CN1154193A - Method and apparatus for reduction of image data compression noise - Google Patents

Abstract

An image encoding module and an image decoding module provide a method of performing transform coding of an original image source, where the method serves to suppress or eliminate blocking artifacts. Encoded data are transmitted by the image encoding module, converted into received data, and transformed into modified data by means of a filtering operation utilizing quantization error terms. The quantization error terms can be derived from quantization error data generated in the image encoding module, or can be provided as a look-up table in an alternative image decoding module. The modified data are converted into reduced-noise image data for reconstruction into a digital image.

Description

Be used to reduce the method and apparatus of image data compression noise

Invention field

The present invention relates to image processing, relate in particular to a kind of method and apparatus that is used to reduce block artefact (blocking artifact), wherein block artefact is caused by image data compression noise, and image data compression noise is produced by the error that quantizes in the conversion picture coding course to cause.

Background of invention

The image processing procedure of conversion picture coding is used in various application, it comprises the electronic switch of photographic image (photographic image), in printing process rendering graphical information, and by electronic communication system transmission of digital pictorial data.In these are used, the original source image (original source image) that provides as a series of electric image signals (electricalimage signal) is encoded, and wherein each signal is all corresponding with the feature of a certain element of original source image or pixel.The electric image conversion of signals is become a two Dimension Numerical Value group of representing the source image pixel.For each colour band that is used (color-band) is equipped with one group of independently numerical value.For example, in ' yuv ' configuration, provide three groups of numerical value, and as known in the art, before further processing, u-group and v-group have been carried out down-sampling (downsample).Usually these numerical value are built into a two-dimentional H * V array of being made up of image data item.So each in the image data array is just corresponding to a certain specific pixel in the original source image, and the characteristic of this pixel is described quantitatively.For example, in a standard display format, the digital source image comprises the image data array of being made up of 640 row and 480 line number values.The conversion encoding of graphs produces one group of new treated H * V numerical value, and they are commonly called the image reconstruction data item, calculate according to the digital source image.Convert the numerical value after handling to a series of new signals of telecommunication, and produce treated digital image by these signals of telecommunication.

The conversion picture coding adopts the orthogonal transform of discrete cosine transform (discrete cosine transform DCT) and so on, converts image data item to the coefficient of frequency item, calculates thereby simplified follow-up processing.For example, the cosine transform picture coding is a kind of like this image processing procedure, in this process, digital source image is carried out two-dimentional forward (forward) discrete cosine transform (FDCT), remove the result of gained then with the item in the quantization table, and use such as processes such as Huffman codings it is carried out entropy coding (entropy encode).Then, storage or transmission coded data, this is more much effective than adopting original figure source image item usually.Subsequently, coded data is deciphered, multiplied each other with it with quantizing item, and adopt inverse cosine conversion (IDCT) to be converted into the image data item of reconstruction.Next, obtain a treated digital image by the image data item of rebuilding.

The orthogonal transform of using in the conversion picture coding course is FDCT and IDCT normally.These conversion are to carry out according to the industrial standard that joint photographic experts group (JPEG) is set up.Description among " JPEG still image data compression standard (the JPEG Still Image DataCompression Standard) " of " ISO draft internation standard (ISO Draft International Standard) 10918-1 " see reference document William B.Pennebaker and Joan L.Mitchell of JPEG in the appendix A.According to Joint Photographic Experts Group, the digitlization source image that provides is a series of image-data matrixes, be formatted as 8 * 8 matrixes usually, and FDCT is used for producing a series of matrix of frequency coefficients.

With two-dimentional FDCT the process that image-data matrix converts frequency-coefficient matrix to can be explained with the matrix notation of following simplification:

S(ν，μ)＝C×s(j，i)×C ^T

Wherein, S (ν μ) is item in the matrix of frequency coefficients, s (j i) is item in image-data matrix, and C is the basis matrix of discrete cosine transform, and C ^TIt is the transposed matrix of C.When being applied to 8 * 8 images-data matrix, two-dimentional FDCT can be provided by following equation: $S(\mathrm{\ν},\mathrm{\μ})=\frac{1}{4}{C}_{\mathrm{\ν}}{C}_{\mathrm{\μ}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}s(j,i)\cos \[(2i+1)\frac{\mathrm{\μ\π}}{16}\]\cos \[(2j+1)\frac{\mathrm{\ν\π}}{16}\]$

Wherein, for k=0, C _k=1/ √ 2; And for k＞0, C _k=1.Generally send and storage before, be with the coefficient of frequency item divided by quantizing and the merchant of gained being rounded up.Merchant who calculates and the difference that rounds up between the gained merchant are the errors of calculation that computing produced that rounds up.This can cause the graphics data compression noise, thereby produces block artefact medium to the height compression level.

