yet the coding of view data uses at JPEG, MPEG1 widely, H.261, MPEG2 and H.263 waiting in a lot of international standards.The latter's standard and then improved code efficiency.That is, compare, be devoted to further to reduce figure place and show identical image quality with the standard that has earlier.

The coding of the view data of motion video is made of frame interior coding and predictive frame coding.For example, in the typical hybrid coding system of MPEG1 standard, continuous frame can be divided into following 3 different types.

(a) frame interior (below, be called the I frame),

(b) predictive frame (below, be called the P frame),

(c) bi-directional predicted frames (below, be called the B frame).

The I frame is independent of other frame and encodes.That is, irrelevant to the compression and other the frame of I frame.For the content of the frame (this frame is exactly the P frame) of predictive coding, the P frame uses a frame of front, detects and compensation is encoded by moving.The moving detection and the compensation of the information of the subsequent frame of B frame by using prediction one frame information of front and B frame intermediate data are encoded.Frame in the past and follow-up frame can be I frame or P frame.The I frame is encoded with internal schema.P frame and B frame are encoded with inner and predictive mode.

As the different in kind of the coding of I frame, P frame and B frame, their compression method also has nothing in common with each other.Because the I frame is for the not prediction of service time that reduces redundancy, so, compare with the B frame with the P frame, need many figure places.

Below, be that example describes with MPEG2.If bit rate 4M bps supposes that image is the image of 30 frame/seconds.Usually, the ratio of the figure place of I, P and B frame use is 6: 3: 1.Therefore, the I frame uses about 420K position/s, and the B frame uses about 70K position/s.This is from the two-way cause of predicting fully because of the B frame.

Figure 14 is the block diagram of structure of the image prediction encoding device of expression prior art.Owing to carry out dct transform according to piece, so, nearest method for encoding images all be based on image segmentation for littler piece.In the frame interior coding, for the data image signal of input, as shown in Figure 14, at first the execution block sampling processing 1001.Secondly, for these pieces after the piece sampling processing 1001, after carrying out dct transform processing 1004, carry out quantification treatment 1005 and stroke Huffman Variable Length Code (VLC:VariableLength Coding; Entropy coding) handles 1006.On the other hand, in the predictive frame coding,, carry out moving compensation deals 1003 for the digital picture of input, and, carry out dct transform for the piece (that is, having carried out the piece of prediction) of moving compensation and handle 1004.Secondly, carry out quantification treatment 1005 and stroke Hough VLC coding (entropy coding) and handle 1006.

The redundancy this point of time between spatial redundancy in the piece of block-based dct transform processing 1004 removals or minimizing process object and moving detection and compensation deals 1002,1003 removals or minimizing consecutive frame is well-known according to the image coding technology that has earlier.In addition, stroke Huffman VLC coding or other entropy codings 1006 of carrying out after dct transform processing 1004 and quantification treatment 1005 are removed through the statistical redundancy between the dct transform coefficient of quantification.But this is handled just at the piece in the image.

As original character, digital picture spatially has bigger redundancy.This redundancy not only exists in the piece in frame, and also exists between piece and piece.But according to as can be known above-mentioned, the method for reality is not used the method for the redundancy of the interblock of removing image.

In present method for encoding images, dct transform handle 1004 or other conversion process because the formation of hardware and the constraints calculated and carry out according to piece.

The redundancy utilization in space reduces based on the conversion process of piece, and still, this is only limited in 1 piece.The redundancy of 2 adjacent interblocks is not considered well.But as long as intraframe coding always consumes a lot of figure places, redundancy just can further reduce.

In addition, handling based on the dct transform of piece and to remove or reduce the redundancy that the time of 2 adjacent interframe was removed or reduced to the redundancy in the space in the piece of processing target and moving prediction and compensation deals, is well-known according to present image coding technique.Saw-tooth sweep, stroke Huffman VLC encoding process or other entropy codings that carries out after dct transform processing and quantification treatment handled the redundancy of removing through the statistics in the dct transform coefficient after quantizing, and is only limited in 1 piece.

Digital picture comprised high spatial redundancy originally.This redundancy not only exists in the inside of piece, and outside exceeding the piece of image, also exist at interblock.In present method, except the prediction of the DC coefficient of JPEG, MPEG1 and MPEG2, use the method for the redundancy of the interblock of removing 1 image.

In MPEG1 and MPEG2, the prediction of DC coefficient is undertaken by the DC value that deducts preceding 1 encoding block from the piece of present coding.Here it is does not have the simple Forecasting Methodology of adaptability or mode switch when prediction is inappropriate.In addition, this includes only the DC coefficient.

Under the present state of this technical field, saw-tooth sweep was used all pieces before stroke is deceived sign indicating number.The data adaptive ground of not attempting in the basis scans.

Figure 22 is the block diagram of structure of the image prediction encoding device of expression prior art.In Figure 22, the image prediction encoding device of prior art has piece sampling unit 2001, dct transform unit 2003, quantifying unit 2004, saw-tooth sweep unit 2005 and entropy coding unit 2006.In this manual, the term of so-called " unit ", indication circuit device.

In frame interior coding (that is, intraframe coding), for the picture signal of input, after execution block sampling processing 2001, directly carry out dct transform and handle, and order is carried out quantification treatment 2004, saw-tooth sweep processing 2005 and entropy coding and handled 2006.On the other hand, in interframe encode (promptly, interframe encode, predictive frame coding) in, after piece sampling processing 2001, moving detection and compensation deals are carried out in unit 2011, secondly, by the view data of piece sampling 2001 and the detected value of unit 2011 are carried out subtraction, obtain predicated error by adder 2002.In addition, for this predicated error, carry out dct transform and handle 203, then, the same quantification treatment 2004, saw-tooth sweep processing 2005 and the entropy coding carried out with the frame interior coding handles 2006.

In the local decoder that is provided with in the image prediction encoding device of Figure 22, re-quantization is handled and the inverse dct transform processing is carried out in unit 2007 and 2008.In interframe encode, the moving predicted value that detects and compensate is by adder 2009 and the predicated error addition of being constructed again by

unit

2007 and 2008, this addition value represents to have carried out partly decoded image data, and this decoded image data storage is in the frame memory 2010 of local decoder.At last, bit stream is 2010 outputs from the entropy coding unit, to the other side's image prediction/decoding device transmission.

Figure 23 is the block diagram of structure of the image prediction/decoding device of expression prior art.Bit stream is decoded by variable length decoder (VLD:Variable Length Decoding) unit (or entropy decoding unit) 2021, then, for having carried out decoded image data, in

unit

2023 and 2024, carry out re-quantization processing and inverse dct transform and handle.In the frame interior coding, moving detection that forms in unit 2027 and the predicted value that compensated and the predicated error addition of being constructed again by adder 2025 form local decode image data.Carried out the frame memory 1026 of the image data storage of decoding partly to local decoder.

In present image coding technique, dct transform handle or other conversion process because the formation of hardware and the constraints calculated and be that carry out on the basis with the piece.Redundancy utilization on the space reduces based on the conversion of piece.But this is just in piece.The redundancy of adjacent interblock takes into full account.Particularly the frame interior coding that always will consume a large amount of positions is not paid attention to especially.

The 1st purpose of the present invention aims to provide image prediction encoding device and method and the image prediction/decoding device and the method for the predicted image data in span zone simply, at a high speed and accurately.

In addition, thus the 2nd purpose of the present invention is to provide with the image prediction encoding device of prior art compares image prediction apparatus and method and image prediction/decoding device and the method that the redundancy that can remove in the piece can more effectively be encoded or decode view data with image prediction/decoding device.

In addition, thus the 3rd purpose of the present invention be to provide by the decoding important conversion coefficient relevant with the character of the inside of view data concentrate on piece different zones problem and can improve image prediction encoding device and method and the image prediction/decoding device and the method for the efficient of entropy coding processing to the correct scan method of piece decision.

In addition, the 4th purpose of the present invention is to provide the recording medium of each step that writes down above-mentioned image prediction/encoding method or image prediction decoding method.

Disclosure of an invention

The image prediction encoding device of the present invention the 1st aspect is characterised in that to have:

Cutting unit is divided into the coded image data of input the view data of mutually adjacent a plurality of zonules;

The 1st generation unit, when in the view data of mutual adjacent a plurality of zonules of cutting apart by above-mentioned cutting unit, the view data of the zonule of wanting processing target being encoded, will be with the view data of above-mentioned zonule of wanting processing target the view data of the adjacent zonule that has rebuild as the view data of the intra-frame prediction zonule of above-mentioned zonule of wanting processing target, with the view data of above-mentioned intra-frame prediction zonule as the view data of optimum prediction zonule and generate the difference of the view data of the view data of zonule of above-mentioned processing target and above-mentioned optimum prediction zonule, it is the view data of difference zonule;

Coding unit is encoded to the view data of the difference zonule that generated by above-mentioned generation unit;

Decoding unit is to being decoded by the view data of the coded difference zonule of above-mentioned coding unit; With

The 2nd generation unit will be by being generated the view data of having carried out the zone that rebuilds by the view data addition of the view data of the difference zonule of above-mentioned decoding unit decodes and above-mentioned optimum prediction zonule.

In addition, the image prediction encoding device of the 2nd aspect of the present invention is characterised in that to have:

The 1st generation unit, when in mutual adjacent a plurality of zonules of cutting apart by above-mentioned cutting unit, the zonule of processing target being encoded, by input in only will be with the view data of the zonule of the above-mentioned processing target adjacent view data of rebuilding the zonule, represent above-mentioned coded image data whether the significant view data shown in the significant meaningful signal as the view data of the intra-frame prediction zonule of the zonule of above-mentioned processing target, with the view data of above-mentioned intra-frame prediction zonule as the view data of optimum prediction zonule and generate the difference of the view data of the view data of zonule of above-mentioned processing target and above-mentioned optimum prediction zonule, it is the view data of difference zonule;

Decoding unit is encoded to the view data of the difference zonule that generated by above-mentioned generation unit;

The 2nd generation unit will be by being generated the view data of having carried out the zonule that rebuilds by the view data addition of the view data of the difference zonule of above-mentioned decoding unit decodes and above-mentioned optimum prediction zonule.

In addition, the image prediction/decoding device of the 3rd aspect of the present invention is characterised in that to have:

Analytic unit is analyzed DS of image encoded and the output image differential signal imported;

Decoding unit according to the difference image signal from above-mentioned analytic unit output, is encoded to the view data of regeneration difference zonule;

Line storage, storage are used to generate the view data of view data of the intra-frame prediction zonule of appointment;

Generating unit, handle by the view data of above line memory is carried out prediction signal, will be with the view data of the above-mentioned regeneration difference zonule adjacent pictorial data of having rebuild is as the view data of intra-frame prediction zonule, export the view data of above-mentioned intra-frame prediction zonule as the view data of optimum prediction zonule; With

Adder unit, view data addition with the optimum prediction zonule of the view data of the regeneration difference zonule of above-mentioned decoding unit and above-mentioned generating unit, thereby output is used to generate the view data of the intra-frame prediction zonule of addition result, stores into simultaneously in the above line memory.

In addition, the image prediction/decoding device of the 4th aspect of the present invention is characterised in that to have:

Analytic unit is analyzed the DS of importing of image encoded, and output image differential signal, dynamic vector signal and control signal;

Decoding unit will be decoded as the view data of regeneration difference zonule from the difference image signal of above-mentioned analytic unit output;

Control unit according to the control signal output switching signal from above-mentioned analytic unit output, moves selectively in order to control moving compensating unit and generating unit;

Frame memory, the picture data of having rebuild of storage appointment;

Line storage, storage are used to generate the view data of the view data of prediction zonule in the assigned picture;

Moving compensating unit, the switching signal by replying above-mentioned control unit move compensation deals to the dynamic vector signal of input, and the rise time is predicted the view data of zonule and exports as the view data of optimum prediction zonule from above-mentioned frame memory;

Generating unit, by the switching signal of replying above-mentioned control unit the view data of above line memory is carried out prediction signal and handles, will be with the view data of the above-mentioned regeneration difference zonule adjacent view data of having rebuild as the view data of intra-frame prediction zonule, with the view data of above-mentioned intra-frame prediction zonule as the view data of optimum prediction zonule and export; With

Adder unit, by with the view data of the regeneration difference zonule of above-mentioned decoding unit and the optimum prediction zonule addition of above-mentioned generating unit, the view data of having rebuild of output addition result, simultaneously, in above-mentioned frame memory, and the image data storage of view data that will only be used for generating above-mentioned intra-frame prediction zonule is to the above line memory with above-mentioned image data storage of having rebuild.

In addition, the image prediction/decoding device of the 5th aspect of the present invention is characterised in that to have:

Analytic unit is analyzed the DS of image encoded of input, and exports compression shape signal and image difference signal;

The 1st decoding unit will be the regeneration shape signal from the compression shape signal decoding of above-mentioned analytic unit output;

Fig. 2 decoding unit will be decoded as the view data of regeneration difference zonule from the difference image signal of above-mentioned analytic unit output;

Generating unit, handle by the view data of above line memory being carried out prediction signal, only will be from the view data of having rebuild adjacent by the significant view data shown in the above-mentioned regeneration shape signal as the view data of intra-frame prediction zonule, the view data of above-mentioned intra-frame prediction zonule is exported as the pictorial data of optimum prediction zonule with the view data of above-mentioned regeneration difference zonule; With

Adder unit, the image data storage of view data that only will be used for generating above-mentioned intra-frame prediction zonule by the view data of the optimum prediction zonule addition of the view data of the regeneration difference zonule of above-mentioned the 2nd decoding unit and above-mentioned generating unit being exported addition result simultaneously is to the above line memory.

