CN1852442A - Layering motion estimation method and super farge scale integrated circuit - Google Patents
Layering motion estimation method and super farge scale integrated circuit Download PDFInfo
- Publication number
- CN1852442A CN1852442A CN 200510093217 CN200510093217A CN1852442A CN 1852442 A CN1852442 A CN 1852442A CN 200510093217 CN200510093217 CN 200510093217 CN 200510093217 A CN200510093217 A CN 200510093217A CN 1852442 A CN1852442 A CN 1852442A
- Authority
- CN
- China
- Prior art keywords
- image
- predicted macroblock
- motion estimation
- image block
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
Diversiform sampling rates are adopted to carry out prediction for motion estimation of macro block: under low sampling rate, searching in larger range; in original image, searching in smaller range so as to ensure accuracy rate of search as well as executing speed. Very large scale integrated circuit of implementing the method is composed processing array including N*N pieces of processing units, control circuit, storage area, and comparison arbiter. Advantages are: simple structure, low power consumption, small size, search speed ahead of complete search algorithm. Meanwhile, without increasing amount of calculation and size of processing array, the Very large scale integrated circuit supports advanced prediction mode in MPEG-4 standard.
Description
Technical field
The present invention relates to method for estimating and very lagre scale integrated circuit (VLSIC) among a kind of MPEG-4, specifically, relate to a kind of layering motion estimation method and very lagre scale integrated circuit (VLSIC) among the MPEG-4.
Background technology
Current era, information technology and computer internet develop rapidly, multimedia messages has become the main carrier that the mankind obtain information, the multimedia information data amount is very big, information storing device and communication network have all been proposed very high request, hinder the great bottleneck that people effectively obtained and used information thereby become, therefore study multi-medium data compaction coding method efficiently, significant with compressed format storage and transmitting digitized multimedia messages.International Standards Organization has set up moving picture experts group tissue (MPEG) in 1998, through expert group's unremitting effort, formulated several video compression standards, and wherein the MPEG-4 standard has become international now research focus.
MPEG-4 only handles discrepant element between picture frame and the frame and gives up identical part, has significantly reduced data volume.Estimation is to remove the effective ways of time redundancy, and so-called time redundancy is meant the existence that a large amount of identical elements are arranged between picture frame and the frame.Provide block-based estimation and compensation technique to handle time redundancy between each frame effectively among the MPEG-4.MPEG-4 thinks can both find corresponding blocks by each image block of present frame in the next frame image, on this basis, the corresponding blocks of next frame image just can add that a predicted motion vector rebuilds by the image block of former frame.Estimation among the MPEG-4 is exactly to find out the predicted motion vector of each image block in the present frame.The size of above-mentioned image block is 16 * 16,8 * 8 etc., and 16 * 16,8 * 8 are meant, image block all is 16 or 8 pixels at the width of x direction and y direction, and the entire image piece is the square-shaped image zone that is made of 256 or 64 picture elements.In MPEG, handled image block is 16 * 16, the image block that will calculate predicted motion vector in the present frame is called as predicted macroblock, the image block that carries out matching operation with current image block in the next frame image is called as the searching image piece, and the entire image zone that current image block will be done to search in the next frame image is called search window.
The research of motion estimation algorithm is more deep, has proposed many effective methods, such as full-search algorithm, three-step approach, rhombus algorithm or the like.Simultaneously, along with the development in hand-held video terminal market, more and more urgent to the requirement of low-power consumption, undersized process chip.
In full-search algorithm, need each and the image block of the identical size of predicted macroblock in the search window of the predicted macroblock in the predicted macroblock in the present frame and the next frame be mated, though thereby the full-search algorithm performance is good, algorithm is regular, being convenient to hardware realizes, but amount of calculation is too big, the computing array that needs is also huger, power consumption is higher, be not suitable for the demand of existing market, though other algorithm amounts of calculation are few, algorithm is irregular, the concrete outcome that predicted macroblock and searching image piece had mated before each step of algorithm depended on is not suitable for hardware and realizes.