The image data item that adopts two-dimentional IDCT computing to obtain to rebuild, can state as with the matrix notation of simplifying:

r(j，i)＝C ^T×R(ν，μ)×C

Wherein, r (j i) is the image data item of rebuilding, and R (ν is to remove the coefficient of frequency item that quantizes μ).When being applied to 8 * 8 matrix of frequency coefficients, corresponding two-dimentional IDCT can be provided by following equation: $r(j,i)=\frac{1}{4}\underset{\mathrm{\ν}=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{\μ}=0}{\overset{7}{\mathrm{\Σ}}}{C}_{\mathrm{\ν}}{C}_{\mathrm{\μ}}R(\mathrm{\ν},\mathrm{\μ})\cos \[(2i+1)\frac{\mathrm{\μ\π}}{16}\]\cos \[(2j+1)\frac{\mathrm{\ν\π}}{16}\]$

The method of the block artefact problem of various solutions has been described in relevant field.The United States Patent (USP) that is presented to H.Malvar has disclosed a kind of method and apparatus that is used to handle n-dimension word signal the 4th, 754, No. 492, and wherein n-dimension word signal contains two adjacent digital sampling values at least.The equipment that list of references discloses comprises a composite space operator (composite spatial operator), and the basic function of its use is similar with conventional DCT/IDCT basic function, but it is characterized in that it extends in the piece adjacent in the input signal slightly.

The United States Patent (USP) that is presented to people such as Downing has disclosed a kind of method that image is carried out digital processing for the 5th, 220, No. 616, in the method, corresponding image array is divided into some, subsequently each piece is carried out conversion, thus compressed image.With the one dimension operator matrix boundary pictorial data element is scanned successively, make the discontinuous place on border level and smooth.

Be presented to the United States Patent (USP) the 5th of Yamaoka, 357, disclosed a kind of equipment that is used to compress the expansion image for No. 584, this equipment comprises one and estimates circuit, it estimates and provides a compressed coefficient of optimizing by comparing original image data piece and treated pictorial data to the predetermined compressed coefficient.This comprises each image data block is carried out the device of this comparison in pixel ground one by one with reference to estimating circuit, so that block noise data (blocknoise data) are provided.

The United States Patent (USP) the 5th that is being presented to Yuan, 367, in No. 385, disclosed a kind of method and apparatus that the block encoding pictorial data is handled, wherein carry out the pixel correction, with the numerical difference between between near ' external world ' pixel near the block edge in ' this locality ' pixel and the adjacent block reconstructed pixels block edge that reduces to select by low-pass filtering.

The United States Patent (USP) that is presented to Jeong has disclosed a kind of method and apparatus of eliminating block artefact the 5th, 384, No. 849, wherein comprises a block artefact measuring appliance in code device.Block artefact measuring appliance receives the original frame data that is delayed, and produces a frame data error, this error poor corresponding between original frame data and the frame data that receive the back recovery.

In the prior art, do not disclose by corrected received to frequency component reduce or eliminate the conversion picture coding course of block artefact.Therefore, the purpose of this invention is to provide a kind of equipment that carries out picture coding, the method for its use is by the coefficient of frequency filtering (filter) that receives is reduced block artefact, thus the compensation influence that quantization error produced.

Another object of the present invention is that a kind of method and apparatus that can be used to reduce block artefact in the image data files that meets the JEPG requirement that receives is provided.

Summary of the invention

The present invention is a kind of image coding/decoding equipment that carries out transition coding, and the method for its use can suppress or eliminate block artefact.After finishing the conversion picture coding, send coded data and convert thereof into the reception pictorial data, will receive the pictorial data lapped transform subsequently and become coefficient of frequency, so that revise by the filtering operation that adopts the quantization error matrix.The quantization error matrix can be derived by the quantization error data that produce in the code device, is perhaps provided by the look-up table in the decoding device.Coefficient of frequency through revising is converted into the image data item that noise reduces, and is used for the reconstructing digital image.

The accompanying drawing summary

Describe below and characterize novel characteristics of the present invention.In conjunction with the accompanying drawings referring to the embodiment of the following description of the present invention, the reader will be better understood mechanism of the present invention and method with and other objects and advantages.Wherein,

Fig. 1 is the schematic diagram of conventional image processing system, and this system adopts conversion picture coding and decoding that pictorial data is compressed;

Fig. 2 is a functional-block diagram, shows the process that with the equipment that comprises picture coding module and decoder module the original source picture inversion is become the image reconstruction of noise minimizing according to the present invention;

Fig. 3 is a block diagram, shows the performed calculation step of picture coding module among Fig. 2;

Fig. 4 is a block diagram, shows the performed calculation step of image decoding module among Fig. 2; And

Fig. 5 is another embodiment of image decoding module shown in Figure 4.

Detailed description of the present invention

Fig. 1 shows the image processing system that carries out the routine of Data Compression of Image by conversion picture coding and decoding.The original source picture inversion is become to comprise the digital source image of image data item with the digital imagery transducer (digtizer) of optical scanner 11 or video camera 13 and so on.Image data item both can have been used image-data diskette 15 or similar storage medium stores, also can directly send to conversion image coding device 21, so that convert compressed image data to.Subsequently, conversion image coding device 21 is reconfigured to a digital image with compressed image data, and sends it to such as image autput devices such as monitor 23 or printing equipments 25.Another kind method is, compressed image data is stored in the conversion image coding device 21, and perhaps filing is kept on the packed data floppy disk 27, perhaps sends to remote image processing system, thereby carries out image reconstruction.