In addition, the image prediction/decoding device of the 6th aspect of the present invention is characterised in that to have:

Analytic unit is analyzed the DS of image encoded of input and is exported compression shape signal, image difference signal, dynamic vector signal and control signal;

The 2nd decoding unit will be decoded as regeneration difference zonule from the difference image signal of above-mentioned analytic unit output;

Frame memory, the reproduced picture data of storage appointment;

Moving compensating unit, according to replying and from the dynamic vector signal of above-mentioned analytic unit output from the switching signal of above-mentioned control unit output, the view data of time of origin prediction zonule by the reproduced picture data of above-mentioned frame memory are moved compensation deals, and export as the view data of optimum prediction zonule;

Generating unit, from the switching signal of above-mentioned control unit output the view data of above line memory is carried out prediction signal and handles by replying, from the view data view data adjacent, that rebuild of above-mentioned regeneration difference zonule only will be by the significant view data shown in the above-mentioned regeneration shape signal as the view data of intra-frame prediction zonule, the view data of above-mentioned intra-frame prediction zonule is exported as the view data of optimum prediction zonule; With

Adder unit, by with the view data of the regeneration difference zonule of above-mentioned the 2nd decoding unit and the optimum prediction zonule addition of above-mentioned generating unit, the view data of having rebuild of output addition result, simultaneously, above-mentioned image data storage of having rebuild in above-mentioned frame memory, and only will be used for generating the image data storage of above-mentioned intra-frame prediction zonule to the above line memory.

The image prediction encoding device of the 7th aspect of the present invention is characterised in that to have:

Sampling unit is sampled as a plurality of view data of the pixel value that comprises two-dimensional arrangements respectively with the picture signal of input;

Converter unit will be transformed to the coefficient data of the domain transformation of appointment by the view data of the piece of above-mentioned sampling unit sampling;

Block storage is stored decoded coefficient data;

Predicting unit according to the coefficient data of the piece of constructing again before in storing above-mentioned block storage into, forms the coefficient data of a plurality of prediction pieces to the coefficient data by the piece of above-mentioned converter unit transformation;

The decision unit, decision is selected and the coefficient data of the prediction piece that delivery efficiency is best from the coefficient data of a plurality of prediction pieces of being formed by above-mentioned predicting unit, and will represent that the identifier of the prediction piece of above-mentioned selection sends to image prediction/decoding device by the form of indicating bit;

The 1st adder unit deducts the coefficient data of the predicated error of exporting subtraction result from the coefficient data of the current block of now by the coefficient data of the prediction piece that will be selected by above-mentioned decision unit;

Quantifying unit is carried out quantification treatment to the coefficient data from the predicated error of above-mentioned the 1st adder unit output;

Coding unit carries out entropy coding and the coefficient data of the predicated error that will encode sends to image prediction/decoding device to the coefficient data of the predicated error of above-mentioned quantifying unit;

Inverse quantization unit is carried out re-quantization to the coefficient data of the predicated error of above-mentioned quantifying unit and is handled, and exports decoded coefficient data;

The 2nd adder unit by exporting decoded coefficient data from the coefficient data of the prediction piece of above-mentioned decision unit output and the coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, stores in the above-mentioned block storage simultaneously; With

Inverse transformation block generates decoded view data by carrying out inverse transformation from the coefficient data of the piece of above-mentioned the 2nd adder unit output.

In addition, the image prediction encoding device of the 8th aspect of the present invention is characterised in that to have:

Sampling unit is sampled as a plurality of view data of the pixel value that comprises two-dimensional arrangements with the picture signal of input;

Quantifying unit is carried out quantification treatment with the coefficient data of the domain transformation of above-mentioned converter unit;

Block storage is stored decoded coefficient data;

Predicting unit according to the coefficient data of the piece of constructing again before in storing above-mentioned block storage into, forms the coefficient data of a plurality of prediction pieces to the coefficient data by the piece of above-mentioned converter unit conversion;

Coding unit the coefficient data of the predicated error of above-mentioned the 1st adder unit is carried out entropy coding, and the coefficient data of the predicated error that will encode sends to image prediction/decoding device;

The 2nd adder unit, by the coefficient data of the predicated error of above-mentioned the 1st adder unit and the coefficient data of the current block that will quantize from the coefficient data addition of the prediction piece of above-mentioned decision unit output being decoded and exporting, store into simultaneously in the above-mentioned block storage;

Inverse quantization unit will carry out that re-quantization is handled and output from the coefficient data of the current block of above-mentioned the 2nd adder unit output; With

Inverse transformation block is carried out inverse transformation by the coefficient data with the current block of above-mentioned inverse quantization unit and is generated decoded view data.

In addition, the image prediction encoding device of the 9th aspect of the present invention is characterised in that to have:

Compensating unit is also exported by the view data of piece of input move the view data that compensation deals generate the piece predicated error of move compensation;

The 1st adder unit will deduct from the view data of the piece of above-mentioned sampling unit output from the view data of the predicated error of the piece of above-mentioned compensating unit output, and the view data of the piece of output subtraction result;

Converter unit will be transformed to the coefficient data of the domain transformation of appointment from the view data of the piece of above-mentioned the 1st adder unit output;

Block storage is stored decoded coefficient data;

Predicting unit according at the coefficient data that stores the piece of constructing again before the above-mentioned block storage into, forms the coefficient data of a plurality of prediction pieces to the coefficient data by the piece of above-mentioned converter unit conversion;

The decision unit, decision is selected and the coefficient data of the prediction piece that delivery efficiency is best from the coefficient data of a plurality of prediction pieces of being formed by above-mentioned predicting unit, and the identifier that will represent above-mentioned selected prediction piece sends to image prediction/decoding device by the form of indicating bit;

The 2nd adder unit deducts and exports the coefficient data of the predicated error of subtraction result from the coefficient data of the current block of now by the coefficient data of the prediction piece that will be selected by above-mentioned decision unit;

Quantifying unit will be carried out quantification treatment from the coefficient data of the predicated error of above-mentioned the 2nd adder unit output;

Coding unit the coefficient data of the predicated error of above-mentioned quantifying unit is carried out entropy coding, and the coefficient data of the predicated error that will encode sends to image prediction/decoding device;

Inverse quantization unit is carried out re-quantization with the coefficient data of the predicated error of above-mentioned quantifying unit and is handled, and exports decoded coefficient data;

The 3rd adder unit by exporting decoded coefficient data from the coefficient data of the prediction piece of above-mentioned decision unit output and the coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, stores in the above-mentioned block storage simultaneously;

Inverse transformation block generates decoded view data by carrying out inverse transformation from the coefficient data of the piece of above-mentioned the 3rd adder unit output; With

The 4th adder unit, by will be from the carrying out of above-mentioned compensating unit output the view data of predicated error of piece of moving compensation and decoded view data addition of above-mentioned inverse transformation block, decoded view data is exported to above-mentioned compensating unit.

In addition, the image prediction encoding device of the 10th aspect of the present invention is characterised in that to have:

Compensating unit is by moving the view data that compensation deals generated and exported the predicated error of the piece that has carried out moving compensation to the view data of piece of input;

The 1st adder unit will deduct from the view data of the piece of above-mentioned sampling unit output from the view data of the predicated error of the piece of above-mentioned compensating unit output and export the view data of the piece of subtraction result;

Block storage is stored decoded coefficient data;

The 2nd adder unit, the coefficient data by the prediction piece that will be selected by above-mentioned decision unit deducts the coefficient data of the predicated error of output subtraction result from the coefficient data of the current block of now;

Coding unit the coefficient data of the predicated error of above-mentioned the 2nd adder unit is carried out entropy coding, and the coefficient data of the predicated error that will encode sends to image prediction/decoding device;

The 3rd adder unit, by the coefficient data of the predicated error of above-mentioned the 2nd adder unit and the coefficient data of the current block that will quantize from the coefficient data addition of the prediction piece of above-mentioned decision unit output being decoded and exporting, store into simultaneously in the above-mentioned block storage;

Inverse quantization unit will carry out that re-quantization is handled and output from the coefficient data of the current block of above-mentioned the 3rd adder unit output;

Inverse transformation block is carried out inverse transformation by the coefficient data with the current block of above-mentioned inverse quantization unit and is generated decoded view data; With

The image prediction/decoding device of the 11st aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 7th aspect of the present invention is provided with accordingly, it is characterized in that having:

Extract the unit out, from the reception data that receive by above-mentioned image prediction encoding device, extract indicating bit out;

Block storage is stored decoded coefficient data;

Other predicting unit, according to the prediction piece of representing by the indicating bit of above-mentioned extraction unit extraction, use coefficient data of decoded before storing above-mentioned block storage into, to the coefficient data of the current block of now of in above-mentioned reception data, comprising, generate and the coefficient data of prediction of output piece;

Decoding unit carries out the entropy decoding with above-mentioned reception data, and exports the coefficient data of decoded predicated error;

Inverse quantization unit will carry out that re-quantization is handled and output from the coefficient data of the predicated error of above-mentioned decoding unit output;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned other predicting unit output and coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, the coefficient data of the current block of now is decoded and export, store into simultaneously in the above-mentioned block storage; With

Other inverse transformation block will be carried out inverse transformation from the coefficient data of the current block of above-mentioned the 3rd adder unit output, and be exported the view data of decoded current block.

In addition, the image prediction/decoding device of the 12nd aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 8th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

Other predicting unit, according to the prediction piece of representing by the indicating bit of above-mentioned extraction unit extraction, use in storing above-mentioned block storage into before decoded coefficient data, the coefficient data of the current block of now of comprising in above-mentioned reception data is generated and the coefficient data of prediction of output piece;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned predicting unit output and coefficient data addition from the predicated error of above-mentioned decoding unit output, the coefficient data of the current block of now is decoded and export, store into simultaneously in the above-mentioned block storage;

Inverse quantization unit will be carried out re-quantization from the coefficient data of the predicated error of above-mentioned the 3rd adder unit output and be handled; With

Other inverse transformation block will be carried out inverse transformation from the coefficient data of the current block of above-mentioned inverse quantization unit output, and be exported the view data of decoded current block.

In addition, the image prediction/decoding device of the 13rd aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 9th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

Other predicting unit, according to the prediction piece of representing by the indicating bit of above-mentioned extraction unit extraction, use in storing above-mentioned block storage into before decoded coefficient data, to the coefficient data of the current block of now of in above-mentioned reception data, comprising generate and the coefficient data of prediction of output piece not;

The 3rd adder unit, by will from the coefficient data of the prediction piece of above-mentioned other predicting unit output with from the coefficient data addition of the predicated error of above-mentioned inverse quantization unit output the coefficient data of the current block of now is decoded and exports, store into simultaneously in the above-mentioned block storage;

Other inverse transformation block will be carried out inverse transformation from the coefficient data of the current block of above-mentioned the 3rd adder unit output and exported the view data of decoded current block;

Other compensating units are by exporting the moving prediction error data that compensates to move compensation deals from the view data of the current block of above-mentioned other inverse transformation block output; With

The 5th adder unit will deduct from the view data of the current block of above-mentioned other inverse transformation block outputs from the prediction error data of the moving compensation of above-mentioned other compensating units output and export decoded view data of subtraction result.

In addition, the image prediction/decoding device of the 14th aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 10th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned predicting unit output and coefficient data addition from the predicated error of above-mentioned decoding unit output, the coefficient data of the current block of now is decoded and export, simultaneously, be stored in the above-mentioned block storage;

Inverse quantization unit will carry out that re-quantization is handled and output from the coefficient data of the predicated error of above-mentioned the 3rd adder unit output;

Other inverse transformation block will be carried out inverse transformation from the coefficient data of the current block of above-mentioned inverse quantization unit output and be exported the view data of decoded current block;

The 5th adder unit will deduct decoded view data of output subtraction result from the prediction error data of the moving compensation of above-mentioned other compensating unit output from the view data of the current block of above-mentioned other inverse transformation block outputs.

The image prediction encoding device of the 15th aspect of the present invention is characterised in that to have:

Block storage is stored decoded coefficient data;

The decision unit, decision is selected and the coefficient data and the scan method of the prediction piece that delivery efficiency is best from the coefficient data of a plurality of prediction pieces of being formed by above-mentioned predicting unit, and the identifier that will represent the prediction piece of above-mentioned selection and scan method sends to image prediction/decoding device by the form of indicating bit;

Quantifying unit will be carried out quantification treatment from the coefficient data of the predicated error of above-mentioned the 1st adder unit output;

Scanning element, to the coefficient data of the predicated error of above-mentioned quantifying unit, according to carrying out scan process by the scan method of above-mentioned decision unit decision, and the coefficient data of the predicated error of output scanning after handling;

Coding unit will carry out entropy coding from the coefficient data of the predicated error after the scan process of above-mentioned scanning element output, and the coefficient data of the predicated error that will encode sends to image prediction/decoding device;

The 2nd adder unit by exporting decoded coefficient data from the coefficient data of the prediction piece of above-mentioned decision unit output and the coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, is stored in the above-mentioned block storage simultaneously; With

Inverse transformation block generates decoded view data by the coefficient data from the piece of above-mentioned the 2nd adder unit output is carried out inverse transformation.

In addition, the image prediction encoding device of the 16th aspect of the present invention is characterised in that to have:

Converter unit will be transformed to the coefficient data of the domain transformation of appointment by a plurality of view data of above-mentioned sampling unit sampling;

Quantifying unit will quantize from the coefficient data of the next shift zone of above-mentioned converter unit;

Block storage is stored decoded coefficient data;

Scanning element, to the coefficient data of the predicated error of above-mentioned the 1st adder unit, according to carrying out scan process by the scan method of above-mentioned decision unit decision, and the coefficient data of the predicated error of output scanning after handling;

The 2nd adder unit, by with the coefficient data of the predicated error of above-mentioned the 1st adder unit and coefficient data addition from the prediction piece of above-mentioned decision unit output, the coefficient data of the current block that quantized is decoded and export, be stored to simultaneously in the above-mentioned block storage;

Inverse transformation block is carried out inverse transformation by the coefficient data with the current block of above-mentioned re-quantization and is generated decoded view data.