Summary of the invention
Technical problem to be solved by this invention is: design a kind of layering motion estimation method and very lagre scale integrated circuit (VLSIC), being used for digital video coding uses, the size and the power consumption of the video coding chip of this method of employing and very lagre scale integrated circuit (VLSIC) are greatly reduced, and support the substandard advanced prediction mode of MPEG-4.The advanced prediction mode of MPEG-4 is meant in estimation, and the motion estimation result of 16 * 16 predicted macroblock not only is provided, and the motion estimation result of 48 * 8 subimage blocks in the predicted macroblock also will be provided.
The object of the present invention is achieved like this: a kind of layering motion estimation method, and this method may further comprise the steps:
A) original image is carried out at least low sampling rate sampling, when each low sampling rate is sampled, call the match search of N * N image block repeatedly and realize the match search of predicted macroblock, obtain the motion vectors of predicted macroblock under low sampling rate, the width of described N presentation video piece or height;
B) directly handle original image, with A) in the motion vectors of predicted macroblock under low sampling rate obtained be that initial position is made match search, obtain the final motion vectors of predicted macroblock.
The value of described N gets 8,4 or 2.
Described steps A) comprising:
A1) original image is carried out 4: 1 sampling, in the scope of predicted macroblock corresponding position [4,3], make match search, obtain motion vectors V1;
A2) original image is carried out 2: 1 sampling, with A1) in the V1 that obtains be that initial position is made the match search in [2,1] scope, obtain the motion vectors of predicted macroblock under low sampling rate.
Described steps A 2) further comprise:
A21) adopt median method from the motion vectors of three adjacent macroblocks of predicted macroblock, to obtain motion vectors V2;
A22) original image is carried out 2: 1 sampling, respectively with A1) in the V1 and the A21 that obtain) in the V2 that obtains be that initial position is made the match search in [2,1] scope, obtain the motion vectors of predicted macroblock under low sampling rate.
Described steps A 1) comprising:
Original image is carried out 4: 1 sampling, and predicted macroblock is sampled to one 4 * 4 current image block, and the match search in [4,3] scope is broken down into current image block and carries out at the match search of 4 original position predicted macroblock in the corresponding position.
Described steps A 2) comprising:
Original image carried out 2: 1 sampling, predicted macroblock is sampled to one 8 * 8 current image block, predicted macroblock is in the corresponding position [2,1] match search that is broken down into 44 * 4 image blocks of the match search in the scope carries out, the absolute error of these 44 * 4 image blocks and corresponding searching image piece and add up the absolute error that obtains current 8 * 8 image blocks with.
Described step B) comprising:
Directly handle predicted macroblock, the size of current image block is 16 * 16, predicted macroblock is in the corresponding position [2,1] match search that is broken down into 16 4 * 4 image blocks of the match search in the scope carries out, the absolute error of these 16 4 * 4 image blocks and corresponding searching image piece and add up the absolute error that obtains current image block with.
Described step B) further comprise, realize the advanced prediction mode of MPEG-4 in the following manner:
Predicted macroblock is in the corresponding position [2,1] match search that is broken down into 16 4 * 4 image blocks of the match search in the scope carries out, 4 groups of 48 * 8 subimage blocks that these 16 4 * 4 image blocks are divided into again predicted composition macro block respectively, the absolute error of 4 group of 4 * 4 image block and add up and export, the comparison, the motion vectors that obtains 48 * 8 subimage blocks of predicted macroblock is that the senior of MPEG-4 predicts the outcome.
The object of the present invention is achieved like this: a kind of hierarchical motion estimation very lagre scale integrated circuit (VLSIC) is characterized in that: this circuit comprises pe array, data storage area, controller and comparison arbiter, wherein:
Data storage area storage predicted macroblock data and search window data;
Pe array comprises N * N processing unit, the absolute error that is used to calculate N * N N * N searching image piece in the scope of current N * N image block and a certain position [N/2, N/2-1] with;
Relatively decision device is used for the result of calculation according to pe array output, obtains motion vectors and gives controller;
Controller is used for: deliver to pe array from the data storage area current N * N image block data and searching image blocks of data, the operation of controlled processing unit array, give the comparison decision device result of calculation of pe array, and from comparing the motion vectors that decision device obtains predicted macroblock.