According to the present invention, picture coding module 35 shown in Figure 2 and image decoding module 37 realize the Data Compression of Image and the decoding function of conventional conversion image coding device.Digital imagery transducer 10 converts original source image 31 to one or more groups colour band data, and every group of colour band data all independently are identified as a digital source image 33.Digital source image 33 is offered picture coding module 35, as one group of two dimension H * V image data item.Picture coding module 35 is transformed into compressed image data with digital source image 33, and sends it to image decoding module 37 by transmission medium 19 (it can be electric network or other communication system).Image decoding module 37 receives compressed image data, and compressed image data is transformed into the digital image 39 that noise reduces.Image autput device 20 converts low noise digital image 39 in the low noise image 41 of reconstruction.In some applications, before the compressed image data that picture coding module 35 is produced sends to image decoding module 37, its storage or filing can be kept at such as in the temporary storage medium such as disk 29.

The picture coding module

Hereinafter with reference to Fig. 3 the performed processing procedure of picture coding module 35 is described.Digital source image 33 comprises one group of H * V image data item, uses s ₀(z, y) expression.As known in the art, this picture group image data item is divided into a two-dimensional array by many N * N image-data matrix 51 formations.Each image-data matrix 51 comprises N ²Individual with s (j, i) Biao Shi item, 0≤i wherein, j≤N-1.Constitute and to be positioned at each image data item that graphics image data matrix array p is listed as the capable pictorial data matrix of q and to determine by following equation:

s _q，p(j，i)＝s ₀(j+Nq，i+Np)

Wherein, 0≤i+Np≤H-1, and 0≤j+Nq≤V-1.

In calculation step 61, adopt orthogonal transform that each image-data matrix 51 is transformed into one and comprise frequency-coefficient entry S _{Q, p}(conversion is based on following forward orthogonal transform formula for ν, N μ) * N frequency-coefficient matrix 53:

S _q，p(ν，μ)＝C×s _q，p(j，i)×C ^T

Wherein, 0≤μ, ν≤N-1, and C is the positive-going transition basis matrix.

In calculation step 62, with corresponding pro rata quantification (scaled quantization term) Q (ν, μ) each coefficient of frequency item S (ν in removal frequency-coefficient matrix 53, μ), a wherein pro rata quantification acquisition from a N * N quantization table 73, this quantization table 73 then is used in combination with the scale factor of representing with κ herein 71.The degree of the Data Compression of Image that the picture coding module is carried out is proportional to the numerical value of scale factor 71.The value of scale factor 71 also is used for determining the parameter obtain from the look-up table of filtering parameter group 75 _kAnd β _kNumerical value.For example, in a preferred embodiment, when given or when having determined the value of κ, Table I has just been determined image-data tape y, the parameter of u and v _kAnd β _kSuch as described in more detail below, filtering parameter group 75 is sent to image decoding module 37, is used to weaken quantizing noise.

The value of α and β during the given scale factor κ of Table I
				????κ	??α y，β y	????α u，β u	????α v，β v
???κ≤8	????0	????0	????0
				?8＜κ≤24	????1	????2	????2
24＜κ≤32	????2	????4	????4
				???κ＞32	????3	????6	????6

Carry out calculation step 62 and can produce, wherein be positioned at matrix array p and be listed as the capable merchant-coefficient matrix of q and comprise the merchant's item Qu that provides by following formula by the two-dimensional matrix array that N * N merchant-coefficient matrix 55 constitutes _{Q, p}(ν, μ): ${\mathrm{Qu}}_{q,p}(\mathrm{\ν},\mathrm{\μ})=\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}$

In calculation step 63, will discuss a Qu _{Q, p}(ν μ) rounds up, and becomes the lower numerical value of precision, thereby produces quantification-coefficient matrix 57, the quantification that this matrix comprises-merchant Qc _{Q, p}(ν μ) is provided by following expression formula: ${\mathrm{Qc}}_{q,p}(\mathrm{\ν},\mathrm{\μ})=\mathrm{round}\left({\mathrm{Qu}}_{q,p}(\mathrm{\ν},\mathrm{\μ})\right)=\mathrm{round}\left(\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}\right)$

In calculation step 65, derive N * N poor-coefficient matrix 77.Difference-coefficient matrix 77 comprises a difference Dc _{Q, p}(ν, μ), they are to go into business-each merchant Qu of coefficient matrix 55 according to following equation _{Q, p}(ν deducts the corresponding quantification-merchant Qc in quantification-coefficient matrix 57 in μ) _{Q, p}(ν μ) obtains: ${\mathrm{Qc}}_{q,p}(\mathrm{\ν},\mathrm{\μ})=\mathrm{abs}|\left(\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}\right)-\mathrm{round}\left(\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}\right)|$

