CN1460968A - Integer wavelet conversion method based on digital signal processor - Google Patents
Integer wavelet conversion method based on digital signal processor Download PDFInfo
- Publication number
- CN1460968A CN1460968A CN 03148079 CN03148079A CN1460968A CN 1460968 A CN1460968 A CN 1460968A CN 03148079 CN03148079 CN 03148079 CN 03148079 A CN03148079 A CN 03148079A CN 1460968 A CN1460968 A CN 1460968A
- Authority
- CN
- China
- Prior art keywords
- data
- transform
- inverse
- coefficient
- wavelet
- 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.)
- Granted
Links
Images
Landscapes
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The integer wavelet transformation method based on digital signal processor is characterized by that when the wavelent is undergone the process of positive transformation, firstly, dividing the image data into several data sections, then respectively making line data of current data section undergo the process of line positive transformation, and making column data of current data section undergo the process of column positive transformation, and when the vavelet is undergone the process of inverse transformation, firstly, dividing the image data into several data sections, then respectively making column inverse transformation of column data of current data section and line inverse transformation of line data of current data section, crossly making line and column transformation until the whole image data are processed completely.
Description
Technical field
The present invention relates to a kind of small wave converting method, particularly a kind of integer wavelet transformation method based on digital signal processor.
Background technology
At present, wavelet transformation is as the international standard of JPEG2000, is widely used in fields such as the static texture information compression of JEPG2000, MPEG4 and Medical Image Compression.The multiresolution that wavelet transformation has is resolved, can block embedded bitstream, characteristics such as no mosaic have the incomparable advantage of conventional discrete cosine transform (DCT), but wavelet transformation is owing to carry out computing to entire image, rather than as the computing of dct transform piecemeal, therefore the degree of parallelism of computing is not high in the conversion, and hardwareization is very difficult.
Yet in a lot of real systems,, consider the requirement of speed and area, need realize compressibility with chip as digital camera, videophone, field camera and palm PC etc.With the DCT of traditional piecemeal computing relatively, do not make full use of the parallel work-flow of wavelet transformation in the present existing 2D DWT chip structure, though reduced the control complexity, reduced the hardware utilization factor also, increased chip cost.Therefore, the processing capability in real time of research wavelet transformation is of great practical value.
For two dimensional image, traditional small wave converting method is at first whole row of image to be done line translation, afterwards, after all finishing, line translation does rank transformation again, if carry out the wavelet transformation of multilayer, then, proceed two-dimensional wavelet transformation to the low frequency coefficient in the upper left corner behind the last two-dimensional wavelet transformation 1/4.Calculation step and 1,2 identical, the ranks number each divided by 2, if desired wavelet coefficient is quantized, then need after all wavelet transformation numbers of plies finish, carry out quantization operation to each coefficient.
As can be seen, for hardware was realized, above operation had several very big drawbacks:
1, the degree of parallelism of computing is very poor, all is that order is carried out between each step.With two dimensional image size M * N, carrying out one deck wavelet transformation is example, wherein needs M line translation circulation, need N rank transformation circulation, and each line translation needs N/2 circulation, each rank transformation to need M/2 circulation, amounts to (N/2) * M+ (M/2) * N=M * N circulation.
2, computing is carried out entire image, and the memory headroom of hardware DSP is very limited, when picture size is big, view data can only be put in the outer storage space of sheet, and arithmetic speed is reduced greatly.With the CIF form is example, and piece image is of a size of 352 * 288 * 2=198KB, and the ram in slice of TI C6000 DSP can not be satisfied the demand, and when picture size bigger the time, the deficiency of this memory headroom is just more outstanding.
3, in the wavelet transformation of a two dimensional image, a coefficient need be called at least twice (go filtering once, row filtering once) repeatedly, and the outer data dispatch of sheet reduces algorithm efficiency in the frequent sheet.
4, quantization operation is carried out separately after wavelet transformation is finished, and needs M * N cycling, has increased the execution time of computing greatly.