Described processing unit is realized current N * N image block and a certain position [N/2 under the control of controller, N/2-1] in N * N the searching image piece in the scope piece absolute error and calculating, the relative position of described searching image piece in [N/2, N/2-1] scope fixed.
Described processing unit is realized current N * N image block and a plurality of search N * absolute error of N image block and the calculating that adds up under the control of controller.
Described processing unit comprises that further an accumulator group of being made up of 4 accumulators and size are 4 register, in order to realize the senior prediction of MPEG-4, wherein 4 accumulators be responsible for adding up respectively 48 * 8 subimage blocks of predicted macroblock and searching image piece absolute error and, register is responsible for preserving the motion vectors of 4 number of sub images pieces.
The mode that described controller is used for from the data storage area searching image blocks of data is delivered to pe array is:
Controller is used for dividing the searching image blocks of data odd and even number row pixel value two-way to send into described pe array with pipeline system, the odd column data are than sluggish 4 cycles of even column data, and the controlled processing unit array every its value object of four cycles by wherein one the tunnel being converted to another road.
The mode that described controller is used for from the data storage area current N * N image block data is delivered to pe array is:
Controller is used for the present image blocks of data is sent into pe array with parallel mode, and in the same cycle, all processing units use same pixel value in the current image block.
The present invention is a kind of layering motion estimation method and very lagre scale integrated circuit (VLSIC), undertaken by carrying out match search and the match search of using all is decomposed into a plurality of basic search unit under multiple sample rate, thereby obtained following good effect: search speed obviously is better than full-search algorithm; The array structure of very lagre scale integrated circuit (VLSIC) is simple, low in energy consumption, size is little, under the prerequisite that does not increase amount of calculation and computing array, utilizing the accumulator group of being made up of 4 accumulators increase in processing unit and size is that 4 register just can be supported the substandard advanced prediction mode of MPEG-4.
Description of drawings
Fig. 1 is the layering motion estimation method flow chart.
Fig. 2 is sub-block search precedence diagram.
Fig. 3 is the circuit overall schematic.
Fig. 4 is a PE array junctions composition.
Fig. 5 is the PE structure chart.
Fig. 6 is the accumulator group structure chart that is used for advanced prediction mode.
Fig. 7 is the level overlapping performance plot of data.
Fig. 8 is the input timing figure of PE array.
Embodiment
For making purpose of the present invention, technical scheme and advantage clearer, following with reference to the accompanying drawing embodiment that develops simultaneously, the present invention is described in more detail, in following introduction, the N value is 4, the match search of image block realizes by the match search that calls 4 * 4 image blocks repeatedly that all the number of PE equals 4 * 4 in the PE array, promptly 16.
The present invention is a kind of layering motion estimation method, and this method realizes that among the MPEG-4 estimation of 16 * 16 predicted macroblock being divided into four steps the entire method flow process is referring to shown in Figure 1, and sub-block search order is referring to shown in Figure 2.
101) original image is carried out 4: 1 sampling, under sampling in 4: 1, in one 4 * 4 image block of 4 picture element width of x direction, 4 picture element width of y direction, extract the sampled value of a picture element as this image block, in MPEG-4, the size of predicted macroblock is 16 * 16, is sampled as 14 * 4 current image block under sampling in 4: 1.The search best match position obtains motion vectors V1 in the scope of predicted macroblock corresponding position [4,3].In predicted macroblock corresponding position [4,3] search best match position in the scope, be broken down into current image block and carry out in the basic search unit of 4 original position, the basic search unit is meant the match search of one 4 * 4 current image block in the scope of a certain position [2,1].In the present invention, the criterion that the search best match position adopts is absolute error and (SAD) criterion, and the searching image piece best with the current image block coupling is: with the image block of current image block in the absolute value sum minimum of same position place picture element value difference.。
Step 101) corresponding position [4 described in, 3] scope be meant with the predicted macroblock correspondence position be the center, the x coordinate is [4,3] in the scope, the y coordinate is also [4,3] 8 * 8 in the scope, in predicted macroblock corresponding position [4,3] search best match position in the scope, exactly predicted macroblock be the searching image piece at center with each point in these 8 * 8, totally 64 searching image pieces mate.