Derive thus a series of poor-coefficient matrix 77, each matrix is corresponding to a frequency-coefficient matrix 53 in the matrix array.In calculation step 66, to each poor-coefficient matrix 77 summations, average, and multiply by a corresponding quantization item, thereby produce a N * N quantization error matrix 79, the item E that this matrix comprises ₀(ν, μ) can derive by following equation: $E_0(\mathrm{\ν},\mathrm{\μ})=Q(\mathrm{\ν},\mathrm{\μ})\underset{q}{\mathrm{\Σ}}\underset{p}{\mathrm{\Σ}}\frac{D{c}_{q,p}(\mathrm{\ν},\mathrm{\μ})}{M}$

The sum of poor-coefficient matrix that wherein M is added up.Another kind of situation is that quantization error matrix 79 comprises the item E that derives according to following equation ₁(ν, μ), these are root-mean-square error items: $E_1(\mathrm{\ν},\mathrm{\μ})=Q(\mathrm{\ν},\mathrm{\μ})\sqrt{\underset{q}{\mathrm{\Σ}}\underset{p}{\mathrm{\Σ}}\frac{{\left(D{c}_{q,p}(\mathrm{\ν},\mathrm{\μ})\right)}^2}{M}}$

In calculation step 64, by such as the Z-shaped serializing process of using the Huffman coding, each quantification-coefficient matrix 57 is sent coding, produce one group through sending coded data 59.Then, by transmission medium 19 this group is sent to image decoding module 37 through sending coded data 59.Similarly, by transmission medium 19 quantization error matrix 79, filtering parameter group 75 and quantization table 73 are sent to image decoding module 37.

The image decoding module

Hereinafter with reference to Fig. 4 the performed image processing procedure of image decoding module 37 is described.Data set 59 behind the transmission coding is decoded into one group uses Sr _{Q, p}(ν, μ) Biao Shi reception quantization transform coefficient.In calculation step 91,, each is received coefficient S r with corresponding pro rata quantification item in the quantization table 73 according to following equation _{Q, p}(ν μ) carries out mask and takes advantage of (mask multiple), to produce the conversion coefficient R that takes advantage of through mask _{Q, p}(ν, μ):

R _q，p(ν，μ)＝Sr _q，p(ν，μ)×Q(ν，μ)

Execution calculation step 91 can produce takes advantage of conversion coefficient R a series of comprising through mask _{Q, p}(ν, N μ) * N receives coefficient matrix 81, wherein each conversion coefficient R _{Q, p}(ν is μ) corresponding to a previous quantification-merchant Qc who handles in picture coding module 35 _{Q, p}(ν, μ).

In calculation step 92, according to following transformation by reciprocal direction formula, with a reverse orthogonal transform with conversion coefficient R _{Q, p}(ν μ) is transformed into by r _{Q, p}(j, i) Biao Shi reception image data item:

r _q，p(j，i)＝C ^T×R _q，p(ν，μ)×C

Each receives image data item r _{Q, p}(j is i) corresponding to an image data item s in H * V digital source image 33 ₀(z, y).Thus, can image data item r will be received _{Q, p}(j i) is formatted into one group and comprises s _R(z, y) Xiang H * V receives image-data 83.This formatting procedure can be finished according to following equation:

S _R(j+Nq，i+Np)＝r _q，p(j，i)

0≤i+Np≤H-1 wherein, and 0≤j+Nq≤V-1.

In conventional image decoding system, ensuing conventional calculation step 93 will rebuild and receive image data set 83, to form the reception digital image 43 that dotted line is represented.Receive digital image 43 just and to the ratio of height image compression, show block artefact medium usually.Block artefact causes that by quantizing merchant's item error produces in the calculation step 63 of picture coding module 35.A kind of method that is used for overcoming the quantizing noise problem is to reduce scale factor 71, thereby reduces the quantification to merchant, reduces the generation of block artefact.But, owing to the Data Compression of Image degree can reduce, and handle required computational resource of packed data and the corresponding increase of transmitting time meeting, so do not wish to use this method.

The present invention adopts a kind of trimming process that can compensate error that previous quantization operations causes to suppress or eliminate the generation of block artefact.According to the method that is disclosed, before the reconstructing digital image, constitute the s that receives image-data set 83 by a filtering correction _R(z, y) item.Filtering realizes according to following order of operation: i) in calculation step 94, will receive the s of image-data set 83 _R(z, y) overlapping-be converted into frequency domain, to form a two-dimensional array that constitutes by the frequency-coefficient matrix 85 of N * N through revising; Ii) in calculation step 95, to frequency-coefficient matrix 85 filtering, to produce a series of coefficient matrixes 87 through filtering through revising; Iii) in calculation step 96, will be through coefficient matrix 87 inverse transformations of filtering to spatial domain, to produce an image-data matrix 88 through filtering; Iv) in calculation step 98, from through the image-data matrix 88 of filtering, extracting the item that selected noise reduces, to form one group of bidimensional H * V low noise image-data 89; And v) in calculation step 99, form low noise digital image 39 by low noise image-data set 89.