5, the filter parameter of a lot of wavelet transformations being arranged is floating number, and the computing of floating number is very consuming time for hardware, has influenced the realization of real-time.
Improve though some have also occurred, for example the floating-point operation of wavelet transformation is changed into the lifting computing of integer, add computing with displacement and replace the multiplication and division computing at above-mentioned defective; Adopt the common subexpression technology, with the schemes such as public keys merging in low-pass filter and the Hi-pass filter operation, but these improve the operation efficiency that has just improved wavelet transformation from the angle of calculating formula, fundamentally do not solve the integral operation degree of parallelism that wavelet transformation run on hardware is realized low, consume problems such as internal memory, data repetitive schedule.
Summary of the invention
Technical matters to be solved by this invention be to provide a kind of be convenient to hard-wired, based on the small wave converting method of digital signal processor, original image is carried out the positive inverse transformation of real-time small echo.
The invention provides a kind of integer wavelet direct transform method, comprise the steps: based on digital signal processor
(1) view data is divided into several data segments;
(2) to the capable direct transform of line data of current data section;
(3) column data to the current data section carries out the row direct transform;
(4) to other data segment repeating steps (2), (3).
A kind of integer wavelet inverse transform method based on digital signal processor comprises the steps:
(1) view data is divided into several data segments;
(2) column data to the current data section carries out the row inverse transformation;
(3) to the capable inverse transformation of line data of current data section;
(4) to other data segment repeating steps (2), (3).
The present invention is applicable to the digital signal processor that possesses parallel processing capability that memory headroom is less, original image is carried out the positive inverse transformation of real-time integer wavelet, arithmetic speed is faster than floating-point operation, and the wavelet coefficient after the conversion distributes and obviously is better than traditional floating-point wavelet transformation.In conversion, take multiple preselected quantization parameter, under the prerequisite that does not increase computing expense and time overhead, finish quantification wavelet coefficient.This method is convenient to hardware and is realized, can apply to the Video transmission system of a lot of low code checks etc.
Description of drawings
Fig. 1 is the embodiment process flow diagram that the present invention is based on the integer wavelet direct transform method of digital signal processor;
Fig. 2 is the embodiment process flow diagram that the present invention is based on the integer wavelet inverse transform method of digital signal processor;
Fig. 3 uses the capable direct transform computing of small echo of the present invention synoptic diagram;
Fig. 4 uses small echo row direct transform computing synoptic diagram of the present invention;
Fig. 5 uses small echo row inverse transformation computing synoptic diagram of the present invention;
Fig. 6 uses the capable inverse transformation computing of small echo of the present invention synoptic diagram.
Embodiment
The invention provides a kind of integer wavelet transformation method based on digital signal processor, when direct transform, at first view data is divided into several data segments, then respectively to the capable direct transform of line data of current data section, column data to the current data section carries out the row direct transform, row-column transform is interspersed to carry out, until handling whole view data.
Image division is a variety of for the mode of several data segments has, for the small echo direct transform, view data can be divided into a plurality of systemic circulations, the number of times of systemic circulation can be divided exactly by the line number of view data; Again the each row of data that comprises in described each systemic circulation is divided into a plurality of partial circulatings, the number of times of partial circulating can be divided exactly by the columns of described view data.
As shown in Figure 1, promptly be that the present invention is at the embodiment of coding side process flow diagram, a kind of integer wavelet direct transform method based on digital signal processor is provided, at first view data is divided into a plurality of systemic circulations, the number of times of systemic circulation can be divided exactly by the line number of view data, for example, the ranks number average of view data can be divided exactly by 4, then can be with the line number of view data divided by 4 (steps 101); Then the each row of data that comprises in described each systemic circulation is divided into a plurality of partial circulatings, the number of times of partial circulating can be divided exactly by the columns of described view data, then can be with the columns of view data divided by 4 (steps 102); The line data that comprises in the current systemic circulation is done capable direct transform (step 103) respectively; After the capable filtering of described line data is finished, it is done row direct transform (step 104); Repeating step 103,104 then, until whole loop ends (step 105).