102) adopt median method from the motion vectors of three adjacent macroblocks of predicted macroblock, to obtain the motion vectors V2 of this predicted macroblock.Described three adjacent macroblocks are the MPEG-4 defined, and in prediction before this, the motion vectors of these three macro blocks obtains.
103) original image is carried out 2: 1 sampling, predicted macroblock is sampled as 18 * 8 current image block, and current 8 * 8 image blocks are made the interior match search of [2,1] scope at two initial position V1 and V2 place, obtain a best match position, obtain corresponding motion vectors V3.8 * 8 current image block are in a certain position [2,1] the basic search unit that is broken down into 44 * 4 image blocks of the match search in the scope carries out, in the match search process, the absolute error of 44 * 4 image blocks that add up and obtain current 8 * 8 image blocks absolute error and.
Described step 101) and 102) in, the method that adopts layering motion estimation method to combine with the adjacent macroblocks motion vectors is step 103) two initial matching points are provided, improved the estimation precision.
104) directly handle original image, predicted macroblock is exactly a current image block, and size is 16 * 16, and current image block is made the match search in [2,1] scope at the V3 place, obtains the final motion vectors of predicted macroblock.The basic search unit that match search in [2,1] scope that current image block is done is broken down into 16 4 * 4 image blocks carries out, and the SAD of these 16 4 * 4 image blocks is added up obtains the SAD of current image block.In said process, can realize the advanced prediction mode of MPEG-4 in the following manner: current image block is made up of 16 4 * 4 image blocks, these 16 image blocks can be divided into 4 groups again, 48 * 8 subimage blocks of difference predicted composition macro block, when the SAD that carries out these 16 4 * 4 image blocks adds up, the absolute error of 4 group of 4 * 4 image block and add up and export, the comparison, the motion vectors that obtains 48 * 8 subimage blocks of predicted macroblock is that the senior of MPEG-4 predicts the outcome.
The present invention also provides a kind of very lagre scale integrated circuit (VLSIC), form by PE array, data storage area, controller and comparison arbiter, wherein the data storage area is made up of current data storage area and 2 parts of search data storage area, integrated circuit as shown in Figure 3, among the figure: the current data storage area, be used to store the predicted macroblock data, in running, need to read repeatedly; The search data storage area is used to store the data of whole search window; Controller is used to control reading of current image block and searching image piece, pe array and comparison decision device; The PE array is used to realize the basic search unit; Relatively decision device is used for the result of calculation of each basic search unit is compared judgement.
Described PE array is made up of 16 PE, with the judgement comparator, finishes the match search of 4 * 4 image blocks in the scope of a certain position [2,1] under the control of controller, and this match search is called the basic search unit.Described PE array structure as shown in Figure 4, among the figure: SW, searching image blocks of data; Cur, the current block data; S00, searching image piece the 0th first pixel value of row; PE, processing unit; Acc
*, the accumulator group.
Described PE finishes one 4 * 4 image block and a certain position [2,1] coupling of a piece in 16 image blocks in the scope, obtain the SAD of current 4 * 4 image blocks and this piece, and the SAD of a plurality of 4 * 4 image blocks that can under control, add up, it is 4 register that each PE also has an accumulator group of being made up of 4 accumulators and size, to realize the senior prediction of MPEG-4.The structure of PE is formed as shown in Figure 5, among the figure: Cur, current block data; SW, the searching image blocks of data; Reg, register; Acc, accumulator.
Described PE also comprises the accumulator group that is used for advanced prediction mode, its structure is as shown in Figure 6: 16 current 4 * 4 image blocks that PE calculates are divided into 4 groups with the sad value of search 4 * 4 image blocks, send in 4 accumulators the sad value of 48 * 8 subimage blocks obtaining predicted macroblock of adding up respectively.
