CN104113763A - Optimized integer transform radix applied to image coding - Google Patents
Optimized integer transform radix applied to image coding Download PDFInfo
- Publication number
- CN104113763A CN104113763A CN201310138425.4A CN201310138425A CN104113763A CN 104113763 A CN104113763 A CN 104113763A CN 201310138425 A CN201310138425 A CN 201310138425A CN 104113763 A CN104113763 A CN 104113763A
- Authority
- CN
- China
- Prior art keywords
- transform
- integer
- matrix
- image coding
- dct
- 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 Or Coding Systems Of Tv Signals (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention discloses an optimized integer transform radix applied to image coding. Integer DCT matrix filling rules are improved, so that a type of new integer transform radixes is obtained and, the integer transform radixes are named as variety integer DCT radixes. A kind of optimized integer transform radix is obtained through searching a large number of variety integer DCT radixes. The optimized integer transform radix has the following characteristics that: matrix elements are integers; the optimized integer transform radix satisfies an orthogonal constraint condition; a normalized orthogonal matrix and the original DCT radixes have similar decorrelation performance; one-dimensional transform has a complete butterfly fast algorithm; only addition and shifting operation is required in a transform process, and multiplication does not exist, and therefore, the accuracy of transform operation is high; and forward and reverse transform are completely reversible, and no floating point drifts, and therefore, the transform operation is easy to realize through hardware. With the transform radix and transform radix transform processing method applied to video image coding adopted, the complexity of an encoder can be reduced.
Description
one, technical field
The present invention relates to image coding technology field, relate in particular to a kind of image conversion coding method (coding method of class Integer DCT Transform, or excellent integer transform method).
two, background technology
Image Coding refers to meeting under the condition of certain mass (requirement of signal to noise ratio or subjective assessment score), with the technology of institute's inclusion information in less bit number presentation video or image.
Discrete cosine transform (Discrete Cosine Transform) after being proposed by people such as N.Ahmed for 1974, owing to can greatly removing the correlation of pictorial element at transform domain, be considered to the accurate optimal mapping that performance approaches optimal transformation Karhunen-Loeve transformation, be widely used at image and video compression coding field.Conventional dct transform matrix represents by floating number, more multiplying and add operation are adopted, and multiplying needs to consume more system resource in general-purpose computations chip, take more operation time, under limited word length condition, there is truncated error in floating-point operation, and precision is not high, there is drift in Code And Decode data, can cause and not mate phenomenon.Integer DCT Transform adopts integer to replace conventional floating number to represent transformation matrix.The core of conversion is integer transform, calculates without floating number, and accuracy is high; Kernel kernal mapping only needs simple addition and shift operation to complete, and has reduced computation complexity; There is not trueness error in integer arithmetic, ensure the invertibity of coding, solved the matching problem between the positive inverse transformation of encoder and decoder, therefore audio frequency and video compression standard of new generation H.264/AVC, China national audio/video encoding standard AVS etc., standard has all adopted Integer DCT Transform.
The selection of integer translation base plays conclusive effect to the performance of orthogonal transform.Because integer transform basic matrix is not unique, Integer DCT Transform base is one of numerous integer translation bases, and selecting transform-based criterion is that its normalization coefficient matrix and dct transform matrix are approaching as far as possible, to reach best decorrelation performance.And the present invention " class Integer DCT Transform " selection criterion of base is: decorrelation characteristic and the DCT characteristic of transform-based are close as far as possible, and do not pursue its normalization coefficient matrix and DCT matrix approaches, and have the object of complete dish-shaped fast algorithm to reach transform operation.
three, summary of the invention
The object of example of the present invention is to propose a kind of preferred integer translation base that is applied to Image Coding, be intended to solve two aspect problems below: (1) existing dct transform matrixing adopts floating-point multiplication and addition, take hardware resource many, computational accuracy is not high, the positive inverse transformation of DCT can not be mated completely, converts irreversible problem; (2) DCT, integer DCT are without complete fast algorithm, and hardware implementation complexity is high.Preferably the acquisition of integer translation base is that complexity, the raising performance that further reduces image encoder has huge practical value.
