CN101860745A - Change quantization optimizing method for video monitoring system - Google Patents
Change quantization optimizing method for video monitoring system Download PDFInfo
- Publication number
- CN101860745A CN101860745A CN 200910061450 CN200910061450A CN101860745A CN 101860745 A CN101860745 A CN 101860745A CN 200910061450 CN200910061450 CN 200910061450 CN 200910061450 A CN200910061450 A CN 200910061450A CN 101860745 A CN101860745 A CN 101860745A
- Authority
- CN
- China
- Prior art keywords
- reduce
- instruction
- monitoring system
- video monitoring
- quantization
- 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
Landscapes
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention belongs to the technical field of video coding and decoding in signal processing, and particularly relates to a change quantization method. The invention provides a change quantization optimizing method for a video monitoring system, which is applicable to video coding and decoding standards of MPEG, H.26x and AVS and the like. Through the steps of data packing, instruction scrambling, stream circulation and the like, the method makes full use of a register and a functional unit of a CUP, improves the operational parallelism degree, meanwhile reduces date read-write and movement, and reduces the use of a memory so as to reduce the time and spatial complexity of a coder and realize real-time coding and decoding.
Description
Technical field
The invention belongs to the video coding and decoding technology field in the signal processing, be specifically related to a kind of change quantization method.
Background technology
Along with fast development of information technology, the transmission of image and video more and more is subjected to extensive concern.Because the vedio data amount is very big, transmission data bandwidth and data storage capacity there are very high requirement, common bandwidth and storage medium are difficult to satisfy the demands.Therefore, it is necessary data being compressed.The main flow video compression standard mainly comprises the MPEG series standard that Motion Picture Experts Group (Moving Picture Expert Group) formulates at present, the H.26x series standard that International Telecommunications Union (International Telecommunication Union) formulates, and China have independent intellectual property right advanced audio/video encoding standard (Advanced Audio/Video Standard, AVS).Above-mentioned standard can reach tens to hundreds of times compression ratio, for Network Transmission provides possibility.
Yet, MPEG, H.26x, standard such as AVS when bringing high compression rate, self also has high time and space complexity.For some specific occasions, especially video monitoring system often is difficult to direct operation without the program of optimizing.Common video monitoring system is many, and all there are restriction in the operational capability of CPU and the memory capacity of system based on embedded platform, and supervisory control system itself is high to the real-time requirement, therefore, must be optimized program.
Statistics shows that the operand of change quantization module accounts for the 20%-30% of whole video coding framework, becomes the bottleneck in the video coding.Though adopted butterfly computation in the mainstream standard at present, use displacement and add operation to replace original matrix multiplication, but still need a large amount of memory space and high computational complexities continuously.Therefore, be necessary the change quantization module is further optimized, improve arithmetic speed, reduce taking of memory space, thereby realize real-time coding.
Summary of the invention
Main purpose of the present invention provides a kind of less quick conversion and quantification method of resource that takies that is applicable to video monitoring system, and its step comprises:
1. calculate residual error and horizontal transformation.
1.1 the original value and the predicted value of delegation are read in respectively in the corresponding register by the packing reading command.
1.2 data extending.Use instruction that the byte unsigned number of reading in is extended for the double byte signed number.
1.3 use instruction to the high status of 32 bit registers respectively correspondence subtract each other, in one-period, carry out twice subtraction simultaneously.
1.4 high 16 and low 16 s' storage order is convenient to carry out butterfly computation in the adjustment register.
1.5 use dot product command calculations result.
2. vertical conversion.
If use the packing instruction can cause data to be overflowed in the vertical conversion, therefore do not carry out any optimization.
3. quantize.
3.1 use the packing instruction that 2 16 signed numbers are stored in the high 16 and low 16 of same 32 bit registers.
3.2 merging, coefficient multiplies each other.Use instruction to multiply each other, carry out 2 multiplyings simultaneously obtaining the data correspondence in 3.1.
3.3 use displacement to replace division arithmetic.
3.4 calculate the number of nonzero coefficient.
4. complete zero piece detects.
If the nonzero coefficient number of statistics is 0 in 3.4, then need not carry out step 5 and step 6, directly jump to step 7.Otherwise jump to step 5.
5. inverse quantization and horizontal inverse transformation.
5.1 each line data is carried out multiply-add operation.
5.2 each line data is carried out horizontal anti-change, and method is with step 1.3-1.5.
6. vertical inverse transformation.
6.1 the coefficient of each row is packaged in 32 bit registers.