Very lagre scale integrated circuit (VLSIC) of the present invention is realized a basic search unit in the following manner, i.e. the match search of one 4 * 4 current image block in the scope of a certain position [2,1].In this course, use the level overlapping characteristic of data.Level overlapping characteristic is meant that two searching image pieces that successively mate with current image block have overlapping, promptly have common some columns certificates as shown in Figure 7.Characteristic in view of the above, the searching image blocks of data is divided into odd even two row and takes the mode of streamline to enter the PE array, wherein the even column data are than sluggish 4 cycles of odd column, the searching image blocks of data at the input timing of PE array as shown in Figure 8, among the figure: C () represents predicted macroblock, S () represents the searching image piece, and these data of black matrix data representation are taken from the odd column of searching image piece, and other data are taken from the even column of searching image piece.Each PE selects a wherein circuit-switched data by the multiplexer of the alternative of controller control, with the data computation difference of current 4 * 4 image blocks, and difference is sent into accumulator add up.Before beginning calculating, need make data in registers group, progressively be displaced to state shown in Figure 4 through the preload of data in 16 cycles, multiplexer is in high-impedance state before this, and accumulator is not accepted data from PE.Multiplexer enters operating state then, selects a circuit-switched data to send into PE from two paths of data, with current image block data computation difference, the result is sent into accumulator add up.In preceding four computing cycles, sequence number be 0～3 and 8～11 PE from the odd column value, other PE are from the even column value.Four all after dates, they just change by the object of value, sequence number be 0～3 and 8～11 PE from the even column value, after this other unit are changed every four cycles once more from the odd column value, up to finishing a search of 4 * 4.In the computational process, the data of two column registers continue to keep shifting function.At 16 all after dates, the PE array obtains in current 4 * 4 image blocks and a certain position [2, the 1] scope sad value of totally 16 searching image pieces, and these sad values are input to the comparison decision device by each PE, and relatively decision device is determined minimum value wherein.The searching image piece that each PE calculates is [2,1] relative position in the scope is fixed, relatively decision device only need judge minimum sad value is from which PE, just can obtain corresponding motion vectors, do not need to add the adjunct circuit of dedicated memory location, make very lagre scale integrated circuit (VLSIC) structure of the present invention more succinct.
In the present invention, all search of layering motion estimation method are all by calling realization repeatedly to the basic search unit.Illustrate respectively below how very lagre scale integrated circuit (VLSIC) of the present invention needing in the described layering motion estimation method realizes each function of search of using.
Current 4 * 4 current image block are in the corresponding position [4,3] search in the scope: be broken down into 4 basic search unit and carry out, in these four basic search unit, respectively be written into whole [4 from the search data storage area to the PE array by control circuit, 3] data in certain position [2, the 1] scope in the scope.
Current 8 * 8 image blocks are in a certain position [2,1] match search in the scope: the basic search unit that is broken down into 44 * 4 image blocks carries out, in search procedure, by control circuit the basic search unit is controlled, the SAD of these 44 * 4 image blocks that add up obtains the SAD of current 8 * 8 image blocks.Specifically, just the result is outputed in the comparison decision device with the piece that PE in the basic search unit has calculated 4 * 4, to carry out clear operation simultaneously different, in this search, calculated the not zero clearing of accumulator of PE behind one 4 * 4 the piece, in the period of the preload next one 4 * 4 blocks of data, keep its value constant, wait for next 4 * 4 calculating, calculated four 4 * 4 ability continuously the result is exported, promptly need to add up 64 times.
Current 16 * 16 image blocks are in a certain position [2,1] match search in the scope: the basic search unit that is broken down into 16 4 * 4 image blocks carries out, by control circuit the basic search unit is controlled, the SAD of these 16 4 * 4 image blocks added up obtain the SAD of current image block, specifically, in this search, calculated the not zero clearing of accumulator of PE behind one 4 * 4 the piece, in the period of the preload next one 4 * 4 blocks of data, keep its value constant, wait for next 4 * 4 calculating, calculated 16 4 * 4 ability continuously the result has been exported, promptly needed to add up 256 times.
The advanced prediction mode of MPEG-4: carrying out 16 * 16 image blocks in a certain position [2,1] in the match search process in the scope, the sad value of 4 group of 4 * 4 image block of 48 * 8 subimage blocks of forming 16 * 16 predicted macroblock respectively is under the instruction of controller, be imported in the accumulator group of PE going in the accumulator of 8 * 8 subimage blocks, by add up the respectively SAD of this 4 group of 4 * 4 image block of these 4 accumulators, and the output result has just obtained the senior of MPEG-4 and has predicted the outcome to register.