In calculation step 94, carry out lapped transform and be meant the process that an array that is made of N * N matrix is carried out the forward orthogonal transform receiving image-data set 83, wherein this N * N matrix array obtains from image-data set 83 by ' overlapping ' process.Can obtain to constitute these image data item from receiving image-data matrix group 83 according to following equation through overlapping image-data matrix:

v _s，r(j，i)＝s _R(j+ωs，i+ωr)

Wherein, 1≤ω≤N-1,0≤j+ ω s≤V, and 0≤i+ ω r≤H.Constitute to be positioned at and be listed as capable each the image data item v of s through superimposed image-data matrix through superimposed image-data matrix array r _{S, r}(j, i) expression.The value of constant integer ω has been determined the lap that uses in the overlapping process, and it is the overlapping parameter 84 that image decoding module 37 is provided.In a preferred embodiment, ω is set to N/2.As can be seen, use overlapping process and can obtain such matrix array, in this matrix array, some row and column of image data item is the common through the superimposed image data matrix of corresponding pairs.For example,

v _{S+1, r}(j, i)=v _{S, r}(j+N-ω, i) and v _{S, r+1}(j, i)=v _{S, r}(j, i+N-ω)

For 0≤i, j≤ω-1.By overlapped image-data matrix is carried out filtering operation, avoided the discontinuity of matrix, and suppressed or eliminated the generation of block artefact to matrix.

To convert to through overlapping image-data matrix by the forward orthogonal transform and to comprise through frequency of amendment-coefficient entry Sv _{S, r}(ν, N * N μ) are through correction factor matrix 85, wherein through frequency of amendment-coefficient entry Sv _{S, r}(ν μ) is provided by following matrix form:

Sv _s，r(ν，μ)＝C×v _s，r(j，i)×C ^T

In calculation step 95, utilize 79 pairs in filtering parameter group 75 and quantization error matrix to carry out filtering and calculate through the correction factor matrix, to produce the coefficient matrix 87 of N * N through filtering, the coefficient matrix 87 of these warps comprises the frequency-coefficient entry Sf through filtering _{S, r}(ν, μ), they obtain according to following equation: ${\mathrm{Sf}}_{s,r}(\mathrm{\ν},\mathrm{\μ})=\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ})\]\×\[\frac{\[S{v}_{s,r}(v,\mathrm{\μ}){\]}^2}{\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2+\mathrm{\α}\[E_x(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\]}^{\mathrm{\β}}$

Can obtain the numerical value of α and β from filtering parameter group 75, and E _x(ν μ) is quantization error matrix 79, E for example as defined above ₀(ν, μ) and E ₁(ν, μ).

In calculation step 96, comprise through filtering vision data item rf by using inverse orthogonal transformation according to following matrix form, will being transformed into through the coefficient matrix 87 of filtering _{S, r}(j, i) through filtering vision-data matrix 88:

rf _s，r(j，i)＝C ^T×Sf _s，r(ν，μ)×C

Owing in calculation step 94, carried out overlapping process, so each image data item that is comprised through the image-data matrix 88 of filtering is all than obtaining required many of low noise digital image 39.Therefore, in calculation step 98, only extraction and merging are positioned at each N ' through filtering vision-data matrix * N ' core submatrix rk (m, n) Nei image data item, thus one group of bidimensional H * V low noise image-data 89 formed, wherein, 0≤m, n≤N '-1, and N '=N-ω, ω is overlapping parameter 84.

((N-1) image data item in capable is walked in the row of ω/2-1) and (N-ω/2), and corresponding to filtering vision data matrix rf the 0th to walk to the by deleting _{S, r}(j, i) (row of ω/2-1) and (N-ω/2) row image data item through filtering in being listed as to (N-1) can obtain each core submatrix rk to the 0th row to the _{S, r}(m, n).Core submatrix item rk _{S, r}(m is n) with the corresponding image data item rf through filtering in filtering vision-data matrix _{S, r}(j, i) relation between is provided by following formula: ${\mathrm{rk}}_{s,r}\{j-\left(\frac{\mathrm{\ω}}{2}\right),i-\left(\frac{\mathrm{\ω}}{2}\right)\}=r{f}_{s,r}(j,i)$

For (ω/2)≤i, and j≤N-(ω/2+1).

According to following equation,, just can form image-data set 89 that noise reduces by the kernel matrix item being merged into a bidimensional H * V array:

s _F(n+sN’，m+rN’)＝rk _s，r(n，m)

0≤m+rN '≤N-1 wherein, and 0≤n+sN '≤V-1.In calculation step 99, rebuild the low noise image-data set 89 corresponding, thereby form low noise digital image 39 with receiving image-data set 83.Compare with the reception digital image 43 that routine obtains, filtering operation 95 has reduced or has eliminated the influence of the block artefact that is produced because of quantification in low noise digital image 39.