Method of the invention process is achieved in that integer translation base builds matrix
hfor:
It is characterized in that: wherein
a 0 ,
a 1 ,
a 2 ,
a 3 ,
a 4 ,
a 5 ,
a 6 be 1 ~ 15 integer, excellent integer transform base element
a 0 ,
a 1 ,
a 2 ,
a 3 ,
a 4 ,
a 5 ,
a 6 value be { Isosorbide-5-Nitrae, 3,2,5,6,8}.Described matrix
hmust meet orthogonality,
for diagonal matrix, that is:
In formula
, orthogonality requirement
,
.
Preferably integer translation base is for receiving 8 × 8 Image Residual data of Image Coding prediction module output
x , carrying out transition coding processing, the matrix relationship of the positive inverse transformation of its two dimension is expressed as:
Wherein
x for original input residual error data matrix,
y for transformation results,
g for
h normalization matrix,
for matrix dot multiplication, so integer transform is reversible.
For the preferred integer translation base transformation matrix of Image Coding
h middle all elements is integer, and conversion is integer arithmetic entirely, normalization dot product
computing can with quantize to merge, in conversion final stage processing, normalized can narrow down to suitable size by the conversion coefficient having amplified.
The preferred integer translation base of Image Coding, is characterized in that, two-dimentional class Integer DCT Transform can be decomposed into one-dimensional transform twice.One dimension class integer DCT direct transform matrix expression is
, wherein
x, Xfor a dimensional vector before and after conversion, be launched into direct transform evaluator and be expressed as:
Wherein
with
,
with
,
with
,
with
between these four groups of data, form butterfly computation relation, computing obtains four groups of new data
with
,
with
,
with
,
with
, between four groups of new data, can again utilize butterfly computation, its result is multiplied by respectively different coefficients, obtains last transformation results through applying butterfly computation for the third time.Its computing flow process as shown in Figure 1, has complete butterfly fast algorithm.
four, brief description of the drawings
Fig. 1 is the flow chart of preferred integer transform fast algorithm of the present invention;
Fig. 2 is the application example of the present invention in video encoder flow chart.
five, embodiment
In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and embodiment, the present invention is further elaborated, for convenience of explanation, only show the part relevant to the invention process example.Should be appreciated that the concrete embodiment that this place is described, only for explaining the present invention, not in order to limit the present invention.
The present invention proposes the integer transform that a class is new and (is called: " class integer DCT " conversion, or preferred integer transform) and fast algorithm, the decorrelation performance of preferred class integer translation base orthonormalization matrix and dct transform is close, alternative Integer DCT Transform role in Image Coding.
case study on implementation
The MPEG4 video encoder Xvidcore that increases income is the coding and decoding video of telotism, comprise infra-frame prediction, estimation and inter prediction, dct transform and the entropy several parts composition of encoding, encoder input is the video sequence of QCIF, and output is the video file of MPEG4 form.Input video two field picture in piecemeal, frame/inter prediction, form 8 × 8 residual error data pieces and send into conversion, quantization modules, finally by the harmless entropy output squeezing video data of encoding, detailed process is shown in accompanying drawing 2.In order to analyze preferred integer transform performance, positive the DCT in encoder inverse transformation operator source code is changed to the positive inverse operator of preferred integer transform of the present invention, and insert Performance Evaluation code, the decorrelation efficiency of calculation code device,
pSNR, the isoparametric population mean of compression ratio, the modules such as dct transform in accompanying drawing 2, decorrelation efficiency estimation, quantification, inverse quantization, inverse dct transform are seen in concrete insertion position.