In the introduction of the foregoing description, the N value is 4, and the value of N can also be 8,2 among the present invention, and embodiment is similar to the above embodiments.
The present invention is not limited to the concrete introduction of the foregoing description.The present invention can further upgrade, and perhaps new in addition to feature disclosed in this invention combination also can be upgraded or newly makes up any method disclosed in this invention or process.
Claims (14)
1, a kind of layering motion estimation method is characterized in that, this method may further comprise the steps:
A) original image is carried out at least low sampling rate sampling, when each low sampling rate is sampled, call the match search of N * N image block repeatedly and realize the match search of predicted macroblock, obtain the motion vectors of predicted macroblock under low sampling rate, the width of described N presentation video piece or height;
B) directly handle original image, with A) in the motion vectors of predicted macroblock under low sampling rate obtained be that initial position is made match search, obtain the final motion vectors of predicted macroblock.
2, layering motion estimation method according to claim 1 is characterized in that, the value of N gets 8,4 or 2.
3, layering motion estimation method according to claim 1 is characterized in that, described steps A) comprising:
A1) original image is carried out 4: 1 sampling, in the scope of predicted macroblock corresponding position [4,3], make match search, obtain motion vectors V1;
A2) original image is carried out 2: 1 sampling, with A1) in the V1 that obtains be that initial position is made the match search in [2,1] scope, obtain the motion vectors of predicted macroblock under low sampling rate.
4, layering motion estimation method according to claim 3 is characterized in that, described steps A 2) further comprise:
A21) adopt median method from the motion vectors of three adjacent macroblocks of predicted macroblock, to obtain motion vectors V2;
A22) original image is carried out 2: 1 sampling, respectively with A1) in the V1 and the A21 that obtain) in the V2 that obtains be that initial position is made the match search in [2,1] scope, obtain the motion vectors of predicted macroblock under low sampling rate.
5, layering motion estimation method according to claim 3 is characterized in that, described steps A 1) comprising:
Original image is carried out 4: 1 sampling, and predicted macroblock is sampled to one 4 * 4 current image block, and the match search in [4,3] scope is broken down into current image block and carries out at the match search of 4 original position predicted macroblock in the corresponding position.
6, layering motion estimation method according to claim 3 is characterized in that, described steps A 2) comprising:
Original image carried out 2: 1 sampling, predicted macroblock is sampled to one 8 * 8 current image block, predicted macroblock is in the corresponding position [2,1] match search that is broken down into 44 * 4 image blocks of the match search in the scope carries out, the absolute error of these 44 * 4 image blocks and corresponding searching image piece and add up the absolute error that obtains current 8 * 8 image blocks with.
7, layering motion estimation method according to claim 1 is characterized in that, described step B) comprising:
Directly handle predicted macroblock, the size of current image block is 16 * 16, predicted macroblock is in the corresponding position [2,1] match search that is broken down into 16 4 * 4 image blocks of the match search in the scope carries out, the absolute error of these 16 4 * 4 image blocks and corresponding searching image piece and add up the absolute error that obtains current image block with.
8, layering motion estimation method according to claim 1 is characterized in that, described step B) further comprise, realize the advanced prediction mode of MPEG-4 in the following manner:
Predicted macroblock is in the corresponding position [2,1] match search that is broken down into 16 4 * 4 image blocks of the match search in the scope carries out, 4 groups of 48 * 8 subimage blocks that these 16 4 * 4 image blocks are divided into again predicted composition macro block respectively, the absolute error of 4 group of 4 * 4 image block and add up and export, the comparison, the motion vectors that obtains 48 * 8 subimage blocks of predicted macroblock is that the senior of MPEG-4 predicts the outcome.
9, a kind of hierarchical motion estimation very lagre scale integrated circuit (VLSIC) is characterized in that: this circuit comprises pe array, data storage area, controller and comparison arbiter, wherein:
Data storage area storage predicted macroblock data and search window data;
Pe array comprises N * N processing unit, the absolute error that is used to calculate N * N N * N searching image piece in the scope of current N * N image block and a certain position [N/2, N/2-1] with;
Relatively decision device is used for the result of calculation according to pe array output, obtains motion vectors and gives controller;
Controller is used for: deliver to pe array from the data storage area current N * N image block data and searching image blocks of data, the operation of controlled processing unit array, give the comparison decision device result of calculation of pe array, and from comparing the motion vectors that decision device obtains predicted macroblock.