In addition, the image prediction encoding device of the 17th aspect of the present invention is characterised in that to have:

Block storage is stored decoded coefficient data;

The decision unit, decision is selected and the coefficient data and the scan method of the prediction piece that delivery efficiency is best from the coefficient data of a plurality of prediction pieces of being formed by above-mentioned predicting unit, and the identifier that will represent above-mentioned selected prediction piece and scan method sends to image prediction/decoding device by the form of indicating bit;

The 3rd adder unit by exporting decoded coefficient data from the coefficient data of the prediction piece of above-mentioned decision unit output and coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, simultaneously, is stored in the above-mentioned block storage;

Inverse transformation block by carrying out inverse transformation from the coefficient data of the piece of above-mentioned the 3rd adder unit output, generates decoded view data; With

In addition, the image prediction encoding device of the 18th aspect of the present invention is characterised in that to have:

Compensating unit by the view data of piece of input move compensation deals, generates and exports the view data of the predicated error of the piece that has carried out moving compensation;

The 1st adder unit will deduct the view data of the piece of output subtraction result from the view data of the predicated error of the piece of above-mentioned compensating unit output from the view data of the piece of above-mentioned sampling unit output;

Block storage is stored decoded coefficient data;

Scanning element, to the coefficient data of the predicated error of above-mentioned the 2nd adder unit, according to carrying out scan process by the scan method of above-mentioned decision unit decision, and the coefficient data of the predicated error of output scanning after handling;

The 3rd adder unit, by with the coefficient data of the predicated error of above-mentioned the 2nd adder unit and coefficient data addition from the prediction piece of above-mentioned decision unit output, the coefficient data of the current block that quantized is decoded and export, simultaneously, be stored in the above-mentioned block storage;

The image prediction/decoding device of the 19th aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 15th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

Inverse scan unit, the scan method of representing according to the indicating bit of being extracted out by above-mentioned extraction unit is carried out inverse scan to the coefficient data from the predicated error of above-mentioned decoding unit output and is handled, and the coefficient data of the predicated error of output inverse scan after handling;

Inverse quantization unit, the coefficient data of the predicated error after will handling from the inverse scan of above-mentioned inverse scan unit output carries out re-quantization and handles and export;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned other predicting unit output and coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, the coefficient data of the current block of now is decoded and export, simultaneously, be stored in the above-mentioned block storage; With

In addition, the image prediction/decoding device of the 20th aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 16th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned predicting unit output and coefficient data addition from the predicated error of above-mentioned inverse scan unit output, the coefficient data of the current block of now is decoded and export, be stored to simultaneously in the above-mentioned block storage;

Inverse quantization unit will carry out that re-quantization is handled and output from the coefficient data of the predicated error of above-mentioned the 3rd adder unit output; With

In addition, the image prediction/decoding device of the 21st aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 17th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned other predicting unit output and coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, the coefficient data of the current block of now is decoded and export, simultaneously, be stored in the above-mentioned block storage;

Other inverse transformation block will be carried out inverse transformation from the coefficient data of the current block of above-mentioned the 3rd adder unit output, and be exported the view data of decoded current block;

Other moving compensating units are by move compensation deals, the moving prediction error data that compensates of output to the view data from the current block of above-mentioned other inverse transformation block output; With

The 5th adder unit will deduct decoded view data of output subtraction result from the prediction error data of the moving compensation of above-mentioned other compensating unit output from the view data of the current block of above-mentioned other inverse transformation block output.

In addition, the image prediction/decoding device of the 22nd aspect of the present invention is the image prediction/decoding device that the image prediction encoding device with the 18th aspect of the present invention is provided with accordingly, it is characterized in that having:

Block storage is stored decoded coefficient data;

The 3rd adder unit, by will be from the coefficient data of the prediction piece of above-mentioned predicting unit output and coefficient data addition from the predicated error of above-mentioned inverse scan unit output, the coefficient data of the current block of now is decoded and export, simultaneously, be stored in the above-mentioned block storage;

Other inverse transformation block will be carried out inverse transformation from the coefficient data of the current block of above-mentioned inverse quantization unit output, and be exported the view data of decoded current block;

In addition, the image prediction/encoding method of the 23rd aspect of the present invention is characterised in that: comprise the step that each unit of above-mentioned image prediction encoding device is replaced into each step respectively.

In addition, the image prediction decoding method of the 24th aspect of the present invention is characterised in that: comprise the step that each unit of above-mentioned image prediction/decoding device is replaced into each step respectively.

In addition, the recording medium of the 25th aspect of the present invention is the program recording medium that record comprises each step of above-mentioned image prediction/encoding method.

In addition, the recording medium of the 26th aspect of the present invention is the program recording medium that record comprises each step of above-mentioned image prediction decoding method.

The simple declaration of accompanying drawing

Fig. 1 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 1.

Fig. 2 is the input picture that will be input to the image prediction encoding device of Fig. 1 ideograph when being divided into 8 * 8 piece.

Fig. 3 is the input picture that will be input to the image prediction encoding device of Fig. 1 ideograph when being divided into the Delta Region.

Fig. 4 is the block diagram of structure that is illustrated in the embodiment 1 of the prediction signal generator that uses in the image prediction encoding device of Fig. 1.

Fig. 5 is the block diagram of structure that is illustrated in the embodiment 2 of the prediction signal generator that uses in the image prediction encoding device of Fig. 1.

Fig. 6 is the block diagram of structure that is illustrated in the embodiment 3 of the prediction signal generator that uses in the image prediction encoding device of Fig. 1.

Fig. 7 is the block diagram of structure that is illustrated in the embodiment 4 of the prediction signal generator that uses in the image prediction encoding device of Fig. 1.

Fig. 8 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 2.

Fig. 9 is the ideograph of an example of input picture with significant pixel of the input picture of the image prediction encoding device of input Fig. 1 and Fig. 8, i.e. expression.

Figure 10 is the ideograph of an example of input picture with significant pixel of the input picture of the image prediction encoding device of input Fig. 1 and Fig. 8, i.e. expression.

Figure 11 is the ideograph of an example of the input picture of the image prediction encoding device of input Fig. 1 and Fig. 8, the i.e. expression input picture that do not have significant pixel.

Figure 12 is the block diagram of structure of the image prediction/decoding device of the expression embodiment of the invention 3.

Figure 13 is the block diagram of structure of the image prediction/decoding device of the expression embodiment of the invention 4.

Figure 14 is the block diagram of structure of the image prediction encoding device of expression prior art.

Figure 15 is the ideograph of the image in the adaptive dct transform zone that is used to illustrate that intra-frame prediction is used.

Figure 16 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 5.

Figure 17 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 6.

Figure 18 is the block diagram of structure of the dct transform regional prediction circuit of expression Figure 16 and Figure 17.

Figure 19 is the ideograph of image of an example of coding method of DC/AC prediction of the dct transform regional prediction circuit of expression Figure 18.

Figure 20 is the block diagram of structure of the image prediction/decoding device of the expression embodiment of the invention 7.

Figure 21 is the flow chart of coding/decoding method of DC/AC prediction of the image prediction/decoding device of expression Figure 20.

Figure 22 is the block diagram of structure of the image prediction encoding device of expression prior art.

Figure 23 is the block diagram of structure of the image prediction encoding device of expression prior art.

Figure 24 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 8.

Figure 25 is the block diagram of structure of the image prediction/decoding device of the expression embodiment of the invention 8.

Figure 26 is the ideograph of the image of the structure of the macro block of frame of expression embodiment 8 and piece and expression block predication.

Figure 27 is the ideograph that is used for illustrating the image of the horizontal sweep order that the coefficient scanning at embodiment 8 uses.

Figure 28 is the ideograph that is used for illustrating the image of the vertical scanning order that the coefficient scanning at embodiment 8 uses.

Figure 29 is the ideograph that is used for illustrating the image of the saw-tooth sweep order that the coefficient scanning at embodiment 8 uses.

Figure 30 is the flow chart that is illustrated in the pattern decision processing of using among the embodiment 8.

Figure 31 is the ideograph of image of relation of piece of the default mode decision of expression embodiment 8.

The form of the best that carries out an invention

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

The 1st embodiment group comprises embodiment 1～embodiment 4.

(embodiment 1)

Fig. 1 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention.

In Fig. 1, the 101st, input terminal, 102 is the 1st adders, the 103rd, encoder, the 106th, lead-out terminal, the 107th, decoder, 110 is the 2nd adders, the 111st, line storage, the 112nd, prediction signal generator.

Below, the structure and the action of image prediction encoding device are described.To be input to input terminal 101 as the view data of the process object of encoding.Here, the view data of input is divided into a plurality of adjacent zonules.

The image of the view data of importing when expression is divided into 8 * 8 zonule in Fig. 2, the image of the view data of importing when expression is divided into the triangle zonule in Fig. 3.The view data of a plurality of zonules is encoded in proper order, and the view data of the zonule of processing target is input to adder 102 by input terminal 101 and line 113.On the other hand, the view data of intra-frame prediction zonule takes place in prediction signal generator 112, and the view data that is taken place is exported to adder 102 by line 121 as the view data of optimum prediction zonule.

Deduct in the pixel value of adder 102 is imported the respective pixel value of the optimum prediction zonule of prediction signal generator 112 from the zonule of processing target view data, generate the view data of the difference zonule of subtraction result, to encoder 103 outputs, carry out compressed encoding and handle.In the present embodiment, encoder 103 has dct transform device 104 and quantizer (Q) 105, and the view data of difference zonule is transformed to the picture signal of frequency field by dct transform device 104, obtains dct transform coefficient.Secondly, dct transform coefficient carries out quantification treatment by quantizer 105.The DCT coefficient of the zonule that has quantized,, for example stores on the recording medium such as CD or by communication line and transmits (not shown) and then after being transformed to the code of a string variable length or fixed length to lead-out terminal 106 output by line 116.

Simultaneously, the DCT coefficient input decoder 107 of the zonule that has quantized, here, it is inverse dct transform devices 109 that this decoder 107 has inverse quantizer 108, is the view data of decompress(ion) difference zonule with the image data decoding of the zonule imported.In the present embodiment, the DCT coefficient of the zonule of input is undertaken after re-quantization handles by inverse quantizer 108, and the DCT coefficient of re-quantization is transformed to the picture signal of area of space by inverse discrete cosine transformer (below, be called the inverse dct transform device) 109.The view data of the decompress(ion) difference zonule that obtains like this is to adder 110 outputs, the optimum prediction picture signal that adder 110 will be exported by line 121 and line 122 from prediction signal generator 112 and the view data addition of decompress(ion) difference zonule, generate the view data of the zonule rebuild, the pixel value of having rebuild that is used for generating the intra-frame prediction picture signal in the view data with the zonule that rebuild stores line storage 111 into.Prediction signal generator 112 such view data that the intra-frame prediction zonule takes place as described below.That is, the pixel value of the view data of having rebuild that the view data of the zonule of 112 generations of prediction signal generator and processing target is adjacent is as the pixel value of the view data of intra-frame prediction zonule.

In Fig. 2, to establish piece 200 and be the zonule of processing target, the pixel value of the then adjacent view data of having rebuild is a ₀, a ₁, a ₂, a ₃, a ₄, a ₅, a ₆, a ₇, b ₀, b ₁, b ₂, b ₃, b ₄, b ₅, b ₆, b ₇In Fig. 3, to establish triangle 301 and be the zonule of processing target, the pixel value of the then adjacent view data of having rebuild is g ₀, g ₁, g ₂, g ₃, g ₄, f ₀, f ₁, f ₂, f ₃, f ₄, f ₅, f ₆, f ₈In addition, the triangle 300 of establishing Fig. 3 is the zonule of processing target, then adjacent rebuild the pixel value of view data be e ₀, h ₀, h ₁..., h ₄These pixel values are stored in the line storage 111.Prediction signal generator 112 access line memories 111, with the pixel value of adjacent view data as the pixel value of the view data of intra-frame prediction zonule and read.

Fig. 4 and Fig. 5 are illustrated respectively in the block diagram of the structure of the embodiment 1 of the prediction signal generator that uses in the image prediction encoding device of Fig. 1 and embodiment 2.

In Fig. 4, for the zonule of processing target, adjacent pixels value a in the horizontal direction ₀, a ₁, a ₂..., a ₆, a ₇From line storage 111 input prediction signal generators 112, the generator 401 in the prediction signal generator 112 generates the view data 403 of intra-frame prediction zonule by same pixel being exported repeatedly for example 8 times in the horizontal direction.Here, the view data 403 of intra-frame prediction zonule is used when not existing in the vertical direction adjacent pixels for the zonule of processing target.

In Fig. 5, for the zonule of processing target, at vertical direction adjacent pixels value b ₀, b ₁, b ₂, b ₆, b ₇From line storage 111 input prediction signal generators 112, the generator 402 in the prediction signal generator 112 generates the view data 404 of intra-frame prediction zonule by pixel being exported repeatedly for example 8 times in vertical direction.Here, the view data 404 of intra-frame prediction zonule is used when not existing in the horizontal direction adjacent pixels for the zonule of processing target.When existence simultaneously and horizontal direction and vertical direction adjacent pixels value, embodiment 3 as shown in Figure 6 generates the view data of intra-frame prediction zonule like that.