6.2 use dot product command calculations result.
6.3 use with saturated packing instruction the result in 6.2 carried out the clamper operation.
7. reconstruct.
According to corresponding predicted value and residual computations reconstruction value.
Embodiment
Below at the instruction set of C64x+, be example with the AVS standard, specify better embodiment of the present invention.Step is as follows:
1. calculate residual error and horizontal transformation.
1.1 at first with the delegation of the original value of 8x8 piece, promptly 88 signless integers are read in 2 32 general registers, corresponding predicted value read method is the same.
1.2 48 signless integers in each register that obtains in 1.1 are extended for 16 signed integers, exist in 2 32 general registers.
1.3 use the SUB2 instruction that the high low level of the register that obtains in 1.2 is carried out subtraction respectively, in 1 cycle, obtain 2 operation results, thereby reduce the running time of half.
1.4 according to the AVS identifying code, butterfly conversion for the first time need be carried out following calculating:
b[0]＝curr_blk[yy][0]+curr_blk[yy][7]；
b[1]＝curr_blk[yy][1]+curr_blk[yy][6]；
B[0 wherein]～b[7] be the result who obtains in 1.3.In one-period, carry out twice calculating in order to use the ADD2 instruction, need to adjust the storage order of high low level in the register, make curr_blk[yy] [1] and curr_blk[yy] [0] be in the high 16 and low 16 of a register, curr_blk[yy respectively] [6] and curr_blk[yy] [7] be in the high 16 and low 16 of another register respectively.Other compute classes seemingly.
1.5, need carry out following calculating during the butterfly conversion according to the AVS identifying code:
b[0]＝(tmp[0]+tmp[1])＜＜3；
b[1]＝(tmp[0]-tmp[1])＜＜3；
b[2]＝((tmp[2]*10)+(tmp[3]＜＜2))；
b[3]＝((tmp[2]＜＜2)-(tmp[3]*10))；
For first statement, constant [8,8] can be merged into a coefficient 0x00080008, use the DOTP2 instruction once to calculate and finish.Other compute classes seemingly.
2. vertical conversion.
Overflow for fear of data, vertical conversion only uses displacement and addition to replace matrix multiplication, does not use the packing operation.
3. quantize.
Residual error after the conversion that step 2 is obtained is packaged as one 32 figure place in twos, uses the MPY2 instruction to do multiplying respectively with high 16 and low 16, carries out 2 multiplication in 1 cycle.Replace division with shift operation, reduce computational complexity.The number of nonzero coefficient after the while statistic quantification.
4. complete zero piece detects.
If the nonzero coefficient number of statistics is 0 in the step 3, show that then current block is complete zero piece, need not carry out inverse quantization and inverse transformation, directly jump to step 7.Otherwise jump to step 5.
5. inverse quantization and horizontal inverse transformation.
5.1 each line data is carried out multiply-add operation, and concrete grammar is with step 3.
5.2 each line data is carried out horizontal anti-change, and method is with step 1.3-1.5.
6. vertical inverse transformation.
6.1 the coefficient of each row is packaged in 32 general registers in order.
6.2 use dot product command calculations result.Method is with step 1.3-1.5.
6.3 use the result in 6.2 is limited in the interval of [0,255] with saturated packing instruction SPACK2.
7. reconstruct.
If directly jump to step 7 from step 4, then reconstruction value equals predicted value.If to step 7, then reconstruction value equals predicted value and residual error sum through step 5,6.
Claims (8)
1. the change quantization optimizing method in the video monitoring system is characterized in that: will obtain residual error, conversion, quantification, inverse quantization, inverse transformation, 6 parts of reconstruct and be merged into same function, and reduce the expense of function call, thereby improve the speed of service.
2. method according to claim 1 is characterized in that: will obtain the residual sum horizontal transformation and combine, thereby reduce the number of times that reads and writes data.
3. method according to claim 1 is characterized in that: use the packing instruction that 48 signless integers or 2 16 signed integers are packaged as 1 32 figure place, read in the general register.Use instructions such as ADD4, ADD2, SUB4, SUB2, MPYU2 to carry out simultaneously 2 times or 4 arithmetical operations.
4. method according to claim 1 is characterized in that: adjust alignment sequence of data in each row or the row, thereby can use the dot product instruction to replace multiplication and add operation.
5. method according to claim 1 is characterized in that: use with saturated packing instruction dateout is carried out the clamper operation, reduce data number of times relatively.