Wherein said step 103 comprises: the current line data are done the Lazy conversion, with the coefficient of parity column wherein (that is to say earlier all even coefficients are placed continuously, thereafter all strange coefficients are also being placed continuously then) placed apart; Carry out one-dimensional wavelet transform, wherein one-dimensional wavelet transform comprises odd-transform and even number conversion again.
Described step 104 comprises: the data to the same row of each row are done the lazy conversion, and the coefficient of parity rows wherein is placed apart; Carry out one-dimensional wavelet transform then, wherein one-dimensional wavelet transform comprises odd-transform and even number conversion again.
In described small echo direct transform process, if desired wavelet coefficient is quantized, then be 2 power by default quantization parameter, in the process of wavelet transformation, finish quantification simultaneously.
Wavelet coefficient after the described small echo direct transform adopts a coefficient of dilatation k to be optimized, k=1.2~1.5, and it can be transformed into multiplication of integers and shift operation.
Described small echo direct transform is finished in the sheet internal buffer, and to the data before and after the conversion in sheet and the carrying between the sheet external memory district, the ranks transposition of Lazy conversion and matrix is to use the quick direct memory visit of digital signal processor to finish.
Accordingly, in decoding end, the invention provides a kind of integer wavelet inverse transform method based on digital signal processor, at first view data is divided into several data segments, column data to the current data section carries out the row inverse transformation respectively, to the capable inverse transformation of line data of current data section, the row line translation interts to be carried out, until handling whole view data then.
Equally, the mode of partitioned image can be by view data being divided into a plurality of systemic circulations, and the number of times of systemic circulation can be divided exactly by the columns of view data; The every column data that comprises in described each systemic circulation is divided into a plurality of partial circulatings, and the number of times of partial circulating can be divided exactly by the line number of described view data.
As shown in Figure 2, promptly be the present invention at the embodiment of decoding end process flow diagram, at first view data is divided into a plurality of systemic circulations, the number of times of systemic circulation can be divided exactly by the columns of view data, for example, with the columns of described view data divided by 4 (steps 201); Then the every column data that comprises in described each systemic circulation is divided into a plurality of partial circulatings, the number of times of partial circulating can be divided exactly by the line number of described view data, and the line number that for example is described view data is divided by 4 (steps 202); The part rows data that comprise in the current systemic circulation are done row inverse transformation (step 203) respectively; After the row inverse transformation of described part rows data is finished, it is done capable inverse transformation (step 204); Repeating step 203,204 is until whole loop ends (step 205).
Described step 203 comprises: each column data is done the one dimension inverse wavelet transform, and wherein the one dimension inverse wavelet transform comprises odd number inverse transformation and even number inverse transformation again; To carry out anti-Lazy conversion through the coefficient after the inverse wavelet transform, the coefficient of parity column will wherein be placed (just returning to the placement location of original image, i.e. the order of odd even odd even) at interval.
Described step 204 comprises: each row is done the one dimension inverse wavelet transform with data line, and wherein the one dimension inverse wavelet transform comprises odd number inverse transformation and even number inverse transformation again; To carry out anti-Lazy conversion through the coefficient after the inverse wavelet transform, the coefficient of parity column will wherein be placed at interval.
In described inverse wavelet transform process, if desired wavelet coefficient is carried out inverse quantization, then dequantized coefficients is preset as 1/ (2 powers), in the process of inverse wavelet transform, finish inverse quantization simultaneously.
Wavelet coefficient behind the described inverse wavelet transform adopts a coefficient of dilatation k to be optimized, k=1.2~1.5, and it can be transformed into multiplication of integers and shift operation
Described inverse wavelet transform is finished in the sheet internal buffer, and to the data before and after the conversion in sheet and the carrying between the sheet external memory district, the ranks transposition of Lazy conversion and matrix is to use the quick direct memory visit of digital signal processor to finish.