Image decoding module-another embodiment

For example can export the module that meets JPEG requirement data for the conventional picture coding module of use provides the image processing that sends coded data groups 103 to use, and can adopt another kind of image decoding module 1O1 as shown in Figure 5 to realize decoding.Generally, conventional picture coding module provides a quantization table 105 by transmission medium 19, but quantization error matrix or filtering parameter are not provided.Therefore, in calculation step 95, image decoding module 101 is used as the quantization error matrix with quantization error table 109.Used filtering parameter is to utilize the decoding parametric table such as Table II to obtain α from filtering parameter group 75 ' in the calculation step 95 _kAnd β _kNumerical value provide.

The same with above-mentioned preferred embodiment, by calculation step 92 and 94, can convert a series of coefficient matrixes 85 to sending coded data groups 103 through revising.In calculation step 121, with the pro rata quantification of the DC in the quantization table 105 (that is the pro rata quantification Q (0,0) in first row and first row) conduct estimation scale factor 107 of κ ' expression.According to Table II, scale factor 107 is acted on filtering parameter group 75 ', thereby produce α _kAnd β _kNumerical value.

The value of α and β during the given estimation scale factor of table 3 κ '
				????κ’＝DC	????α y，β y	????α u，β u	????α v，β v
????κ’≤8	????0	????0	????0
				???8＜κ’≤24	????1	????2	????2
???24＜κ’≤32	????2	????4	????4
				????κ’＞32	????3	????6	????6

Also scale factor 107 is acted on quantification-errors table 109, the estimation of selecting the table that comprises N * N noise-mask item from quantification-errors table 109 quantizes error matrix E _ε(ν, μ).Quantization error matrix E _ε(ν, the noise-mask item that is comprised in μ) is derived with empirical data and method of estimation, and has considered the various image attributes that the image processing personnel generally should be noted that.The quantization error Table III to V be the example that this class is used for the experience derivation noise mask of 8 * 8 frequencies-coefficient matrix.

In calculation step 95, use and estimate to quantize error matrix E _ε(ν, the α that obtains μ) and from filtering parameter group 75 ' _kAnd β _kNumerical value, the coefficient matrix 85 through revising is carried out filtering operation, thereby produces coefficient matrix 87 through filtering.The same with in the preferred embodiment by carrying out calculation step 96,98 and 99, can obtain the digital image 38 that noise reduces.

Quantization error Table III scale factor=2; Y image-data tape

????0.0????0.6????0.3????0.3????0.6????0.8????1.0????1.0 ????0.6????0.6????0.3????0.6????0.6????0.8????1.1????1.2 ????0.6????0.6????0.6????0.6????0.8????1.0????1.2????1.2 ????0.6????0.6????0.6????0.6????0.8????1.1????1.2????1.1 ????0.6????0.6????0.6????0.8????0.9????1.0????1.2????1.0 ????0.6????0.6????0.8????1.1????1.1????1.1????1.0????0.8 ????0.8????0.8????0.9????1.1????1.1????1.0????0.8????0.7 ????0.8????0.9????1.1????1.1????0.9????0.9????0.7????0.6

Quantization error Table IV scale factor=8; U image-data tape

????0.0????2.2????2.2????2.6????2.7????2.2????1.7????1.5 ????2.2????2.2????2.2????2.3????2.3????1.8????1.4????1.4 ????2.5????2.2????2.3????2.5????2.1????1.6????1.3????1.2 ????2.7????2.5????2.6????2.3????1.7????1.4????1.2????1.1 ????3.1????2.6????2.2????1.8????1.5????1.3????1.1????1.1 ????2.7????2.1????1.7????1.4????1.3????1.1????1.1????1.0 ????2.1????1.8????1.5????1.2????1.1????1.0????1.0????1.0 ????1.7????1.5????1.3????1.1????1.0????1.0????1.0????1.0

Quantization error Table V scale factor=24; V image-data tape

????0.0????6.1????5.4????5.3????4.2????3.4????2.9????2.7 ????6.1????5.9????5.0????4.3????3.5????3.0????2.6????2.5 ????6.1????5.2????4.7????4.1????3.3????2.7????2.4????2.4 ????5.8????4.8????4.5????3.5????2.9????2.5????2.2????2.2 ????5.3????4.3????3.4????2.8????2.4????2.2????2.0????2.0 ????3.9????3.3????2.7????2.3????2.1????2.0????2.0????1.9 ????3.1????2.7????2.3????2.0????1.9????1.8????1.8????1.8 ????2.5????2.3????2.0????1.8????1.7????1.7????1.7????1.7

Advantage of the present invention is to be suitable for providing a kind of equipment that does not cause block artefact and realize medium Data Compression of Image, and its used method is more effective than the method for using in the present conventional image processing facility.Although described preferred embodiment of the present invention, but for those skilled in the art, obviously can not break away from the present invention and carry out various variations and change, and the applicant has attempted to comprise these variation and the change within connotation of the present invention and the scope in appended claim term.