In Fig. 6, the view data 404 of the view data 403 (with reference to figure 5) of the intra-frame prediction zonule that will be taken place by generator 401 and the intra-frame prediction zonule that taken place by generator 402 is input to adder 500, adder 500 asks average by 2 view data sums with 2 removal inputs with these 2 view data.Like this, export the pixel of adjacent regeneration repeatedly and ask average calculating operation by adder 500 by generator 401,402, so, the view data of intra-frame prediction zonule can be generated at high speed.In addition, also can be by 2 adjacent view data pixel values are carried out the view data that linear interpolation generates the intra-frame prediction zonule.

In Fig. 7, for the zonule of processing target, the pixel value a of adjacent view data aspect horizontal ₀, a ₁, a ₂..., a ₆, a ₇From line storage 111 input generators 401, generator 401 by in the horizontal direction repeatedly output pixel generate the view data of the 1st intra-frame prediction zonule.On the other hand, for the zonule of processing target, at vertical direction adjacent pixels value b ₀, b ₁, b ₂..., b ₆, b ₇Be input to generator 402 from line storage 111, generator 402 by in vertical direction repeatedly output pixel generate the view data of the 2nd intra-frame prediction zonule.The view data of the 1st intra-frame prediction zonule and the view data of the 2nd intra-frame prediction zonule are input to adder 500,, generate the view data of the 3rd intra-frame prediction zonule by these 2 view data are asked on average.

On the other hand, the view data of the zonule of processing target is by line 616 error originated from input calculators 601,602,603.Here, the view data error originated from input calculator 601 of the view data of above-mentioned the 1st intra-frame prediction zonule and the zonule of processing target, Error Calculator 601 is calculated the absolute value of the error of these 2 view data, i.e. the 1st absolute error, and exports to comparator 604.In addition, view data error originated from input calculator 602 with the zonule of the view data of above-mentioned the 2nd intra-frame prediction zonule and processing target, Error Calculator 602 is calculated the absolute value of the error of these 2 view data, i.e. the 2nd absolute error, and exports to comparator 604.In addition, the view data of the zonule of the view data of above-mentioned the 3rd intra-frame prediction zonule and processing target is input to Error Calculator 603, Error Calculator 603 is calculated the absolute value of the error of these 2 view data, i.e. the 3rd absolute error, and exports to comparator 604.

3 absolute errors that comparator 604 will be imported compare mutually, determine absolute error reckling and control switch 605 so that the view data of intra-frame prediction zonule that will be corresponding with it is exported to line 121.Comparator 604 will be used to discern the identifier of view data of the 1st, the 2nd and the 3rd intra-frame prediction zonule simultaneously by the device output of line 615 to receiver side or regeneration side.In receiver side or regeneration side, determine the view data of intra-frame prediction zonule uniquely by this identifier.Like this, the view data of the intra-frame prediction zonule by the use error minimum, the differential signal in the time of can suppressing to encode, thus can reduce the generation figure place.

(embodiment 2)

Fig. 8 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 2, and the part identical with Fig. 1 is marked with identical symbol.

The image prediction encoding device of Fig. 8 is compared with the image prediction encoding device of Fig. 1, it is characterized in that: increased moving detector 700, moving compensator 701, optimal mode selector 703 and frame memory 702.

Below, the structure of the image prediction encoding device of key diagram 8 and action.

The same with embodiment 1, the view data of the zonule of the processing target of input is by input terminal 101 input summers 102, adder 102 deducts the view data of the zonule of processing target from the view data of optimal mode selector 703 by the optimum prediction zonule of line 121 inputs after, with the view data of subtraction result to encoder 103 outputs.Encoder 103 is when the view data of the subtraction result that will import is carried out compressed encoding and passed through lead-out terminal 106 outputs, with the view data of the zonule of compressed encoding after decoder 107 output and carrying out uncompressed encoding, to adder 110 outputs, with the view data of uncompressed encoding and the view data addition of optimum prediction zonule.

Secondly, the same with embodiment 1, the pixel value of the view data that will use for the view data that generates the intra-frame prediction zonule stores in the line storage 111, on the other hand, the pixel value of whole pixels of having rebuild is stored in the frame memory 702.

When importing the view data of next image by input terminal 101, with the view data of the zonule of processing target and the moving detector 700 of the view data of having rebuild input of frame memory 702 storages, moving detector 700 is pressed the activity of method detected image such as piece coupling, by line 705 output dynamic vectors.The dynamic vector of output is when carrying out for example Variable Length Code and storage or transmitting (not shown), to moving compensator 701 transmission.Moving compensator 701 rise time from the image of having rebuild of the indicated frame memory 702 of dynamic vector is predicted the view data of zonule, and to 703 outputs of optimal mode selector.Moving detect handle and moving compensation deals in, the place ahead prediction, rear prediction and bi-directional predicted are arranged, these methods are open in No. 5193004 specification of United States Patent (USP) for example.

On the other hand, the same with embodiment 1, prediction signal generator 112 is in the view data that the intra-frame prediction zonule takes place and to 703 outputs of optimal mode selector.The view data of zonule that simultaneously, will processing target is to 703 outputs of optimal mode selector.Optimal mode selector 703 (for example is chosen in the view data of zonule of processing target error from the view data of the view data of intra-frame prediction zonule and time prediction zonule, the absolute value sum of the difference of each pixel) Zui Xiao view data is exported as the view data of optimum prediction zonule the view data of selecting to adder 102.In addition, the identifier of view data of expression having been selected which prediction zonule by line 709 to receiving the output of a side or the side of regenerating and transmitting.

Like this, by intra-frame prediction being imported in the predictive mode image encoded data, the dynamic vector between just needn't transmission frame, so, can further cut down figure place.

In

embodiment

1 and 2, be the situation that in entire image, has significant pixel.Sometimes, in picture, there are significant pixel and insignificant pixel simultaneously.For example, in the image that utilizes chroma key to take, the pixel of expression subject is significant pixel, and the pixel in expression zones such as blueness as a setting is exactly insignificant pixel.The regeneration that by the structure of significant object and shape thereof being encoded and transmit, can carrying out object is the unit shows.For such input picture, when generating the view data of intra-frame prediction zonule, just can not use insignificant pixel value with prediction signal generator 112.

Fig. 9～Figure 11 is the ideograph that expression has the input picture of significant pixel and insignificant pixel.In the present embodiment, whether meaningful for remarked pixel, use shape signal.Shape signal laid that appointed method is carried out compressed encoding and to receiving the transmission of a side or the side of regenerating.As shape is carried out Methods for Coding, methods such as link coding are arranged.The shape signal that has compressed is carried out decompress(ion) regeneration again, and with shape signal such intra-frame prediction signal that generates as described below of regenerating.

In Fig. 9, pattern curve 800 is boundary lines, and the direction shown in the arrow is the inside of object, and the view data of the inside of object is made of significant pixel.In the pixel of having regenerated adjacent with the zonule 802 of processing target, b ₄, b ₅, b ₆, b ₇Be significant pixel, only make these pixel values repeatedly as the pixel value of the intra-frame prediction zonule of the zonule 802 of processing target.

In addition, in Figure 10, pattern curve 804 is boundary lines, and the direction shown in the arrow is the inside of object, and the view data of the inside of object is made of significant pixel.In the pixel of having regenerated adjacent with the zonule 805 of processing target, a ₄, a ₅, a ₆, a ₇Be significant pixel, only make these pixel values repeatedly as the pixel value of the intra-frame prediction zonule of the zonule 805 of processing target.

In addition, in Figure 11, curve 808 is boundary lines, and the direction shown in the arrow is the inside of object, and the view data of the inside of object is made of significant pixel.In the pixel of having regenerated adjacent with the zonule 810 of processing target, a ₅, a ₆, a ₇, b ₄, b ₅, b ₆, b ₇Be significant pixel, only export these pixel values repeatedly,, will ask average value pixel value these pixel values as the intra-frame prediction zonule of the zonule 810 of processing target in 2 places that pixel value is overlapping.

In Figure 11, for example get the pixel Z of the zonule 810 of processing target ₇₇Value be a ₇And b ₇Mean value.In addition, do not have the place yet, just get the mean value of 2 pixel values adjacent with vertical direction with horizontal direction at 1 pixel value.For example, the plain Z of capture ₁₄Value be a ₅And b ₄Mean value.Generate the view data of the predictive pictures zonule of image with arbitrary shape like this.

In above embodiment, illustrated to be divided into foursquare zonule, still, the invention is not restricted to this, also can be the same with Fig. 3 be leg-of-mutton zonule with picture segmentation.At this moment, image processing is carried out too.

In addition,, also can only use significant pixel value to average as other embodiment, and with the pixel value of this mean value as the intra-frame prediction zonule.Particularly, in Fig. 9, be exactly calculating pixel b ₄, b ₅, b ₆, b ₇Mean value, and with the mean value that calculates pixel value as the intra-frame prediction zonule.In Figure 10, be exactly calculating pixel a ₄, a ₅, a ₆, a ₇Mean value, and with the mean value that calculates pixel value as the intra-frame prediction zonule.In Figure 11, calculate a exactly ₅, a ₆, a ₇, b ₄, b ₅, b ₆, b ₇Mean value, and as the pixel value of intra-frame prediction zonule.

(embodiment 3)

In Figure 12, the 901st, input terminal, the 902nd, data-analyzing machine, the 903rd, decoder, the 906th, adder, the 907th, lead-out terminal, the 908th, controller, the 909th, moving compensator, the 910th, prediction signal generator, the 911st, line storage, the 912nd, frame memory.

Below, the structure and the action of the image prediction/decoding device of Figure 12 are described.In Figure 12, with the view data input data analysis device 902 of compressed encoding, data-analyzing machine 902 is analyzed the view data of input, and pass through line 915 view data of compressed differential zonule is exported to decoder 903, in addition, by line 926 control signal is exported to controller 908, in addition, above-mentioned dynamic vector (only when existing) is exported to moving compensator 909.Decoder 903 has inverse quantizer 904 and inverse dct transform device 905, reverts to the view data of decompress(ion) difference zonule behind the view data decompress(ion) with the difference zonule compressed.

In the present embodiment, the view data of the difference zonule of having compressed is carried out the re-quantization processing by inverse quantizer 904, and the view data of the frequency field behind the re-quantization is changed to the view data of area of space by inverse dct transform device 905.View data input summer 906 with the decompress(ion) difference zonule after the conversion, the view data addition of the optimum prediction zonule that adder 906 sends view data and the driven compensator 923 of the decompress(ion) difference zonule of input or prediction signal generator 922 by switch 913 and line 924, the view data of the regeneration zonule of generation addition results.Adder 906 is passed through line 917 in lead-out terminal 907 outputs in the view data that will rebuild, stores the frame storage into and is equipped with in the device 912.In addition, the pixel value of the view data that will use for the image that generates the intra-frame prediction zonule stores in the line storage 911.

The view data of optimum prediction zonule is determined by controller 908 according to the control signal of data-analyzing machine 902, and the switching of control switch 913.When the view data of intra-frame prediction zonule is selected by controller 908, switch 913 just is connected line 924 with line 922, prediction signal generator 910 is replied the control signal of controller 908 and access line memory 911, with adjacent regeneration pixel value as the pixel value of intra-frame prediction zonule and export.Details about the action of prediction signal generator 910 has been described in detail with reference to Fig. 4, Fig. 5 and Fig. 6.In addition, when the view data of time prediction zonule is selected by controller 908, switch 913 just is connected line 924 with line 923, moving compensator 909 is replied the control signal of controller 908, according to the dynamic vector that sends by line 925 from data-analyzing machine 902, by the view data of frame memory 912 is moved compensation deals, time of origin is predicted the view data of zonule, and exports to adder 906 by switch 913 and line 924.

(embodiment 4)

Figure 13 is the block diagram of structure of the image prediction/decoding device of the expression embodiment of the invention 4.In Figure 13, be marked with identical symbol for the part identical with Figure 12.The image prediction/decoding device of Figure 13 is characterised in that, except the basic structure of the image prediction/decoding device of Figure 12, increased decoded shape device 990.Because the elemental motion of the image prediction/decoding device of Figure 13 is also identical with Figure 12, so, below, only describe different actions in detail.

In the present embodiment, in the view data of compressed encoding, comprise the shape data of compressed encoding.Data-analyzing machine 902 is exported to shape decoder 990 after extracting this shape data out, replys this action, and decoded shape device 990 carries out decompress(ion) regeneration with shape signal.When receiving the transmission of a side or the side of regenerating, import prediction signal generator 910 at the shape signal of having rebuild.Prediction signal generator 910 generates the view data of intra-frame prediction zonule as reference Fig. 9～Figure 11 explanation according to this shape signal of having rebuild.Like this, just can generate the view data of the intra-frame prediction zonule of image, can view data be decoded and regenerate receiving a side or the side of regenerating with arbitrary shape.

Embodiment

3 and 4 feature are to have line storage 911.If there is not line storage 911, just must be from frame memory 912 access be used to generate the pixel of the view data of intra-frame prediction zonule.Pixel generation forecast signal for adjacent zonule must write at high speed or read frame memory.By special-purpose line storage or buffer are set, do not use frame memory at a high speed just can generate the view data of intra-frame prediction zonule at high speed.

In above embodiment, the mean value of a plurality of pixel values also can be the weighted average of appointment.

As mentioned above, according to the 1st embodiment group of the present invention, only will be with the view data of the zonule of the processing target adjacent pixel value of having rebuild is as the pixel value of intra-frame prediction signal, just can generate high-precision prediction signal simply, thereby can obtain to cut down the peculiar effect of the figure place of intraframe coding with the operand lower than prior art.In addition, owing to being provided with line storage 911 in order to store the pixel value of having rebuild that uses that is used to generate the intra-frame prediction signal, so, access pixel value at high speed, thus the intra-frame prediction signal can be generated at high speed.