Below with two dimensional image size M * N, M and N all can be divided exactly by 4, and carrying out one deck integer biorthogonal Daubechies (5,3) wavelet transformation is example, further specifies the present invention, but the invention is not restricted to this embodiment.
Coding side:
1, be M/4 systemic circulation with image division, each systemic circulation comprises N/4 partial circulating again.
2, in each circulation 4~6 line data of image are done capable filtering respectively: at first this line data is done the lazy conversion, the coefficient of parity column wherein is placed apart, carry out one-dimensional wavelet transform then.Wherein one-dimensional wavelet transform comprises odd-transform and even number conversion again.In the process of wavelet transformation, finish quantification simultaneously.
2,4~6 line data that the computing through step 1 wavelet transformation is finished, carry out row filtering simultaneously: at first the data of the same row of each row are done the lazy conversion, the coefficient of parity rows wherein is placed apart, carry out one-dimensional wavelet transform then.Wherein one-dimensional wavelet transform comprises odd-transform and even number conversion again.In the process of wavelet transformation, finish quantification simultaneously.
Repeating step 2,3 is up to loop ends.
From hardware point of view, the period of program is many more, and the execution of code just is not easy to realize flowing water and parallel more, and it is very low to carry out efficient.But if period very little, the required average resource number (such as storage space or the like) that expends is too much in a circulation so, and calculated amount is too complicated, also can influence code and carry out efficient.Therefore, in an embodiment of the present invention, being divided into picturedeep M/4 circulation is a selection of relatively optimizing.Because wavelet transformation is to do capable filtering earlier, after do row filtering, the division of systemic circulation is filtered into the basis with row, so can not select N/4 as big cycle index, the ranks filtering of 4 row coefficients is finished in a systemic circulation, and in addition, the systemic circulation number must be divided exactly by the line number of entire image.In like manner, the division of partial circulating is filtered into the basis with row, has selected N/4 here, that is, a partial circulating is finished the capable filtering of 4 coefficients, and the partial circulating number must be divided exactly by the columns of entire image.Several row of several row of this and carries out image data are reasons, just utilize the characteristics of wavelet transformation, and entire image is cut and handled respectively.
Accordingly, in decoding end:
4, be N/4 systemic circulation with image division, each systemic circulation comprises M/4 partial circulating again.
5, in each circulation 4~6 column data of image are done the row inverse transformation respectively: at first the data of each row are done the one dimension inverse wavelet transform.Wherein the one dimension inverse wavelet transform comprises odd number inverse transformation and even number inverse transformation again.In the process of inverse wavelet transform, finish inverse quantization simultaneously.To carry out anti-lazy conversion through the coefficient after the wavelet inverse transformation at last, the coefficient of parity rows will wherein be placed at interval.
6, several column data of finishing through step 5 inverse wavelet transform are gone inverse transformation simultaneously: at first each row is done the one dimension inverse wavelet transform with data line.Wherein the one dimension inverse wavelet transform comprises odd number inverse transformation and even number inverse transformation again.In the process of inverse wavelet transform, finish inverse quantization simultaneously.To carry out anti-lazy conversion through the coefficient after the wavelet inverse transformation at last, the coefficient of parity column will wherein be placed at interval.
7, repeating step 5,6 is up to loop ends.
Wherein, be example with Daubechies (5,3), the formula of one dimension small echo direct transform is as follows:
s
l (0)Be the even samples of one dimension picture signal x, d
l (0)Be the odd samples of one dimension picture signal x, s
lBe the even samples after the conversion, d
lBe the odd samples after the conversion; α, β are wavelet conversion coefficient, and k is stretching coefficient (α=-0.5, β=0.25, k=1.2~1.5).