Statistic property parameter calculation formula and being described as follows:
y-PSNR:
,
Wherein
,
f(
i,
j) be input picture,
goriginal image is gone back in (i, j) decoding,
compression ratio meter: ,
Wherein
yUV_LEN isinput original YUV ASCII stream file ASCII length,
mPEG4_LEN isoutput MPEG4 ASCII stream file ASCII length
decorrelation efficiency:
If
be 8
8 input residual matrixes,
for converting rear coefficient matrix,
scan by " Z " word the dimensional vector obtaining
,
,
n =64, and column vector is carried out to periodicity and expand
,
yeach rank auto-correlation coefficient
computing formula is:
The auto-correlation extent index of output sequence
computing formula is:
To given input residual error data
x, the decorrelation efficiency of transform operation is defined as:
standard dct transform matrix
pfill method
:
Wherein coefficient
, , , , , , value can be calculated by cosine function, wherein
,
,
,
,
,
,
,
integer DCT direct transform matrix
ifill method
:
Wherein Integer DCT Transform matrix can use integer quotient
,
,
,
,
,
,
vectorial unique expression,
Table 1 is standard DCT, integer DCT, preferably integer translation base is filled element vector table.
Table 2, table 3 have provided each transform-based and have been applied to the Performance Evaluation result that MPEG4 encodes.
Can find out from table 2, table 3, preferably integer transform is applied to MPEG4 coding, and the deviation of the performance index such as its compression ratio, Y-PSNR, decorrelation efficiency and standard dct transform is less than 5%, suitable with integer DCT base.
Before accompanying drawing 1 fast algorithm, two-stage is without multiplication, the last third level needs multiplication of integers operation, can adopt the summation that is shifted by turn to replace multiplication, realize 3 shifting functions of maximum needs, the 3 sub-addition operations of 4Bit × 10Bit parallel multiplier, if the complexity of 1 sub-addition or shifting function is 1, can be with reference to the accompanying drawings 1 to obtain preferred integer transform fast algorithm overall complexity be 66.And the overall complexity of Integer DCT Transform fast algorithm is 148.So the preferred integer transform of the present invention is realized and can be saved 55% gate resource with hardware.
Claims (5)
1. an integer translation base that is applied to Image Coding, is characterized in that: transform-based matrix is 8 × 8 INTEGER MATRICES
h , matrix element fill method is:
Wherein
a 0 ,
a 1 ,
a 2 ,
a 3 ,
a 4 ,
a 5 ,
a 6 be 1 ~ 15 integer, optimum integer transform base element
a 0 ,
a 1 ,
a 2 ,
a 3 ,
a 4 ,
a 5 ,
a 6 value is { Isosorbide-5-Nitrae, 3,2,5,6,8}.
2. the Image Coding integer transform basic matrix as described in claim 1
h ,it is characterized in that
h must meet orthogonality, i.e. requirement
hH t for diagonal matrix:
In formula
k=4 (
a 1 a 5 -
a 2 a 6 ), for meeting orthogonality, must meet
k=0,
a 1 a 5 =
a 2 a 6 .
3. integer translation base as described in claim 1, is characterized in that, for receiving 8 × 8 Image Residual data of Image Coding prediction module output
x , for transition coding processing, the matrix relationship of the positive inverse transformation of its two dimension is expressed as:
Wherein
x for original input residual error data matrix,
y for transformation results,
g for
h normalization matrix,
for matrix dot computing, so integer transform is reversible.
4. the integer translation base of the Image Coding as described in claim 1, is characterized in that transformation matrix
h middle all elements is integer, and conversion be full integer arithmetic, normalization with quantize to merge, converting final stage processing, normalized can narrow down to suitable size by the conversion coefficient having amplified.