The 2nd embodiment group

The 2nd embodiment group comprises embodiment 5～7.

The present invention is in view of the problem of prior art, by not only remove between 2 images or 1 image in the inside of 2 pieces between redundancy, and remove the redundancy of 2 interblocks in 1 image, find further to have improved image coding efficiency.

The present inventor finds that the dct transform coefficient of the same position of adjacent block is very similar as a rule.When particularly closely similar or when comprising identical picture pattern (for example straight line, angle and other), find the degree of approximation height with organizing of the corresponding original image of 2 pieces.According to information theory, identical information just means redundancy.

Exceed this redundancy that exists in the external dct transform of the piece zone, can utilize the adaptive intra-prediction (infra-frame prediction) of former piece and remove or reduce significantly.And VLC entropy coding is after this handled the little entropy that can utilize prediction and is further realized high code efficiency.As the prediction result in this dct transform zone, can reduce significantly to the input of the redundant data of VLC entropy coding circuit.Therefore, can expect to save a lot of positions.Therefore, the image quality of image encoded data is improved clearly.

Present embodiment of the present invention provides the mode of the dct transform coefficient of suitably predicting other pieces.Utilize this mode to remove the redundancy that exceeds adjacent piece and exist, the further entropy of the dct transform coefficient that quantized of minimizing, the result just can reduce and is used for the dct transform coefficient needed figure place of encoding.

According in the former adjacent piece at the dct transform coefficient of same position, can the prediction processing target be the dct transform coefficient of the current block (below, be called current block) of now.Adjacent piece has been decoded when handling.That is the 1DC coefficient prediction 1DC coefficient in decoded adjacent 1 piece before utilizing.In addition, the 2nd coefficients AC 1 is predicted according to the 2nd coefficients AC 1 in identical decoded.Below, handle equally like this.By using this method, for the discrete cosine transform block of encoding in now, can be according to the decoded adjacent piece of obtaining several predictions that is positioned at upper left side, oblique left side, oblique upper right side and upside.For these pieces of having predicted, check by carrying out actual entropy coding.And, after selecting to have the prediction piece of figure place still less, carry out entropy coding, and transmit to the image prediction/decoding device that receives a side or the side of regenerating with additional indicating bit.To image prediction/decoding device report current block is according to which adjacent piece to predict.

The method of present embodiment of the present invention can be predicted the dct transform coefficient of current block.This dct transform coefficient has the high correlation of dct transform coefficient of the piece adjacent with other usually.Its reason is, dct transform is in the identical value that dct transform coefficient is provided or the tendency of same distribution for same piece image.

For the view data of promptly having imported as the frame of frame interior or time prediction, at first carry out dct transform usually and handle according to piece.After obtaining the dct transform coefficient of current block, can be before quantification or carry out the prediction processing in dct transform zone after quantizing.

As shown in figure 15, the dct transform coefficient of current block is decoded, can be that upper left piece B1, top piece B2, upper right piece B3, the piece B4 on the left side predict according to adjacent piece.The piece of having predicted can by with whole dct transform coefficients of current block from be in same position before whole dct transform coefficients of adjacent piece deduct and obtain.In addition, also can obtain by partly deducting the whole dct transform coefficient of dct transform coefficient replacement.

If before quantizing, predict that just the dct transform coefficient of having predicted of the different masses that will predict carries out quantification treatment.Secondly, for dct transform coefficient, carry out entropy coding and handle.This entropy coding is handled identical with the processing of image prediction encoding device, checks that the piece which has been predicted uses as the low level position.

Select to use the prediction piece of low level position, selected prediction piece carries out entropy coding with the indicating bit for prediction decision notice image prediction/decoding device.

In image prediction/decoding device, use indicating bit that the piece of having predicted is decoded.Promptly, for 1 piece, after the dct transform coefficient of having predicted being carried out contrary entropy decoding, the dct transform coefficient corresponding with this piece obtains by the dct transform coefficient and the above-mentioned decoded dct transform coefficient addition of the decoded adjacent block in front that will be represented by indicating bit.At last, inverse dct transform is handled and is applied to the decoded dct transform coefficient corresponding with each piece, can obtain decoded view data.

Present embodiment of the present invention except the spatial redundancy removed by the such conversion of common dct transform, by moving detect and statistical redundancy that compensation is removed by entropy coding in redundancy that interframe is removed and the quantization transform coefficient in piece, the picture coding device of the redundancy that can reduce other kinds that exceed adjacent block, exist in the dct transform zone also is provided.

By the expression prior art image prediction encoding device Figure 14 as can be known, (for example, at MPEG) normally used image prediction encoding device has piece sampling unit 1001, dct transform unit 1004, quantizer 1005 and entropy coder 1006 in the image encoding that formerly has.

In frame interior coding (intraframe coding),, at first carry out the piece sampling processing for the picture signal of input.Directly carrying out dct transform then handles.Then, carrying out quantification treatment and entropy coding handles.On the other hand, in frame interior coding (predictive frame coding), after the piece sampling processing,, move the processing of detecting unit 1002 and moving compensating unit 1003, in addition, also carry out the dct transform processing for the view data of the frame of now of processing target.And then, carry out quantification treatment and entropy coding again and handle.

Here, in entropy coding unit 1006, after quantized value carries out entropy coding, outputting encoded data.So-called entropy coding is exactly by to the short code of recurrent value distribution, to the long code of not recurrent value distribution, is encoded near entropy as average information, thus the mode of cutting down overall size of code significantly.Reversible encoding that Here it is.As entropy coding, motion variety of way, still, in baseline system, use huffman coding.In DC coefficient value that has quantized and AC coefficient value, the method difference of huffman coding, that is, the DC coefficient is represented the mean value of 8 * 8 block of pixels, in general image, has similar value to the mean value majority of adjacent piece.Therefore, after the difference of getting with last, carry out entropy coding.Like this, since value set near 0, so entropy coding is effective.In addition,, for example carry out saw-tooth sweep, 2-D data is transformed to one-dimensional data about the AC coefficient.In addition, the AC coefficient that particularly comprises radio-frequency component much is 0, so will have the value of AC coefficient of the value beyond 0 and its front has several 0 (strokes) to be one group to carry out entropy coding.

The position that rate controller 1007 feedbacks were used in the piece of having encoded in the past, the processing of control quantifying unit 1005, and adjust code bit speed.Here, rate controller 1007 distributes different position amounts according to the character and the utilizable position control routine bit rate of the unit of having encoded in order to target data, each frame and each piece of having encoded that each has been encoded.In addition, re-quantization is handled and the inverse dct transform processing is carried out in the unit 1008 and 1009 as the part of local decoder.Store in the local decoding frame memory 1010 by the local decoder decoded image data, be used to move detect and handle.The 1011st, in order to move the reference frames memory that detects and preserve the former frame of front.And last, 1006 output bit streams transmit to the image prediction/decoding device that receives a side or the side of regenerating from the entropy coding unit.

Figure 15 is the ideograph of the image of the adaptive DCT domain transformation that is used to illustrate that intra-frame prediction is used.

In Figure 15, in the dct transform zone, show the situation that 4 each discrete cosine transform block of 8 * 8 constitute macro blocks.Here, B0 represents to have the current block of the now of 8 * 8 dct transform coefficient.B2 represents the decoded piece adjacent with upside.B1 and B3 represent 2 each piece that decoded oblique upper is adjacent.B4 represents the piece in the dead ahead that the left side is adjacent.As shown in Figure 15, the piece with dct transform coefficient can be predicted according to decoded adjacent a plurality of with dct transform coefficient of 8 * 8.

Should be noted that current block according to which piece is predicted always different.Therefore, determine that this decision can be adaptively provides to the different piece of image prediction/decoding device one side according to the minimum bit service regeulations.This decision utilizes indicating bit notice image prediction/decoding device.Here, the minimum bit service regeulations are used for determining Forecasting Methodology in different a plurality of Forecasting Methodologies that after using each Forecasting Methodology, counting is encoded piece employed and measured.As a result of, the method that causes employed minimum bit amount is chosen as the Forecasting Methodology that should use.

In addition, the dct transform regional prediction can carry out after quantification treatment He before the quantification treatment.

(embodiment 5)

Figure 16 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 5.The image prediction encoding device of Figure 16 is characterised in that the dct transform regional prediction is handled and carried out after quantification treatment.

In Figure 16, carry out the piece sampling by piece sampling unit 1012 earlier for the picture signal of input.And in the frame interior coding, the piece view data of sampling is not carried out the processing of adder 1013, and is input to dct transform unit 1014 by adder 1013.On the other hand, in the predictive frame coding, adder 1013 deducts the moving inspection image data of driven detection and compensating unit 1025 outputs from the piece view data of sampling, and the view data of subtraction result is exported to dct transform unit 1014.And, carry out the dct transform processing by unit 1014 after, carry out quantification treatment by unit 1015.

The dct transform regional prediction is handled and is undertaken by unit 1017, and the 1018th, be used to store in order to predict the block storage of decoded of front.Adder 1016 will deduct from the discrete cosine transform block of the now of quantifying unit 1015 outputs from the decoded adjacent block of dct transform regional prediction unit 1017 outputs.The decision of the adjacent piece that this has been encoded is carried out in dct transform regional prediction unit 1017.At last, for the discrete cosine transform block of having predicted, carry out entropy VLC encoding process by unit 1020, the position of having encoded writes in the bit stream.

The adjacent piece and prediction piece addition of adder 1019 by predicting employed front decoded to the discrete cosine transform block of now.Secondly, for decoded discrete cosine transform block, re-quantization is handled and the inverse dct transform processing is carried out in

unit

1021 and 1022 respectively.Local decoded from the inverse dct transform unit view data input summer 1023 of the pieces of 1022 outputs.Adder 1023 is passed through the view data of the frame of front and decoded view data addition, and the view data that obtains to construct again also is stored in the frame memory 1024.Moving detection and compensation deals are undertaken by unit 1025.Moving detect and the frame of the front that compensation deals are used use frame memory 1024 in order to store.

(embodiment 6)

Figure 17 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 6.The image prediction encoding device of Figure 17 is characterised in that, carried out the dct transform regional prediction and handle before quantification treatment.For the picture signal of input, in unit 1026, carry out the piece sampling processing.Secondly, adder 1027 is carried out subtraction for the predictive frame coding, and the view data of subtraction result is exported to entropy VLC coding unit 1034 and inverse quantization unit 1033 by dct transform unit 1028, adder 1029 and quantifying unit 1030.

The view data of the front piece that block storage 1032 storages are used for the dct transform regional prediction of unit 1031 handles.By adder 1029, will be from the dct transform unit view data of discrete cosine transform block of now of 1028 outputs from according to deducting the discrete cosine transform block of minimum bit service regeulations by 1031 selected fronts, dct transform regional prediction unit.After the view data of the discrete cosine transform block of subtraction result is carried out quantification treatment by quantifying unit 1030, to inverse quantization unit 1033 and 1034 outputs of entropy VLC coding unit.Inverse quantization unit 1033 is handled and is decoded by the view data of the discrete cosine transform block that has quantized of input being carried out re-quantization, and to adder 1055 outputs.Adder 1055 is with the view data addition of the discrete cosine transform block of the front of the view data of decoded discrete cosine transform block and dct transform regional prediction unit 1031, with the image data storage of the piece of the front of addition results in block storage 1032, simultaneously to 1036 outputs of inverse dct transform unit.

The 1036 pairs of view data from the piece of the front of adder 1035 inputs in inverse dct transform unit are carried out the inverse dct transform processing, and the decoded view data after the conversion process is exported to adder 1037.Adder 1037 is with the view data of the frame of the front of driven detection and compensating unit 1025 outputs and the decoded view data addition of 1036 outputs from the inverse dct transform unit, after temporarily storing into the view data of addition results in the frame memory 1038, detect and compensating unit 1025 is exported to moving.

B1. the pattern decision is bright in general

Figure 18 is the block diagram of structure of the dct transform regional prediction circuit 1017,1031 of expression Figure 16 and Figure 17.

In Figure 18, the 1040th, the block storage of the view data of the adjacent piece of the employed front of storage prediction.The current block of now of processing target is input to unit 1041, and unit 1041 from by deducting the adjacent discrete cosine transform block in the front of block storage 1040 storages, obtains the view data of following 4 kinds of prediction discrete cosine transform blocks with the view data of the current block of input.

(a) by the No-Pred piece shown in 1042,

(b) by the Up-Pred piece shown in 1043,

(c) by the Left-Pred piece shown in 1044,

(d) by the Other-Pred piece shown in 1045.

Here, with the above-mentioned 4 kinds of pieces of 2 bit representations.That is, for example " 00 " represents the No-Pred piece, " 01 " expression Up-Pred piece, " 10 " expression Left-Pred piece, " 11 " expression Other-Pred piece.

The No-Pred piece is the view data of the discrete cosine transform block of the now when not having prediction itself.The Up-Pred piece represents to predict the view data of the prediction piece that the piece of use obtains when being the adjacent discrete cosine transform block B2 in top.The Left-Pred piece represents to predict the view data of the prediction piece that the piece of use obtains when being the adjacent discrete cosine transform block B4 in left side.The Other-Pred piece represents to predict the view data of the prediction piece when only the DC coefficient being carried out.Under the situation of Other-Pred, there are 2 kinds of Forecasting Methodologies.That is the view data of the prediction piece that obtains when, Up-DC-Pred (1046) expression is only predicted the DC coefficient according to the adjacent discrete cosine transform block B2 in top.The view data of the prediction piece that Left-DC-Pred (1047) expression obtains when only the DC coefficient being predicted according to the adjacent discrete cosine transform block B4 in left side.For both of these case, also need 1 in order to indicate.For example, use " 0 " expression Up-DC-Pred (1046), use " 1 " expression Left-DC-Pred (1047).