The formula of Daubechies (5,3) one dimension inverse wavelet transform is as follows:
In the formula: d
lBe the high frequency samples of signal, s
lBe the low frequency samples of signal, d
l (1)Be the high frequency samples behind stretching, s
l (1)Be the low frequency samples behind stretching, α, β are wavelet conversion coefficient, s
l (0)Be the even samples after the conversion, d
l (0)Be the odd samples after the conversion; The one dimension picture signal (α=-0.5, β=0.25, k=1.2~1.5) of x for recovering.
| α |=0.5, to its floating-point multiplication, can be transformed into and move to right 1.
| β |=0.25, to its floating-point multiplication, can be transformed into and move to right 2.
Wherein the stretching coefficient k chooses, obtain through experiment, when the k value is 1.2~1.5, the effect of further optimization has been played in distribution for the coefficient value behind the shaping wavelet transformation, further big coefficient has been focused on the upper left corner of entire image two-dimensional array, help further encoding computing, make the ratio of compression of image bigger.
The k value can be transformed into multiplication of integers and shift operation, and promptly k multiply by 2
L, round, obtain new drawing coefficient value k
1, L position then moves to right; 1/k also can be transformed into multiplication of integers and shift operation, and promptly 1/k multiply by 2
L, round, obtain new drawing coefficient value k
2, L position then moves to right.
Coefficient of dilatation k has adopted multiplication of integers and shift operation to realize, therefore, when can utilize this operation wavelet coefficient is quantized, and does not need to waste extra circulation.Default quantization parameter only needs to use shift operation, can finish the quantification of different frequency bands wavelet coefficient in the flexible computing of integer wavelet transformation easily and fast.
Quantification to wavelet coefficient is primarily aimed at high frequency coefficient, and purpose is in order to improve the compression efficiency of follow-up Coding Compression Algorithm.Concrete way is quantization parameter to be preset as 2 power, so just can only finish quantification with shifting function.Determine whether according to the raising of the wavelet transformation number of plies by a parameter preset and lower this initial quantization coefficient with 2 integral multiple.
Because through test and theoretical proof, human eye is so inresponsive to the high-frequency information of level and vertical direction to the high-frequency information of diagonal, so we have relatively high expectations to the quantification of the high-frequency information of diagonal.Because the diagonal high-frequency information will multiply by the coefficient of dilatation of twice high frequency coefficient of row, column in one deck two dimension shaping wavelet transformation, thus just the raising of nature the quantification of diagonal high-frequency information.
For example, at coding side, be 2 to the quantized value of low frequency component
n, then
Can be rewritten as
Quantized value to high fdrequency component is 2
n, then
Can be rewritten as
In decoding end, be 1/2 to the inverse quantization value of low frequency component
n, then
Can be rewritten as
Inverse quantization value to high fdrequency component is 1/2
n, then
Can be rewritten as
This quantization algorithm science and efficient are very high, do not need extra time and computing cost just can finish different default quantification purposes.
Because the data of image information amount is bigger, therefore the data of putting in order frame can not be put in the internal memory of DSP, and frequent in sheet and call data outside the sheet, operation efficiency is reduced, and in the present embodiment, the memory requirements of DSP only is max (M, N) * 2 * 6 a byte, i.e. 6 row/row view data.The integer wavelet transform of present embodiment is to finish in less sheet internal buffer, once handles four lines or four row view data at most, carries out efficient to improve.And to the data carrying between the storage portions in sheet and outside the sheet before and after the integer wavelet computing, the ranks transposition of Lazy conversion and matrix, the present invention adopts the quick direct memory visit of digital signal processor to finish, and transporting velocity is fast, and does not take the processing time of processor.
See shown in Figure 3, with the row be filtered into example, can finish the High frequency filter of two odd samples and the low frequency filtering of two even samples in the circulation.Be filtered into example with row, the computing with traditional algorithm needed N circulation to finish originally only needs N/4 circulation just can finish.Improved operation efficiency greatly.