10, hierarchical motion estimation very lagre scale integrated circuit (VLSIC) according to claim 9 is characterized in that:
Described processing unit is realized current N * N image block and a certain position [N/2 under the control of controller, N/2-1] in N * N the searching image piece in the scope piece absolute error and calculating, wherein, the relative position of described searching image piece in [N/2, N/2-1] scope fixed.
11, hierarchical motion estimation very lagre scale integrated circuit (VLSIC) according to claim 9 is characterized in that:
Described processing unit is realized current N * N image block and a plurality of search N * absolute error of N image block and the calculating that adds up under the control of controller.
12, hierarchical motion estimation very lagre scale integrated circuit (VLSIC) according to claim 9 is characterized in that:
Described processing unit comprises that further an accumulator group of being made up of 4 accumulators and size are 4 register, in order to realize the senior prediction of MPEG-4, wherein 4 accumulators be responsible for adding up respectively 48 * 8 subimage blocks of predicted macroblock and searching image piece absolute error and, register is responsible for preserving the motion vectors of 4 number of sub images pieces.
13, hierarchical motion estimation very lagre scale integrated circuit (VLSIC) according to claim 9 is characterized in that, the mode that described controller is used for from the data storage area searching image blocks of data is delivered to pe array is:
Controller is used for dividing the searching image blocks of data odd and even number row pixel value two-way to send into described pe array with pipeline system, the even column data are than sluggish 4 cycles of odd column data, and the controlled processing unit array every its value object of four cycles by wherein one the tunnel being converted to another road.
14, hierarchical motion estimation very lagre scale integrated circuit (VLSIC) according to claim 9 is characterized in that, the mode that described controller is used for from the data storage area current N * N image block data is delivered to pe array is:
Controller is used for the present image blocks of data is sent into pe array with parallel mode, and in the same cycle, all processing units use same pixel value in the current image block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200510093217 CN1852442A (en) | 2005-08-19 | 2005-08-19 | Layering motion estimation method and super farge scale integrated circuit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200510093217 CN1852442A (en) | 2005-08-19 | 2005-08-19 | Layering motion estimation method and super farge scale integrated circuit |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN1852442A true CN1852442A (en) | 2006-10-25 |
Family
ID=37133909
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200510093217 Pending CN1852442A (en) | 2005-08-19 | 2005-08-19 | Layering motion estimation method and super farge scale integrated circuit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1852442A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101576894B (en) * | 2008-05-09 | 2012-05-30 | 中国科学院半导体研究所 | Real-time image content retrieval system and image feature extraction method |
| CN102647588A (en) * | 2011-02-17 | 2012-08-22 | 北京大学深圳研究生院 | GPU (Graphics Processing Unit) acceleration method used for hierarchical searching motion estimation |
| CN102790884A (en) * | 2012-07-27 | 2012-11-21 | 上海交通大学 | Hierarchical motion estimation-based search method and implementation system thereof |
| CN103139562A (en) * | 2011-11-30 | 2013-06-05 | 富士通株式会社 | Motion estimation method and motion estimation device |
| CN103686188A (en) * | 2012-09-14 | 2014-03-26 | 扬智电子科技(上海)有限公司 | Motion estimation search window mapping method and motion estimation module |
| CN103716639A (en) * | 2013-12-25 | 2014-04-09 | 同观科技（深圳）有限公司 | Search algorithm of frame image motion estimation |
| CN103873874A (en) * | 2014-02-19 | 2014-06-18 | 同观科技（深圳）有限公司 | Full search motion estimation method based on programmable parallel processor |
| CN108900846A (en) * | 2018-07-17 | 2018-11-27 | 珠海亿智电子科技有限公司 | A kind of the two-dimensional directional motion estimation hardware circuit and its method of Video coding |
-
2005
- 2005-08-19 CN CN 200510093217 patent/CN1852442A/en active Pending