Prediction based on incline direction adjacent piece B1, B3 also is possible, and still, this piece prediction of utilizing top or left side of predicting the outcome not good like that be not so use in the present embodiment.

For the piece of all having predicted, check by carry out actual entropy coding processing by unit 1048.In unit 1049, the position of using in the different pieces of having predicted is compared.At last, unit 1050 discrete cosine transform block that decision has been predicted according to the minimum bit service regeulations is exported the discrete cosine transform block of having predicted with indicating bit.That is, select the discrete cosine transform block of having predicted of figure place minimum.

B2. the enforcement of pattern decision

In Figure 19, use for reality, illustrate the part set of the image religion certificate that the defined DC/AC in front predicted.Current block 1101 is 8 * 8 of upper left-hand of current macro, and current block 1102 is 8 * 8 of upper right of current macro.A and B are 8 * 8 adjacent with current block 1101. the top row of being emphasized of current block 1101 and left column are that predict in the same place of A and B according to adjacent block respectively.That is, the lastrow of current block 1101 predicts that according to the lastrow of the piece A above it left column of current block 1101 is predicted according to the left column of the piece B on its left side.According to same order, current block 1102 is predicted according to the current block 1 on the piece D above it and its left side.

If (u v) is the piece that should encode to C, E _iPredicated error during for pattern i, suppose be with predicted value from A (u, v) and/or B (u deducts in respectively the determining v) and obtains.In the enforcement of reality, only use and save 3 maximum patterns of described frequency at B1.

(a) pattern 0: only predict DC

E ₀(0，0)＝C(0，0)-(A(0，0)+B(0，0))/2，

E ₀(u，v)＝C(u，v)，

u≠0；v≠0；u＝0，…，7；v＝0，…，7 …(1)

(b) pattern 1: according to last lateral mass prediction DC/AC

E ₁(0，)＝C(0，)-A(0，)，v＝0，…，7，

E ₁(u，v)＝C(u，v)，

u＝1，…，7；v＝0，…，7 …(2)

(c) pattern 2: according to left side block prediction DC/AC

E ₂(u，0)＝C(u，0)-B(u，0)，u=0，…，7，

E ₂(u，v)＝C(u，v)，

u＝0，…，7；v＝1，…，7. …(3)

The selection of pattern is by calculating the absolute value sum SAD of the error of having predicted to the piece of 4 luminance signals of macro block _ModeiAnd select to have the pattern of minimum value wherein and carry out.

{SAD}_{mode i}

= \underset{b}{Σ} [Ei (0,0) + 32 \cdot \underset{u}{Σ} Ei (u, 0) + 32 \cdot \underset{v}{Σ} Ei (0, v),

i＝0，…， 2；b＝0，…， 3；u， v＝1，…， 7

…(4)

The decision of pattern and application program different relevant that with different encoded bit rate is target can utilize piece baseline and macro block baseline to carry out.Pattern uses the variable length code of following table 1 to encode.

Table 1 DC/AC pattern is shown with VLC

Coefficient	The DC/AC predictive mode	The VLC code
Coefficient	The DC/AC predictive mode	The VLC code	0	0 (only DC)	0
1	1 (vertical DC and AC)	10	0	0 (only DC)	0
1	1 (vertical DC and AC)	10	2	2 (horizontal DC/AC)	11

When after quantification treatment, carrying out the DC/AC prediction, in the adjacent block of the adjacent block of the horizontal direction of going ahead of the rest or vertical direction, usually because the step-length difference of the quantification of using, so, in order correctly to carry out DC/AC prediction, several weight coefficients of the dct transform coefficient that need be used to converts has quantized.

If QacA is the dct transform coefficient that has quantized, the QacB of piece A (referring to Figure 19) is the dct transform coefficient that has quantized of piece B (referring to Figure 19).If QstepA is the employed quantization step of the quantification of piece A, then QstepB is exactly the employed quantization step of quantification of piece B, QstepC be exactly current block C quantification the quantization step that should use, therefore, the conversion equation is as follows:

Q’acA＝(QacA×QstepA)/QstepC …

(5)

Q’acB＝(QacB×QstepB)/QstepC …

(6)

Here, Q ' acA is the dct transform coefficient according to piece A, uses in the prediction of the lastrow of current block C.In addition, Q ' acB is the dct transform coefficient according to piece B, uses in the prediction of the left column of current block C.

(embodiment 7)

In Figure 20, the bit stream of image prediction encoding device is input to entropy VLD decoding unit 1051, carry out variable-length decoding.Decoded view data is restored the view data of discrete cosine transform block by the view data addition of adder 1052 by the discrete cosine transform block adjacent with the front of dct transform regional prediction unit 1053.The adjacent discrete cosine transform block in front knows it is which piece according to the indicating bit that takes out from bit stream, be used to predict in unit 1053.The 1054th, be used to store the block storage of the employed adjacent discrete cosine transform block of prediction.Export to inverse dct transform unit 1055 from the discrete cosine transform block that has restored that adder 1052 obtains.Inverse dct transform unit 1055 carries out inverse dct transform by the discrete cosine transform block to input and handles the view data that generates the dct transform coefficient that has restored, and to adder 1056 outputs.Adder 1056 generates moving detection and compensation and decoded view data and output by with the view data of having restored of inverse dct transform unit 1055 and the view data addition of the frame of the front of moving detection and compensating unit 1057.After this decoded view data temporarily stores the frame memory of storing the view data that is used for the moving previous frame that detects and compensate into, to moving detection and compensating unit 1057 outputs.The view data of moving detection and 1057 pairs of inputs of compensating unit is moved and is detected and compensation deals.

In addition, handle by the opposite recovery corresponding from the decoded view data of adder 1056 output, and restore original view data with the processing of the piece sampling unit 1012,1026 of Figure 16 and Figure 17.

In addition, the 1059th, inverse quantization unit, as shown in figure 17, when before quantification treatment, carrying out the processing of dct transform regional prediction, this inverse quantization unit 1059 is inserted into the position of the 1059a of Figure 20, on the other hand, and as shown in figure 16, when carrying out the processing of dct transform regional prediction after quantification treatment, this inverse quantization unit 1059 just is inserted into the position of the 1059b of Figure 20.

Figure 21 is the flow chart of coding/decoding method of DC/AC prediction of the image prediction/decoding device of expression Figure 20.That is, in Figure 21, illustrate and obtain the DC/AC predictive mode, reconstruct the details of decoding of the bit stream of dct transform coefficient according to adjacent DC/AC predicted value.

At first, in step 1059, indicating bit is decoded according to the bit stream of input, in step 1060, check the sign of indicating bit, if be masked as " 0 ", just step 1061 according on the mean value calculation Dc value of view data of lateral mass and left side block, and enter step 1063.If 1060 is "No", just enter step 1062, in step 1062,, just extract the view data of the left column of left side block out, and enter step 1065 in step 1063 if the Warning Mark of checking is " 10 ".If be "No" in step 1062, just enter step 1064, in step 1064, if the Warning Mark of checking is " 11 ", the view data of the lastrow of lateral mass in step 1065 extraction just, and enter step 1066.At last, in step 1066, will be in the corresponding dct transform coefficient addition of step 1061,1063 or 1065 dct transform coefficients of obtaining or extracting out with current block.

Below, the variation of present embodiment group is described.

(a) above-mentioned sampling unit 1012,1026 also can comprise the cross processing of alternately inserting pixel, so that the pixel of the two-dimensional arrangements in 4 piece groups is made of the odd pixel that is positioned at odd-numbered line in the 1st, in the 2nd, constitute by the even pixel that is positioned at odd-numbered line, in the 3rd, constitute, in the 4th, constitute by the even pixel that is positioned at even number line by the odd pixel that is positioned at even number line.

(b) above-mentioned prediction piece stores in the above-mentioned block storage, is the piece that has restored the front, selects from the piece that the current block that is arranged in Yu encoded is adjacent, can select conversion coefficients all in the piece.

(c) above-mentioned prediction piece stores in the above-mentioned block storage, is the piece that has restored the front, selects from the piece that the current block that is arranged in Yu encoded is adjacent, the subclass that is predetermined can be chosen as the conversion coefficient of piece.

(d) above-mentioned prediction piece stores in the above-mentioned block storage, it is the piece that has restored the front, select from the top piece adjacent of the current block that is arranged in and encoded with the left side, can only use the conversion coefficient of the first from left row of the lastrow of this piece and this piece, remaining conversion coefficient can be set at 0.

(e) above-mentioned prediction piece stores in the above-mentioned block storage, is the piece that has restored the front, selects near the piece that is arranged in the current block of having encoded, and the conversion coefficient of each piece can be weighted with different weighting functions.

(f) above-mentioned prediction piece stores in the above-mentioned block storage, is the piece that has restored the front, selects near the piece that is arranged in the current block of having encoded, can carry out transform operation to the conversion coefficient of each piece.

(g) above-mentioned prediction piece stores in the above-mentioned block storage, is the piece that has restored the front, can be in the hope of being positioned near a plurality of weighted average of the current block of having encoded.

(h) according to decoded view data, be made of 4 pieces that intersect a plurality of groups when forming the pixel of two-dimensional arrangements and restoring original view data, can carry out contrary cross processing to above-mentioned decoded view data, all ask for so that be positioned at the odd pixel of odd-numbered line according to the 1st, the even pixel that is positioned at odd-numbered line is asked for according to the 2nd, the odd pixel that is positioned at even number line is asked for according to the 3rd, and the even pixel that is positioned at even number line is asked for according to the 4th.

As mentioned above, if according to present embodiment group of the present invention, then very big effect is being arranged in order to remove or to reduce aspect the tediously long property in dct transform zone of adjacent interblock, the result can reduce the figure place of use, thereby finally can improve code efficiency greatly.As the example of detailed image predictive coding device, with reference to Figure 18, prediction processing is preferably only undertaken by the adjacent piece that uses upside or left side.

For the sequence that comprises QCIF, can save 6.4% position to upper bit rate coding, can save 20% position to the next bit rate coding.In addition, for example for other QCIF sequences such as test sequence such as Akiyo, Mother and Daughter can save about 10% the position.In addition, can save more position for CIF and CCIR sequence.

As mentioned above, according to the 2nd embodiment group of the present invention, can provide the new image prediction encoding device and the image prediction/decoding device of the code efficiency that increases now.In this device, improving code efficiency does not need complicated apparatus, and its circuit structure can form very simple and easyly.

The 3rd embodiment group

The 3rd embodiment group comprises embodiment 8.

The present inventor is in view of the problem of prior art, consider by not only cut down between 2 images or 1 image in piece inside tediously long property and also cut down the tediously long property of the interblock in the image, improve the tediously long property of image coding efficiency by the scan mode that suitably adopts piece simultaneously.

We find that even identical position, the dct transform coefficient in the adjacent piece is very similar as a rule.The character of the original image of 2 pieces when closely similar or level or vertical line, diagonal and other picture patterns when comprising identical content, we can say that above-mentioned situation is correct.According to the viewpoint of information theory, identical information just means tediously long property.

Exceed the tediously long property that in the dct transform zone, exists outside the piece can by before the adaptive prediction of piece remove or cut down.Here it is, and the VLC entropy coding can obtain the result of the higher code efficiency of the entropy littler than predictive error signal.

Simultaneously, level and vertical structure become the result in the transform block that important dct transform coefficient concentrates on the row of the leftmost side and lastrow, are well-known.Therefore, embodiments of the invention by making the scanning self adaptationization, just can solve the problems referred to above in the scanning of coefficient according to predictive mode.

Thereby embodiments of the invention provide the method for removing the redundancy of interblock according to the dct transform coefficient of other block adaptive ground prediction current block.The information of predicated error, in the predictive mode of the further entropy of lower quantization dct transform coefficient by making the scan method self adaptationization and then dwindle.As a result, can cut down be used for to dct transform coefficient encode the position number.

In order to address this problem, can obtain carrying out the method for the decision of predictive mode according to the bit rate of the reality of utilizing each prediction and scan method to take place.

Embodiments of the invention provide the method for the dct transform coefficient of the present current block of prediction.Dct transform has the view data to identical piece to give the tendency of the distribution of identical value or identical dct transform coefficient, so current block is keeping the good correlation of the dct transform coefficient in the piece adjacent with other usually.

The view data of input is some in the frame of frame interior or temporary transient prediction, at first, the view data of input is carried out dct transform processing based on common piece.After obtaining the dct transform coefficient of current block, can or carry out the prediction in dct transform zone before quantification after quantizing.

Dct transform coefficient in the current block, the former adjacent piece that can be positioned at diagonal (incline direction) according to left side is up predicted.As shown in figure 23, they were decoded in this moment.The error signal that the

DCT coefficient

1 or 1 or more of the piece of having predicted by piece adjacent before deducting from the DCT coefficient of the same position of current block takes place to have predicted.

If predicted before quantification treatment, the predictive error signal of then different predictive modes just carries out quantification treatment.The predictive error signal that has quantized (a series of) view data to sequence before carrying out entropy coding scans.According to the minimum bit service regeulations, the piece of selecting to have predicted, promptly has a prediction piece of minimum position.The coded data of this piece transmits to image prediction/decoding device with employed predictive mode.