In the conventional procedures of wavelet transformation, the row filtering filtering of need being expert at just can be carried out after executing fully.The what is called of people's proposition in the past improves parallel work-flow and also only is limited to respectively in the capable filtering and in the row filtering, and the concurrent operation between the two is not had to consider.See shown in Figure 4ly, the present invention is with row filtering and row filtering operation parallel processing.Row filtering need just not begin to carry out after all capable filtering is finished, as long as those required several line data of row filtering are finished capable filtering, can carry out the row filtering operation to all column data of these several row.Like this, the cycle index of two dimensional image one deck small echo direct transform has become present M * N/16 from the M * N in past.
The row inverse transformation can be finished the inverse transformation of two high frequency samples and two low frequency samples as shown in Figure 5 in circulation.Be changed to example to go contravariant, the computing with traditional algorithm needed N circulation to finish originally only needs N/4 circulation just can finish.Improved operation efficiency greatly.
The row inverse transformation operational method as shown in Figure 6, will go inverse transformation and row inverse transformation the operation parallel processing.The row inverse transformation need just not begin to carry out after all rank transformations are finished, as long as those required several column data of row inverse transformation are finished the row inverse transformation, and can capable inverse transformation computing to all line data of these several row.Like this, the cycle index of two dimensional image one deck inverse wavelet transform has become present M * N/16 from the M * N in past.
Real data with an image is an example below, further specifies the present invention, and with two dimensional image size 16 * 16, carrying out one deck integer biorthogonal Daubechies (5,3) wavelet transformation is example, and raw image data is X:
Traditional method is earlier each line data to be carried out small echo direct transform (row filtering), handles delegation at every turn, and each line translation needs 16/2 circulation, then through after 16 * 16/2 circulations, obtains the Y0 as a result of row filtering:
Again each column data is carried out small echo direct transform (row filtering), handle row at every turn, each rank transformation needs 16/2 circulation, then through after 16 * 16/2 circulations, obtains the Y as a result of one deck small echo direct transform.
One deck wavelet transformation with traditional algorithm calculating 16 * 16 raw data needs 16+16=32 systemic circulation, wherein comprises partial circulating 16/2=8 time in each systemic circulation, so comprise cycle calculations altogether 32 * 8=256 time.
The present invention then is with ranks filtering parallel processing, raw data with 16 * 16 is divided into 16/4=4 systemic circulation, wherein comprise the processing of 4 line data in each systemic circulation, the direct transform of each row of data needs partial circulating 16/4=4 time, finish the direct transform of one deck small echo, only need cycle calculations 4 * 4 * 4=64 time.
Wherein, finish the small echo direct transform of preceding 4 row (0~3) data of original image, be put into the relevant position of matrix of consequence, obtain X1 as a result through after 1 systemic circulation:
Wherein, finish the small echo direct transform of original image 4~7 line data, be put into the relevant position of matrix of consequence, obtain X2 as a result through after 2 systemic circulations:
Wherein, finish the small echo direct transform of original image 8~11 line data, be put into the relevant position of matrix of consequence, obtain X3 as a result through after 3 systemic circulations:
Claims (16)
1, a kind of integer wavelet direct transform method based on digital signal processor is characterized in that comprising the steps:
(1) view data is divided into several data segments;
(2) to the capable direct transform of line data of current data section;
(3) column data to the current data section carries out the row direct transform;
(4) to other data segment repeating steps (2), (3).
2, the method for claim 1 is characterized in that described step (1) comprises the steps:
(1A) view data is divided into a plurality of systemic circulations, the number of times of systemic circulation can be divided exactly by the line number of view data;
(1B) each row of data that comprises in described each systemic circulation is divided into a plurality of partial circulatings, the number of times of partial circulating can be divided exactly by the columns of described view data.
3, the method for claim 1 is characterized in that described step (2) is that the line data that will comprise in the current systemic circulation is done capable direct transform respectively; Step (3) is that the column data that will comprise in the current systemic circulation is done the row direct transform respectively.