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101576894B (en) * | 2008-05-09 | 2012-05-30 | 中国科学院半导体研究所 | Real-time image content retrieval system and image feature extraction method |
| CN102647588B (en) * | 2011-02-17 | 2014-09-24 | 北京大学深圳研究生院 | GPU (Graphics Processing Unit) acceleration method used for hierarchical searching motion estimation |
| CN102647588A (en) * | 2011-02-17 | 2012-08-22 | 北京大学深圳研究生院 | GPU (Graphics Processing Unit) acceleration method used for hierarchical searching motion estimation |
| CN103139562B (en) * | 2011-11-30 | 2016-05-04 | 富士通株式会社 | Method for estimating and device |
| CN103139562A (en) * | 2011-11-30 | 2013-06-05 | 富士通株式会社 | Motion estimation method and motion estimation device |
| CN102790884B (en) * | 2012-07-27 | 2016-05-04 | 上海交通大学 | A kind of searching method based on hierarchical motion estimation and realize system |
| CN102790884A (en) * | 2012-07-27 | 2012-11-21 | 上海交通大学 | Hierarchical motion estimation-based search method and implementation system thereof |
| CN103686188A (en) * | 2012-09-14 | 2014-03-26 | 扬智电子科技(上海)有限公司 | Motion estimation search window mapping method and motion estimation module |
| CN103686188B (en) * | 2012-09-14 | 2017-02-22 | 扬智电子科技(上海)有限公司 | Motion estimation search window mapping method and motion estimation module |
| CN103716639A (en) * | 2013-12-25 | 2014-04-09 | 同观科技（深圳）有限公司 | Search algorithm of frame image motion estimation |
| CN103873874A (en) * | 2014-02-19 | 2014-06-18 | 同观科技（深圳）有限公司 | Full search motion estimation method based on programmable parallel processor |
| CN103873874B (en) * | 2014-02-19 | 2017-06-06 | 同观科技（深圳）有限公司 | A kind of full search method for estimating based on programmable parallel processor |
| CN108900846A (en) * | 2018-07-17 | 2018-11-27 | 珠海亿智电子科技有限公司 | A kind of the two-dimensional directional motion estimation hardware circuit and its method of Video coding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1157956C (en) | Exercise estimator | |
| CN1852442A (en) | Layering motion estimation method and super farge scale integrated circuit | |
| CN1297134C (en) | Moving estimating device and method for reference macro block window in scanning search area | |
| CN1290342C (en) | Apparatus capable of performing both block-matching motion compensation and global motion compensation and method thereof | |
| CN1794814A (en) | Pipelined deblocking filter | |
| CN1767655A (en) | Multi view point video image parallax difference estimating method | |
| CN1921628A (en) | Motion vector detection apparatus and motion vector detection method | |
| CN105681807A (en) | Method and device for calculating sub pixel motion vector based on H264 protocol | |
| CN102790884B (en) | A kind of searching method based on hierarchical motion estimation and realize system | |
| JP2004173224A (en) | Method and apparatus for motion estimation by full binary expression | |
| CN101039430A (en) | Method for scanning quickly residual matrix in video coding | |
| CN101044765A (en) | Apparatus and method for image coding | |
| CN101729893B (en) | MPEG multi-format compatible decoding method based on software and hardware coprocessing and device thereof | |
| CN1703094A (en) | Image interpolation apparatus and methods that apply quarter pel interpolation to selected half pel interpolation results | |
| CN1589028A (en) | Predicting device and method based on pixel flowing frame | |
| CN1214648C (en) | Method and apparatus for performing motion compensation in a texture mapping engine | |
| CN102801982B (en) | Estimation method applied on video compression and based on quick movement of block integration | |
| CN1946178A (en) | VLSI device for movement evaluation and method for movement evaluation | |
| Ismail et al. | High performance architecture for real-time HDTV broadcasting | |
| CN1745587A (en) | Method of video coding for handheld apparatus | |
| CN1520187A (en) | System and method for video data compression | |
| CN1201589C (en) | Motion estimation | |
| CN1780402A (en) | Video image motion compensator | |
| CN104602026B (en) | A kind of reconstruction loop structure being multiplexed entirely encoder under HEVC standard | |
| CN100340118C (en) | Super large scale integrated circuit system structure of moving estimation and data buffer storage method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Open date: 20061025 |