6. method according to claim 1 is characterized in that: add complete zero piece detection module, for complete zero piece, need not carry out inverse quantization and inverse transformation, thereby effectively reduce computation complexity.
7. method according to claim 1 is characterized in that: inverse quantization is combined in horizontal inverse transformation, thereby reduce the reading and writing data number of times.
8. method according to claim 1, it is characterized in that: whole function only needs the continuous memory space of 128 bytes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910061450 CN101860745A (en) | 2009-04-08 | 2009-04-08 | Change quantization optimizing method for video monitoring system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910061450 CN101860745A (en) | 2009-04-08 | 2009-04-08 | Change quantization optimizing method for video monitoring system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101860745A true CN101860745A (en) | 2010-10-13 |
Family
ID=42946347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910061450 Pending CN101860745A (en) | 2009-04-08 | 2009-04-08 | Change quantization optimizing method for video monitoring system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101860745A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104683800A (en) * | 2015-02-11 | 2015-06-03 | 广州柯维新数码科技有限公司 | AVS-based methods for parallel quantization and inverse quantization |
-
2009
- 2009-04-08 CN CN 200910061450 patent/CN101860745A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104683800A (en) * | 2015-02-11 | 2015-06-03 | 广州柯维新数码科技有限公司 | AVS-based methods for parallel quantization and inverse quantization |
| CN104683800B (en) * | 2015-02-11 | 2017-12-15 | 广州柯维新数码科技有限公司 | Parallel quantization and quantification method based on AVS |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100592796C (en) | Video frequency encoder and method for choosing frame inner forecast mode | |
| US10165270B2 (en) | Intra/inter mode decision for predictive frame encoding | |
| US20080063080A1 (en) | Hierarchical Motion Estimation Using Original Frame for Sub-Sampled Reference | |
| CN100556142C (en) | The decoder of decode video data and system | |
| Akramullah et al. | Optimization of H. 263 video encoding using a single processor computer: performance tradeoffs and benchmarking | |
| CN101729893B (en) | MPEG multi-format compatible decoding method based on software and hardware coprocessing and device thereof | |
| US20090112931A1 (en) | Method and a System for Determining Predicted Numbers of Processor Cycles Required for Respective Segments of a Media File for Playback of the Media File | |
| CN101188761A (en) | Method for optimizing DCT quick algorithm based on parallel processing in AVS | |
| CN1589028B (en) | Predicting device and method based on pixel flowing frame | |
| CN105208394B (en) | A kind of real-time digital image compression prediction technique and system | |
| CN102932643B (en) | A kind of expansion variable-block motion estimation circuit being applicable to HEVC standard | |
| Chang et al. | Performance analysis and architecture evaluation of MPEG-4 video codec system | |
| CN102572430A (en) | Method for implementing H.264 deblocking filter algorithm based on reconfigurable technique | |
| CN101715124B (en) | Single-input and multi-output video encoding system and video encoding method | |
| CN104378642A (en) | Quick H.264 fractional pixel interpolation method based on CUDA | |
| US10616585B2 (en) | Encoding data arrays | |
| CN106412597A (en) | Work load balancing in multicore video decoder | |
| CN104038766A (en) | Device used for using image frames as basis to execute parallel video coding and method thereof | |
| CN101860745A (en) | Change quantization optimizing method for video monitoring system | |
| CN102075763A (en) | Intra-frame sub-block predictor circuit for video encoder and method for implementing same | |
| CN100576918C (en) | Standard interframe prediction pixel generation device for digital audio and video coding and decoding technology | |
| CN102625109B (en) | Multi-core-processor-based moving picture experts group (MPEG)-2-H.264 transcoding method | |
| Senn et al. | Joint DVFS and parallelism for energy efficient and low latency software video decoding | |
| CN100469146C (en) | Video image motion compensator | |
| Lan et al. | Mpeg2 decoding complexity regulation for a media processor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C57 | Notification of unclear or unknown address | ||
| DD01 | Delivery of document by public notice |
Addressee: He Qingdi Document name: Notification of Passing Preliminary Examination of the Application for Invention |
|
| C06 | Publication | ||
| PB01 | Publication | ||
| DD01 | Delivery of document by public notice |
Addressee: He Qingdi Document name: Notification of Publication of the Application for Invention |
|
| DD01 | Delivery of document by public notice |
Addressee: He Qingdi Document name: Notification of before Expiration of Request of Examination as to Substance |
|
| DD01 | Delivery of document by public notice |
Addressee: He Qingdi Document name: Notification that Application Deemed to be Withdrawn |
|
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20101013 |