5. the Image Coding integer translation base as described in claim 1, is characterized in that, two-dimentional class Integer DCT Transform can be decomposed into one-dimensional transform twice, and one dimension class integer DCT direct transform matrix expression is
x=
hx, wherein
x, Xfor a dimensional vector before and after conversion, be launched into evaluator and be expressed as:
Wherein
x 0 with
x 7 , x 1 with
x 6 , x 2 with
x 5 , x 3 with
x 4 ,between these four groups of data, form butterfly computation relation, computing obtains four groups of new data
x 0 +
x 7 with
x 0 -x 7 ,
x 1 +
x 6 with
x 1 -x 6 ,
x 2 +
x 5 with
x 2 -x 5 ,
x 3 +
x 4 with
x 3 -x 4 , between four groups of new data, can again utilize butterfly computation, its result is multiplied by respectively different coefficients, obtains last transformation results through applying butterfly computation for the third time, and its operational flowchart provides in Figure of description 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310138425.4A CN104113763A (en) | 2013-04-22 | 2013-04-22 | Optimized integer transform radix applied to image coding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310138425.4A CN104113763A (en) | 2013-04-22 | 2013-04-22 | Optimized integer transform radix applied to image coding |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104113763A true CN104113763A (en) | 2014-10-22 |
Family
ID=51710378
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310138425.4A Pending CN104113763A (en) | 2013-04-22 | 2013-04-22 | Optimized integer transform radix applied to image coding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104113763A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106488235A (en) * | 2015-09-01 | 2017-03-08 | 北京君正集成电路股份有限公司 | A kind of SSE simplified calculation method for rate-distortion optimization and device |
| CN115643404A (en) * | 2022-11-16 | 2023-01-24 | 江西锦路科技开发有限公司 | Image processing method, device and system based on hybrid deep learning |
-
2013
- 2013-04-22 CN CN201310138425.4A patent/CN104113763A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106488235A (en) * | 2015-09-01 | 2017-03-08 | 北京君正集成电路股份有限公司 | A kind of SSE simplified calculation method for rate-distortion optimization and device |
| CN115643404A (en) * | 2022-11-16 | 2023-01-24 | 江西锦路科技开发有限公司 | Image processing method, device and system based on hybrid deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR101013344B1 (en) | Reduction of errors during computation of discrete cosine transform | |
| KR101028676B1 (en) | Transforms with common factors | |
| US8170334B2 (en) | Image processing systems employing image compression and accelerated image decompression | |
| US7379500B2 (en) | Low-complexity 2-power transform for image/video compression | |
| US8170333B2 (en) | Image processing systems employing image compression | |
| US20230419554A1 (en) | Point cloud attribute encoding method, decoding method, encoding device, and decoding device | |
| CN102036075B (en) | Image and digital video coding and decoding methods | |
| WO2019004749A1 (en) | Method and apparatus for performing low complexity computation in transform kernel for video compression | |
| CN102572426B (en) | Method and apparatus for data processing | |
| CN105163130B (en) | A kind of Lossless Image Compression Algorithm method based on discrete Tchebichef orthogonal polynomial | |
| Rojatkar et al. | Image compression techniques: lossy and lossless | |
| Kadhim | Image compression using discrete cosine transform method | |
| US8170335B2 (en) | Image processing systems employing image compression and accelerated decompression | |
| CN104113763A (en) | Optimized integer transform radix applied to image coding | |
| US9584833B2 (en) | Method for coding pulse vectors using statistical properties | |
| US6766341B1 (en) | Faster transforms using scaled terms | |
| De Silva et al. | Exploring the Implementation of JPEG Compression on FPGA | |
| US20240233194A1 (en) | Point Cloud Attribute Encoding Method and Apparatus, Decoding Method and Apparatus, and Related Device | |
| More et al. | JPEG picture compression using Discrete Cosine Transform | |
| CN105472395B (en) | A kind of Lossless Image Compression Algorithm method based on discrete Krawtchouk orthogonal polynomial | |
| US20160049159A1 (en) | Method for coding pulse vectors using statistical properties | |
| CN206962992U (en) | 3 for digital video decoding multiply 3 Integer DCT Transform quantizers | |
| Li et al. | A video coding algorithm using vector-based techniques | |
| Krishna et al. | Incorporation of DCT and MSVQ to Enhance Image Compression Ratio of an image | |
| JP6102707B2 (en) | Digital encoding device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20141022 |