4, the method for claim 1 is characterized in that described step (2) comprises the steps:
The current line data are done the Lazy conversion, the coefficient of parity column wherein is placed apart;
Carry out one-dimensional wavelet transform, wherein one-dimensional wavelet transform comprises odd-transform and even number conversion again.
5, the method for claim 1 is characterized in that described step (3) comprises the steps:
Data to the same row of each row are done the lazy conversion, and the coefficient of parity rows wherein is placed apart;
Carry out one-dimensional wavelet transform, wherein one-dimensional wavelet transform comprises odd-transform and even number conversion again.
6, the method for claim 1 is characterized in that if desired wavelet coefficient being quantized in the described small echo direct transform process, then is 2 power by default quantization parameter, finishes quantification in the process of wavelet transformation simultaneously.
7, the method for claim 1 is characterized in that the wavelet coefficient after the described small echo direct transform adopts a coefficient of dilatation k to be optimized, k=1.2~1.5, and it can be transformed into multiplication of integers and shift operation.
8, the method for claim 1, it is characterized in that described small echo direct transform finishes in the sheet internal buffer, and to the data before and after the conversion in sheet and the carrying between the sheet external memory district, the ranks transposition of Lazy conversion and matrix is to use the quick direct memory visit of digital signal processor to finish.
9, a kind of integer wavelet inverse transform method based on digital signal processor is characterized in that comprising the steps:
(1) view data is divided into several data segments;
(2) column data to the current data section carries out the row inverse transformation;
(3) to the capable inverse transformation of line data of current data section;
(4) to other data segment repeating steps (2), (3).
10, method as claimed in claim 9 is characterized in that described step (1) comprises the steps:
(1A) view data is divided into a plurality of systemic circulations, the number of times of systemic circulation can be divided exactly by the columns of view data;
(1B) the every column data that comprises in described each systemic circulation is divided into a plurality of partial circulatings, the number of times of partial circulating can be divided exactly by the line number of described view data.
11, method as claimed in claim 9 is characterized in that described step (2) is that the column data that will comprise in the current systemic circulation is done the row inverse transformation respectively; Step (3) is that the line data that will comprise in the current systemic circulation is done capable inverse transformation respectively.
12, method as claimed in claim 9 is characterized in that described step (2) comprises the steps:
Each column data is done the one dimension inverse wavelet transform, and wherein the one dimension inverse wavelet transform comprises odd number inverse transformation and even number inverse transformation again;
To carry out anti-Lazy conversion through the coefficient after the inverse wavelet transform, the coefficient of parity column will wherein be placed at interval.
13, method as claimed in claim 9 is characterized in that described step (3) comprises the steps:
Each row is done the one dimension inverse wavelet transform with data line, and wherein the one dimension inverse wavelet transform comprises odd number inverse transformation and even number inverse transformation again;
To carry out anti-Lazy conversion through the coefficient after the inverse wavelet transform, the coefficient of parity column will wherein be placed at interval.
14, method as claimed in claim 9 is characterized in that if desired wavelet coefficient being carried out inverse quantization in described inverse wavelet transform process, then dequantized coefficients is preset as 1/ (2 powers), finishes inverse quantization in the process of inverse wavelet transform simultaneously.
15, method as claimed in claim 9 is characterized in that the wavelet coefficient behind the described inverse wavelet transform adopts a coefficient of dilatation k to be optimized, k=1.2~1.5, and it can be transformed into multiplication of integers and shift operation.