Image prediction/decoding device utilizes the coded data of employed predictive mode and piece that the piece of having predicted is decoded.After the coded data of piece being carried out opposite entropy decoding, the predicated error that has quantized is carried out reverse scan according to employed scan mode.If after prediction processing, carry out quantification treatment, just piece is carried out re-quantization and handle.The piece of constructing again can be by being obtained by dct transform coefficient in the former decoded adjacent piece of predictive mode indication and present dct transform coefficient addition.On the other hand, if before prediction processing, carry out quantification treatment, just the coefficient of constructing is again carried out re-quantization and handle.At last, inverse dct transform is handled the dct transform coefficient that each piece is applied to construct again, can obtain decoded image.

Embodiments of the invention provide to reduce and exceed outside the adjacent piece and the image prediction encoding device and the image prediction/decoding device of the redundancy that exists in the dct transform zone.

(embodiment 8)

Figure 24 is the block diagram of structure of the image prediction encoding device of the expression embodiment of the invention 8.The image prediction encoding device of Figure 24 is compared with the image prediction encoding device of the prior art of Figure 22, it is characterized in that having:

(a) adder 2035,

(b) H/V/Z scanning element 2036,

(c) adder 2038,

(d) block storage 2039 and

(e) has the dct transform regional prediction unit 2040 of quantization conversion.

In frame interior coding (intraframe coding),, after carrying out the piece sampling processing, directly carry out dct transform and handle by unit 2033 by unit 2031 for the picture signal of input.Secondly, the dct transform coefficient by 2034 pairs of 2033 outputs from the dct transform unit in unit carries out quantification treatment.On the other hand, in frame interior coding or interframe encode (predictive frame coding), after the piece sampling processing of unit 2031, adder 2032 deducts the view data of driven detection and compensating unit 2045 outputs from the view data after the piece sampling processing, obtain prediction error data.Then, export to adder 2035 by dct transform unit of handling 2033 and the quantifying unit 2034 of carrying out quantification treatment by carrying out dct transform for this prediction error data.Dct transform coefficient is handled by the dct transform zone of unit 2040 and is predicted, with the dct transform coefficient input summer of having predicted 2035.Adder 2035 deducts the dct transform coefficient of having predicted of dct transform regional prediction unit 2040 from the dct transform coefficient of quantifying unit 2034, and with the dct transform coefficient of the predicated error of subtraction result to H/V/Z scanning element 2036 and adder 2038 outputs.H/V/Z scanning element 2036 is carried out horizontal sweep, vertical scanning or saw-tooth sweep according to selected predictive mode adaptively to the dct transform coefficient of importing, and the dct transform coefficient after the scan process is exported to entropy VLC coding unit 2037.Secondly, the dct transform coefficient of 2037 pairs of inputs of entropy VLC coding unit carries out entropy VLC encoding process, and the bit stream that will encode transmits to the image prediction/decoding device that receives a side or the side of regenerating.

Adder 2038 is by will be from the dct transform coefficient that has quantized of adder 2035 output and the dct transform coefficient addition of having predicted of dct transform regional prediction unit 2040, and the quantification dct transform coefficient data of having been restored.This quantizes the dct transform coefficient data to block storage 2039 and inverse quantization unit 2041 outputs.

In the local decoder in being arranged on this image prediction encoding device, the dct transform coefficient data of having restored of adder 2038 temporarily store in the block storage 2039 of data of 1 piece of storage in order to carry out next prediction after, to 2040 outputs of dct transform regional prediction unit.The quantification dct transform coefficient that inverse quantization unit 2041 will be imported carries out re-quantization to be handled, and to 2042 outputs of inverse dct transform unit, then, the dct transform coefficient that has restored of the 2042 pairs of inputs in inverse dct transform unit carries out inverse dct transform to be handled, and the view data of the piece of now is restored the back to adder 2043 outputs.

In interframe encode, in order to generate local decoded image data, adder 2043 will have been carried out the view data of moving detection and compensation and the view data addition of having restored of inverse dct transform unit 2042 by moving detection and compensating unit 2045, obtain local decoded image data, and store in the frame memory 2044 of local decoder.The

unit

2009,2010,2011 of the structure of adder 2043, frame memory 2044, moving detection and compensating unit 2045 and the prior art of processing and Figure 22 is identical.

At last, bit stream is 2037 outputs from the entropy coding unit, transmit to image prediction encoding device.

Figure 25 is the block diagram of structure of the image prediction/decoding device of the expression embodiment of the invention 8.The image prediction/decoding device of Figure 25 is compared with the image prediction/decoding device of the prior art of Figure 23, it is characterized in that having:

(a) H/V/Z scanning element 2052,

(b) adder 2053,

(c) dct transform regional prediction unit 2055 and

(d) block storage 2054.

In Figure 25, the bit stream of image prediction encoding device is decoded in variable length decoder unit 2051.Decoded data are input to H/V/Z inverse scan unit 2052,, scan by the contrary direction of level, vertical contrary direction or the contrary direction of zigzag respectively according to scan pattern.Data after the scan process are input to adder 2053, adder 2053 is by the data after inverse scan is handled and the prediction error data addition of dct transform predicting unit 2055, obtain decoded dct transform coefficient data, and with it to inverse quantization unit 2056 output, simultaneously, be stored in the block storage 2054.Secondly, the dct transform coefficient data of having encoded that inverse quantization unit 2056 will be imported are carried out the re-quantization processing, obtain the dct transform coefficient of re-quantization, and to 2057 outputs of inverse dct transform unit.The dct transform coefficient data of the 2057 pairs of inputs in inverse dct transform unit are carried out the inverse dct transform processing, restore original view data, and to adder 2058 outputs.In interframe encode, adder 2058 obtains the view data of inverse dct transform unit 2057 and the prediction error data addition of moving detection and compensating unit 2060 decoded partly view data, and exports to external device (ED), simultaneously, store in the frame memory 2059.

In addition, handle by the reverse recovery corresponding, restore and be original view data with the processing of the piece sampling unit 2031 of Figure 24 from the decoded view data of adder 2058 output.

In the above-described embodiments, quantification treatment was carried out before prediction processing.In variation, also can after prediction processing, carry out quantification treatment.At this moment, in local decoder and image prediction/decoding device, before with the predicted value addition, carry out the re-quantization processing.Other details are all identical with the foregoing description.

Figure 26 is illustrated among the embodiment 8 figure of cutting apart the structure of the macro block of the frame that obtains and piece according to piece, and is the ideograph of the image of expression block predication.The enlarged drawing of Figure 26 represents how the prediction data of current block is encoded.Here, (u, v) (u, v) (u v) obtains piece A that can be adjacent according to upside piece C to adjacent piece B with left.Below, illustrate in greater detail this embodiment of the present invention.

C1. predict the sequence number of employed coefficient

The sequence number of predicting employed coefficient is relevant with the sequence of view data.Sign A C_Coeff is used for selecting adaptively the number of the best of the employed coefficient of each image.Sign is listed in the following table 2, transmits to image prediction/decoding device from image prediction encoding device as the part of side information.Fixed length code and the FLC corresponding with sign A C_Coeff are shown in table 2.Here, FLC (Fixed Length Coding; The fixed length coding) be the reversible encoding that distributes the code word of fixed length in order to represent all possible incidents.

Fixed length code, FLC that table 2. is corresponding with sign A C_Coeff

Coefficient	AC-Coeff (prediction is used)	FLC
Coefficient	AC-Coeff (prediction is used)	FLC	0	DC only	000
1	DC+AC1	001	0	DC only	000
1	DC+AC1	001	2	DC+AC1+AC2	010
3	DC+AC1+AC2+AC3	011	2	DC+AC1+AC2	010
3	DC+AC1+AC2+AC3	011	4	DC+AC1+AC2+AC3+AC4	100
5	DC+AC1+AC2+AC3+AC4+AC5	101	4	DC+AC1+AC2+AC3+AC4	100
5	DC+AC1+AC2+AC3+AC4+AC5	101	6	DC+AC1+AC2+AC3+AC4+AC5+AC6	110
7	DC+AC1+AC2+AC3+AC4+AC5+AC6+AC7	111	6	DC+AC1+AC2+AC3+AC4+AC5+AC6	110

Here, ACn is relevant with employed code, be A (0, n) or B (n, 0).

Under the special situation of present embodiment, all AC coefficients of lastrow and the first from left row are used to predict.In the case, when the acquiescence default value, image prediction encoding device and image prediction/decoding device do not need sign.

C2. the conversion of quantization step

When adjacent piece used the different quantization step size of current block, the prediction of AC coefficient was so ineffective.Therefore, this Forecasting Methodology is deformed into prediction data and converts according to the ratio of the quantization step of the piece of the ratio of the quantization step size of present current block and prediction data.This definition uses the equation of following joint C3. to provide.

C3. predictive mode

A plurality of patterns of setting are as follows:

(a) pattern 0: the DC prediction (abbreviating " going up the DC pattern " as) of going up lateral mass according to processing block

E ₀(0，0)＝C(0，0)-A(0，0)，

E ₀(u，v)＝C(u，v) …(7)

(b) pattern 1: carry out the DC prediction (abbreviating " left DC pattern " as) of the piece in left side according to processing block

E ₁(0，0)＝C(0，0)-B(0，0)，

E ₁(u，v)＝C(u，v) …(8)

(c) pattern 2: carry out the DC/AC prediction (abbreviating " going up the DC/AC pattern " as) of the piece of upside according to processing block

E ₂(0，0)＝C(0，0)-A(0，0)，

E ₂(0，v)＝C(0，v)-A(0，v)·Q _A/Q _c，

v＝1，2，…，AC_Coeff，

E ₂(u，v)＝C(u，v) …(9)

(d) mode 3: carry out the DC/AC prediction (abbreviating " left DC/AC pattern " as) of the piece in left side according to processing block

E ₃(0，0)＝C(0，0)-B(0，0)，

E ₃(u，0)＝C(u，0)-B(u，0)·Q _B/Q _C

u＝1，2，…，AC_Coeff，

E ₃(u，v)＝C(u，v) …(10)

C4. adaptive horizontal/vertical/saw-tooth sweep

If given above 4 predictive modes, the efficient of intraframe coding just can further be improved by the scanning of adopting coefficient.

Figure 27, Figure 28 and Figure 29 are respectively the ideographs of image of order that is used to illustrate the employed horizontal sweep of coefficient scanning, vertical scanning and the saw-tooth sweep of embodiment 8.Here, these scannings can scan and reference as H/V/Z in the concentrated area.

C5. significantly pattern decision

In significantly pattern determined, the decision of predictive mode was carried out in image prediction encoding device, and this decision information is used several position information of having encoded of bit stream, transmits significantly to image prediction/decoding device from image prediction encoding device.

Figure 30 is the flow chart that the pattern decision of expression embodiment 8 uses is handled.

In Figure 30, the dct transform coefficient of current block is input to unit 2062, the dct transform coefficient data of the current block of unit 2062 by will input deduct from the dct transform coefficient data of the adjacent block of block storage 2061 carries out the dct transform prediction processing.In unit 2062, carry out the dct transform prediction processing according to 4 patterns that in joint C3., illustrate.Secondly, in H/V/Z scanning element 2063, carry out the scan process of coefficient,, as shown in figure 30, carry out each the corresponding scan process that in joint C4., illustrates here.In addition, the dct transform coefficient data after the scan process transmit to entropy coding unit 2064, carry out Variable Length Code here and handle.Then, in unit 2065, will compare by all positions that different patterns takes place, in unit 2066, the piece of the dct transform coefficient of the predictive mode of the position that the selection generation is minimum.2066 transmit to image prediction/decoding device as bit stream from the unit with the value of predictive mode the position of these dct transform coefficient data.Predictive mode uses and encodes as the fixed length code of following table 3.

The FLC table that table 3. DC/AC/ scan pattern is used

Coefficient	The DC/AC pattern	Scan pattern	The FLC code
Coefficient	The DC/AC pattern	Scan pattern	The FLC code	0	Last DC	Saw-tooth sweep	00
1	Left side DC	Saw-tooth sweep	01	0	Last DC	Saw-tooth sweep	00
1	Left side DC	Saw-tooth sweep	01	2	Go up (DC+AC)	Horizontal sweep	10
3	A left side (DC+AC)	Vertical scanning	11	2	Go up (DC+AC)	Horizontal sweep	10

C6. the decision of default mode

In the embodiment 2 of pattern decision, image prediction encoding device has identical predictive mode decision function jointly with image prediction/decoding device.Image prediction encoding device and the decision relevant directivity of the image prediction/decoding device decoded DC coefficient value decision that all basis is adjacent with current block with predictive mode.That is, in the decision of default mode, several rules are used in the decision of default mode, carry out in image prediction encoding device and image prediction/decoding device.And the additional information data of expression pattern decision does not transmit to image prediction/decoding device from image prediction encoding device.

Figure 31 is the ideograph of the image of the relation of piece during the default mode of expression embodiment 8 determines.That is, Figure 31 represents that each piece and forecasting object are the position relation of current block.

In Figure 31, piece C is the current block of the process object in the now forecast.Piece A is the piece of the upside of the current block C in the prediction.Piece B is the piece that is positioned at current block C left side.Piece C ' be and the piece A of current block C diagonal position and the piece between the piece B.

At first, the direction of decision DC.Use other determining method, determine whether in prediction the same with the AC coefficient.In order to determine, the summation of the difference of the absolute value of predictive coefficient and the absolute value of nonanticipating coefficient are compared, judge which is little.This indication to image prediction/decoding device is used 1.Predicted whether in the decision of the directivity of DC prediction and AC coefficient, used following formula.Table 3 is conclusions of possible 4 conclusions.