16, method as claimed in claim 9, it is characterized in that described inverse wavelet transform finishes in the sheet internal buffer, and to the data before and after the conversion in sheet and the carrying between the sheet external memory district, the ranks transposition of anti-Lazy conversion and matrix is to use the quick direct memory visit of digital signal processor to finish.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 03148079 CN1266650C (en) | 2003-06-30 | 2003-06-30 | Integer wavelet conversion method based on digital signal processor |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 03148079 CN1266650C (en) | 2003-06-30 | 2003-06-30 | Integer wavelet conversion method based on digital signal processor |
Publications (2)
Publication Number | Publication Date |
---|---|
CN1460968A true CN1460968A (en) | 2003-12-10 |
CN1266650C CN1266650C (en) | 2006-07-26 |
Family
ID=29591411
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 03148079 Expired - Lifetime CN1266650C (en) | 2003-06-30 | 2003-06-30 | Integer wavelet conversion method based on digital signal processor |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1266650C (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101996411A (en) * | 2009-08-21 | 2011-03-30 | 中国人民解放军国防科学技术大学 | Fast direction wavelet transforming method for image compression |
CN104202609A (en) * | 2014-09-25 | 2014-12-10 | 深圳市云朗网络科技有限公司 | Video coding method and video decoding method |
CN104519366A (en) * | 2014-12-04 | 2015-04-15 | 广东中星电子有限公司 | Video coding transforming and quantifying method and device |
-
2003
- 2003-06-30 CN CN 03148079 patent/CN1266650C/en not_active Expired - Lifetime
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101996411A (en) * | 2009-08-21 | 2011-03-30 | 中国人民解放军国防科学技术大学 | Fast direction wavelet transforming method for image compression |
CN101996411B (en) * | 2009-08-21 | 2013-08-21 | 中国人民解放军国防科学技术大学 | Fast direction wavelet transforming method for image compression |
CN104202609A (en) * | 2014-09-25 | 2014-12-10 | 深圳市云朗网络科技有限公司 | Video coding method and video decoding method |
CN104519366A (en) * | 2014-12-04 | 2015-04-15 | 广东中星电子有限公司 | Video coding transforming and quantifying method and device |
CN104519366B (en) * | 2014-12-04 | 2019-05-14 | 广东中星微电子有限公司 | A kind of Video coding change quantization method and apparatus |
Also Published As
Publication number | Publication date |
---|---|
CN1266650C (en) | 2006-07-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN1222110C (en) | Self adaptive variable length decoding method for image data | |
CN100348045C (en) | Resolution changing method and apparatus | |
CN1260971C (en) | Image coding appliance and image coding method | |
CN1419787A (en) | Quality based image compression | |
WO2007021615A2 (en) | Simd lapped transform-based digital media encoding/decoding | |
CN1431828A (en) | Optimum scanning method for change coefficient in coding/decoding image and video | |
CN1798341A (en) | Adaptive coefficient scan order | |
CN105847800B (en) | Method for compressing image and system based on all phase discrete sine biorthogonal conversion | |
CN1295696A (en) | Compressing and decompressing images | |
CN1565129A (en) | Coding and decoding method and apparatus using plural scanning patterns | |
CN1558682A (en) | Embedded image processor | |
CN1717049A (en) | Wavelet changeable VLSI structure based on line | |
CN1306826C (en) | Loop filter based on multistage parallel pipeline mode | |
CN1878308A (en) | Orthogonal transformation method for image and video compression | |
CN1914926A (en) | Moving picture encoding method and device, and moving picture decoding method and device | |
CN1266650C (en) | Integer wavelet conversion method based on digital signal processor | |
CN1230000C (en) | Scanning method of changing coefficient block in video frequency coding and decoding and its device | |
CN1665302A (en) | Image processing method and image processing device | |
CN1889690A (en) | Discrete cosine transforming method operated for image coding and video coding | |
CN1222171C (en) | Image transformation apparatus and method | |
CN1173574C (en) | Method for carrying out coding and decoding of wavelet video by digital signal processor | |
CN1874510A (en) | Integral discrete cosine transform method in use for encoding video | |
CN1445999A (en) | Method and device of integer transform applicable to image coding and video frequency coding | |
CN1533183A (en) | Dispatching licence transmitting method in radio communication system | |
CN1692625A (en) | Image conversion device, image conversion method, and recording medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CX01 | Expiry of patent term |
Granted publication date: 20060726 |
|
CX01 | Expiry of patent term |