(A1) if

(B(0，0)-C’(0，0)＜C’(0，0)-A(0，0))

(11) time, then

E(0，0)＝C(0，0)-A(0，0) …(12)

(a1) if

(Σ_{v = 1}^{7} C (0, v) &GreaterEqual; Σ_{v = 1}^{7} C (0, v) - A (0, v)) - - - (13)

The time, then

E(0，v)＝C(0，v)-A(0，v)·Q _A/Q _C，v＝1，…，7，

…(14)

(a2) if above-mentioned (13) formula is false, then

E(0，v)＝C(0，v) …(15)

(A2) if above-mentioned (11) formula is false, then

E(0，0)＝C(0，0)-B(0，0) …(16)

(b1) if

(Σ_{v = 1}^{7} C (u, 0) &GreaterEqual; Σ_{v = 1}^{7} C (u, 0) - B (u, 0)) - - - (17)

The time, then

E(u，0)＝C(u，0)-B(u，0)·Q _B/Q _C，v＝1，…，7，

…(18)

(b2) if above-mentioned (17) formula is false, then

E(u，0)＝C(u，0) …(19)

In addition, for other all coefficients, then

E(u，v)＝C(u，v) …(20)

The FLC table that the DC/AC scan pattern of table 4. acquiescence decision is used

FLC	The acquiescence decision	The DC/AC scan pattern
FLC	The acquiescence decision	The DC/AC scan pattern	00	\|B(0，0)-C’(0，0)\|＜\|C’(0，0)-A(0，0)\|	Last DC saw-tooth sweep
01	\|B(0，0)-C’(0，0)\|≥\|C’(0，0)-A(0，0)\|	Left side DC saw-tooth sweep	00	\|B(0，0)-C’(0，0)\|＜\|C’(0，0)-A(0，0)\|	Last DC saw-tooth sweep
01	\|B(0，0)-C’(0，0)\|≥\|C’(0，0)-A(0，0)\|	Left side DC saw-tooth sweep	10	\|B(0，0)-C’(0，0)\|＜\|C’(0，0)-A(0，0)\|	Last horizontal sweep (DC+AC)
11	\|B(0，0)-C’(0，0)\|≥\|C’(0，0)-A(0，0)\|	Left side vertical scanning (DC+AC)	10	\|B(0，0)-C’(0，0)\|＜\|C’(0，0)-A(0，0)\|	Last horizontal sweep (DC+AC)

In the foregoing description 8, the dct transform coefficient prediction processing is undertaken by 2040 pairs of unit quantized transform coefficients data, still, the invention is not restricted to this, and is the same with the embodiment 6 of Figure 17, also can carry out non-quantized transform coefficient data.At this moment, in the image prediction/decoding device of correspondence, in Figure 25, with inverse quantization unit 2056, mobile being inserted between inverse scan unit 2052 and the adder 2053.

Below, the variation of embodiment 8 is described.

(a) piece sampling unit 2031 also can comprise the cross processing of alternately inserting pixel, so that the pixel of the two-dimensional arrangements in 4 piece groups is made of the odd pixel that is positioned at odd-numbered line in the 1st, in the 2nd, constitute by the even pixel that is positioned at odd-numbered line, in the 3rd, constitute, in the 4th, constitute by the even pixel that is positioned at even number line by the odd pixel that is positioned at even number line.

(b) above-mentioned prediction piece stores in the above-mentioned block storage, is the piece that has restored in the past, selects from the piece that the current block that is arranged in Yu encoded is adjacent, also can select all coefficient datas in the piece.

(c) above-mentioned prediction piece stores in the above-mentioned block storage, and the piece that has restored before being is selected from the piece that the current block that is arranged in Yu encoded is adjacent, also can select the subclass that the is predetermined coefficient data as piece.

(d) above-mentioned prediction piece stores in the above-mentioned block storage, the piece that has restored before being, select from the top piece adjacent of the current block that is arranged in and encoded, can only use the coefficient data of the first from left row of the lastrow of this piece and this piece, remaining coefficient data is set at 0 with the left side.

(e) above-mentioned prediction piece stores in the above-mentioned block storage, from before select according to said reference the piece that restored, also can communicate the subclass of the coefficient data more than 1 or 1 of the lastrow that decides use only to comprise this piece or the first from left row by image prediction encoding device and image prediction/decoding device.

(f) above-mentioned prediction piece stores in the above-mentioned block storage, select according to said reference according to the piece that has restored in the past, also can use the subclass of the coefficient data more than 1 or 1 of the lastrow that only comprises this piece or the first from left row by image prediction encoding device decision, and the sign that will represent the number of the subclass that determined and coefficient data periodically inserts in the data that image prediction/decoding device sends the notice image prediction/decoding device.

(g) above-mentioned prediction piece stores in the above-mentioned block storage, select according to said reference according to the piece that has restored in the past, the coefficient data of each piece also can be according to carrying out multiplying with the quantization step size of the current block of having encoded with the ratio that the ratio of the quantization step size of prediction piece equates.

(h) above-mentioned prediction piece stores in the above-mentioned block storage, selects according to said reference according to the piece that has restored in the past, and the coefficient data of each piece also can be weighted with different weighting functions.

(i) above-mentioned prediction piece stores in the above-mentioned block storage, selects according to said reference according to the piece that has restored in the past, also can carry out the transform operation of appointment to the coefficient data of each piece.

(j) above-mentioned prediction piece is adjacent with the current block of having encoded, stores in the above-mentioned block storage, the weighted average of the piece that has restored before also can be used as and obtaining.

(k) scan method can comprise

(I) each row of coefficient data from left to right, from lastrow begin to next line finish and the horizontal sweep of scanning,

(II) each row of coefficient data from lastrow to next line, from the first from left row begin to the rightest row finish and the vertical scanning of scanning,

(III) coefficient data from the most left coefficient data of lastrow to the rightest coefficient data of next line, along at least a kind of scan method the saw-tooth sweep of diagonal scanning.

(1) above-mentioned prediction piece stores in the above-mentioned block storage, selects according to said reference according to the piece that has restored in the past, and the predictive mode of above-mentioned prediction piece comprises

(i) only expression is positioned at processing target current block upside this piece piece, that be called the DC coefficient mean value go up most and the 1st pattern that the most left coefficient data is used to predict,

(ii) only will be positioned at the 2nd pattern that the D C coefficient of piece in left side of the current block of processing target is used to predict,

(iii) will be positioned at processing target current block upside piece lastrow the DC coefficient and comprise the 3rd pattern that the AC coefficient more than 0 or 0 of radio-frequency component is used to predict,

(iv) will be arranged in processing target current block the left side piece the first from left row the DC coefficient and comprise at least 1 predictive mode of the 4th pattern that the AC coefficient more than 0 or 0 of radio-frequency component is used to predict, the coefficient data of above-mentioned predicated error also can scan with the scan method of saw-tooth sweep.

(m) above-mentioned prediction piece stores in the above-mentioned block storage, select according to said reference according to the piece that has restored in the past, the coefficient data of above-mentioned predicated error scans according to a kind in the above-mentioned scan method, predicts that the Forecasting Methodology of the coefficient data of above-mentioned predicated error can comprise

(i) the DC coefficient of piece of upside that only will be positioned at the current block of processing target be used to predict, to the coefficient data of above-mentioned predicated error with saw-tooth sweep carry out scan process the 1st pattern,

The DC coefficient of piece in left side that (ii) only will be positioned at the current block of processing target is used to predict, to the coefficient data of above-mentioned predicated error with saw-tooth sweep carry out scan process the 2nd pattern,

(iii) will be positioned at the DC coefficient of lastrow of piece of upside of current block of processing target and the AC coefficient more than 0 or 0 that comprises radio-frequency component be used to predict, to the coefficient data of above-mentioned predicated error with horizontal sweep carry out scan process the 3rd pattern,

(iv) will be arranged at least a kind of the 4th pattern that the DC coefficient of the first from left row of piece in left side of current block of processing target and the AC coefficient more than 0 or 0 that comprises radio-frequency component are used to predict, the coefficient data of above-mentioned predicated error is carried out with vertical scanning scan process.

(n) according to above-mentioned decoded view data, thereby restore when being original view data from the pixel that a plurality of groups of being made of 4 pieces that intersect form two-dimensional arrangements, for the odd pixel that is positioned at odd-numbered line is all asked for according to the 4th according to the 3rd even pixel of asking for, be positioned at even number line according to the 2nd odd pixel of asking for, be positioned at even number line according to the 1st even pixel of asking for, be positioned at odd-numbered line, also can carry out contrary cross processing to above-mentioned decoded view data.

(o) image prediction encoding device and image prediction/decoding device also can use the same rule that is predetermined to decide above-mentioned predictive mode.

(p) image prediction encoding device and image prediction/decoding device also can use the same rule that is predetermined to decide above-mentioned scan method.

As mentioned above,, cut down or remove and exceed outside the adjacent piece and the redundancy that exists in the dct transform zone is very effective, can cut down the figure place of using the position, thereby can improve the efficient of coding significantly according to the 3rd embodiment group of the present invention.As the instrument of new video compression algorithm language, this also is useful.

In above embodiment, image prediction encoding device and image prediction/decoding device have been described, but, the invention is not restricted to this, both can be to comprise the image prediction/encoding method that structural elements such as the each several part of above-mentioned image prediction encoding device, each unit is replaced into the step of each step respectively, also can be to comprise the image prediction decoding method that structural elements such as the each several part of above-mentioned image prediction/decoding device, each unit is replaced into the step of each step respectively.At this moment, for example, in storage device, microprocessor unit (MPU), central arithmetic processing apparatus controllers such as (CPU) carry out image forecasting encoding processing and/or the processing of above-mentioned image prediction decoding by carrying out this degree to each step of above-mentioned image prediction/encoding method and/or above-mentioned image prediction decoding method as procedure stores.

In addition, the present invention also can be the program recording medium that record comprises each step of above-mentioned image prediction/encoding method and/or above-mentioned image prediction decoding method.This recording medium has posting field for example and is divided into the disc-shape that fan-shaped or posting field vortex shape is divided into each piece, for example is CD or magnetic recording disks such as photomagneto disk or floppy disk such as CD-ROM, DVD.

The possibility of utilizing on the industry

As mentioned above, according to image prediction encoding device of the present invention, have:

The 1st generation unit, when in the view data of mutual adjacent a plurality of zonules of cutting apart by above-mentioned cutting unit the view data of the zonule of processing target being encoded, will be with the view data of the zonule of above-mentioned processing target the view data of the adjacent zonule that has rebuild as the view data of the intra-frame prediction zonule of the zonule of above-mentioned processing target, the view data of above-mentioned intra-frame prediction zonule is generated the difference part of the view data of the view data of above-mentioned processing target zonule and above-mentioned optimum prediction zonule as the view data of optimum prediction zonule, it is the view data of difference zonule;

Fig. 2 generation unit by will be by the view data of the difference zonule of above-mentioned decoding unit decodes and the view data addition of above-mentioned optimum prediction zonule, generates the view data of having carried out the zonule that rebuilds.

Therefore, only will be with the view data of the zonule of the process object adjacent pixel value of having rebuild is as the pixel value of intra-frame prediction sequence number, compare with prior art, just can generate high-precision prediction signal simply, thereby can obtain to cut down the peculiar effect of the figure place of intraframe coding with low operand.

In addition, according to image prediction encoding device of the present invention, have:

Block storage is stored decoded coefficient data;

The 1st adder unit, the coefficient data by the prediction piece that will be selected by above-mentioned decision unit deducts the coefficient data of the predicated error of output subtraction result from the coefficient data of the current block of now;

Coding unit the coefficient data of the predicated error of above-mentioned quantifying unit is carried out the entropy sign indicating number, and the coefficient data of the predicated error that will encode sends to image prediction/decoding device;

The 2nd adder unit by exporting decoded coefficient data from the coefficient data of the prediction piece of above-mentioned decision unit output and coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, is stored in the above-mentioned block storage simultaneously; With

Therefore, the new image prediction encoding device and the image prediction/decoding device of the code efficiency increase that makes now can be provided.In this device, do not need to be used to improve the complicated apparatus of code efficiency, this circuit structure can form very simple and easyly.

Block storage is stored decoded coefficient data;

The 2nd adder unit by exporting decoded coefficient data from the coefficient data of the prediction piece of above-mentioned decision unit output and coefficient data addition from the predicated error of above-mentioned inverse quantization unit output, simultaneously, is stored in the above-mentioned block storage; With

Inverse transformation block by the coefficient data from the piece of above-mentioned the 2nd adder unit output is carried out inverse transformation, generates decoded view data.

Therefore, cut down or remove and exceed outside the adjacent piece and the redundancy that exists in domain transformation is very effective, can cut down the figure place of using the position, thereby can improve the efficient of coding significantly, as the instrument of new video compression algorithm language, this also is useful.

Claims

1. the image prediction decoding equipment of the incoming bit stream of the conversion coefficient that comprises variable length code of being used to decode comprises:

Variable-length decoder is used for the conversion coefficient of described variable length code is decoded as a series of conversion coefficients;

Predicting unit is used for DC coefficient according to the quantification of the current piece of the DC coefficient prediction of a tightly adjacent piece of selecting adaptively from last lateral mass or left side block; And

Inverse quantizer is used for the conversion coefficient re-quantization is become the numerical value of re-quantization;

The DC coefficient of the piece of having selected of DC coefficient that wherein is used to predict the quantification of current block is to calibrate by the quantization step that uses described current piece.

2. the image prediction decoding equipment of the incoming bit stream of the conversion coefficient that comprises variable length code of being used to decode comprises:

Predicting unit is used for AC coefficient according to the quantification of the current piece of the conversion coefficient prediction of a tightly adjacent piece of selecting adaptively from last lateral mass or left side block; And

The conversion coefficient of the piece of having selected of AC coefficient that wherein is used to predict the quantification of current block is to calibrate by the ratio of the quantization step of quantization step that uses described current piece and the described piece of having selected.