Claims (51)

1. One kind in the conversion picture coding and compression process that the graphics picture signals are converted to digital image, be used to reduce because of quantizing to cause the method for block artefact influence, wherein compression degree is determined by a scale factor κ and a quantization table, picture intelligence provides as series of electrical signals, each signal of telecommunication is corresponding with the feature of certain element of bidimensional image, and each picture element is configured to a bidimensional H * V array, it is characterized in that, said method comprising the steps of:

   Convert described series of electrical signals to one group of numerical value, described each numerical value describe quantitatively a corresponding picture element feature, described numerical value s ₀(z, y) expression, 0≤y≤H-1 wherein, and 0≤z≤V-1;

   Described numerical value group is formatted into a plurality of N * N pictorial data matrixes with subscript ' p ' and ' q ' identification, and described each image-data matrix comprises by s _{Q, p}(described each image data item is made up of a described numerical value of determining according to following relation for j, i) Biao Shi image-data item:

   s _q，p(j，i)＝s ₀(j+Nq，i+Np)

   0≤i wherein, j≤N-1;

   Described each image-data matrix is transformed into a N * N frequency-coefficient matrix, and described frequency-coefficient matrix comprises by S _{Q, p}(the described step of described each the image-data matrix of conversion adopts orthogonal transform basis matrix C to realize according to following formula for ν, μ) Biao Shi frequency-coefficient entry:

   S _q，p(ν，μ)＝C×s _q，p(j，i)×C ^T；

   (ν μ) removes described each frequency-coefficient entry S with a pro rata quantification Q _{Q, p}(ν, μ), producing a plurality of N * N merchant-coefficient matrix, wherein said pro rata quantize be from quantization table, obtain and use the scale factor correction, described merchant-coefficient matrix comprises by Qu _{Q, p}(described merchant's item is derived according to following formula for ν, μ) Biao Shi merchant's item: ${\mathrm{Qu}}_{q,p}(\mathrm{\ν},\mathrm{\μ})=\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})};$

   Described each merchant item rounded up becomes the lower numerical value of precision, thereby produces a quantifications-coefficient matrix, and described quantification-coefficient matrix comprises by the definite Qc of following formula _{Q, p}(ν, μ) represented quantification-coefficient entry: ${\mathrm{Qc}}_{q,p}(\mathrm{\ν},\mathrm{\μ})=\mathrm{round}\left(\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}\right);$

   From each described merchant Qu _{Q, p}(ν deducts corresponding described quantification-coefficient entry Qc in μ) _{Q, p}(ν, μ), thereby form a plurality of N * N poor-coefficient matrix, described poor-coefficient matrix comprises a difference Dc who derives according to following formula _{Q, p}(ν, μ): ${\mathrm{Qc}}_{q,p}(\mathrm{\ν},\mathrm{\μ})=\mathrm{abs}|\left(\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}\right)-\mathrm{round}\left(\frac{S_{q,p}(\mathrm{\ν},\mathrm{\μ})}{Q(\mathrm{\ν},\mathrm{\μ})}\right)|;$

   By to error term summation and average, derive one and comprise E ₀(ν, μ) Xiang quantization error matrix,, described error term comprise described poor-function of coefficient matrix;

   Select one group of filtering parameter of representing with α and β;

   Described each quantization parameter item be multiply by a corresponding quantification item, and to produce the transform coefficient matrix that a plurality of N * N take advantage of through mask, the described transform coefficient matrix of taking advantage of through mask comprises the R that derives according to following formula _{Q, p}(ν, μ) represented item:

   R _q，p(ν，μ)＝Qc _q，p(ν，μ)×Q(ν，μ)；

   One reverse orthogonal transform is acted on the described transform coefficient matrix of taking advantage of through mask, to produce a plurality of r that use according to following transformation by reciprocal direction formula _{Q, p}(j, i) Biao Shi N * N receives image-data matrix:

   r _q，p(j，i)＝C ^T×R _q，p(ν，μ)×C；

   According to following equation, described reception image-data matrix is formatted into one group comprises s _R(z, y) Xiang H * V receives image-data:

   s _R(j+Nq，i+Np)＝r _q，p(j，i)

   0≤i+Np≤H-1 wherein, and 0≤j+Nq≤V-1;

   According to following equation, described reception image data set is constituted a plurality of V that comprise _{S, r}(j, i) Xiang N * N is through superimposed image-data matrix:

   v _s，r(j，i)＝s _R(j+ωs，i+ωr)；

   According to following equation, described being transformed into through superimposed image-data matrix comprised Sv _{S, r}(ν, μ) coefficient matrix through revising:

   Sv _s，r(ν，μ)＝C×v _s，r(j，i)×C ^T；

   Become coefficient matrix through filtering with described through the correction factor matrix conversion with described quantization error matrix and described filtering parameter, described coefficient matrix through revising comprises uses Sf _{S, r}(ν, μ) Biao Shi item are changed the described step of described coefficient matrix through revising and are carried out according to following equation: ${\mathrm{Sf}}_{s,r}(\mathrm{\ν},\mathrm{\μ})=\[{\mathrm{Sv}}_{s,r}(\mathrm{\ν},\mathrm{\μ})\]\×\[\frac{\[{\mathrm{Sv}}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\[{\mathrm{Sv}}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2+\mathrm{\α}\[E_0(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\]}^{\mathrm{\β}};$

   According to following transform, described coefficient matrix through filtering is transformed into image-data matrix through filtering,

   Rf _{S, r}(j, i)=C ^T* Sf _{S, r}(ν, μ) * C; And

   Convert described image-data matrix to a series of signals of telecommunication through filtering through filtering, thereby the described signal of telecommunication through filtering can be configured to one by bidimensional H * V array of forming through the filtering vision element, described each through the signal of telecommunication of filtering corresponding to a described feature through the filtering vision element.
2. 2. the method for claim 1 is characterized in that, described quantization error matrix E ₀(ν, μ) determine according to following relational expression: $E_0(\mathrm{\ν},\mathrm{\μ})=Q(\mathrm{\ν},\mathrm{\μ})\underset{q}{\mathrm{\Σ}}\underset{P}{\mathrm{\Σ}}\frac{D{c}_{q,p}(\mathrm{\ν},\mathrm{\μ})}{M}$ 。
3. 3. the method for claim 1 is characterized in that, described quantization error matrix E ₀(ν, μ) determine according to following relational expression: $E_0(\mathrm{\ν},\mathrm{\μ})=Q(\mathrm{\ν},\mathrm{\μ})\sqrt{\underset{q}{\mathrm{\Σ}}\underset{p}{\mathrm{\Σ}}\frac{{\left(D{c}_{q,p}(\mathrm{\ν},\mathrm{\μ})\right)}^2}{M}}\·$
4. 4. the method for claim 1 is characterized in that, described orthogonal transform basis matrix C is a dct basis plinth matrix.
5. 5. the method for claim 1, it is characterized in that, change described described step through filtering vision-data matrix comprise from described each through filtering vision-data matrix, extract the step of a core submatrix, and merge a plurality of described core submatrixs to form step by the described bidimensional H * V array that constitutes through the filtering vision element.
6. 6. method as claimed in claim 5, it is characterized in that, the step of described extraction core submatrix comprise that deletion is described and in filtering vision-data matrix, the 0th walk to the (image-data item of (N-1) row is walked in the row of ω/2-1) and (N-ω/2), and delete described in filtering vision-data matrix the 0th row (row of ω/2-1) and (N-ω/2) row are to the image-data item of (N-1) row to the.
7. 7. the method for claim 1 is characterized in that, ω is set to N/2.
8. 8. the method for claim 1 is characterized in that, N equals 8.
9. One kind in the conversion picture coding and compression process that the graphics picture signals are converted to digital image, be used to reduce because of quantizing to cause the method for block artefact influence, wherein compression degree is determined by a scale factor κ and a quantization table, picture intelligence provides as series of electrical signals, each signal of telecommunication is corresponding with the feature of certain element of bidimensional image, and each picture element is configured to a bidimensional H * V array, it is characterized in that, said method comprising the steps of:

   With an orthogonal transform basis matrix bidimensional H * V array that picture element constitutes is carried out the conversion picture coding, to form an encoded image-data set, the step of described conversion picture coding comprises a quantization operations;

   With described orthogonal transform basis matrix described encoded image-data set is decoded, to form an image-data set through decoding;

   Convert described image-data set to a plurality of coefficient of frequency items with described orthogonal transform basis matrix through decoding;

   Calculate the filtering of described a plurality of coefficient of frequency item by a filtering, described filtering is calculated and is comprised described coefficient of frequency item be multiply by a coefficient filtering item to form the step of a plurality of coefficient of frequency items through filtering; And

   Convert described coefficient of frequency to digital image that a noise reduces through filtering.
10. 10. method as claimed in claim 9, it is characterized in that, the described step of changing described image-data set through decoding comprises described image through decoding-step through superimposed image-data matrix array of data set formation, and converting the step of a matrix of frequency coefficients array with described to through superimposed image-data matrix array with described orthogonal transform basis matrix, described matrix of frequency coefficients comprises described coefficient of frequency item.
11. 11. method as claimed in claim 9 is characterized in that, also comprises the step that derives described coefficient filtering item.
12. 12. method as claimed in claim 11 is characterized in that, the described step that derives described coefficient filtering item comprises the step that derives a filtering parameter at least.
13. 13., it is characterized in that described filtering parameter is the function of described scale factor κ as method as described in the claim 12.
14. 14. method as claimed in claim 11 is characterized in that, the described step that derives described coefficient filtering item comprises the step that derives a quantization error matrix.
15. 15. method as claimed in claim 9 is characterized in that, described orthogonal transform basis matrix is a dct basis plinth matrix.
16. 16. method as claimed in claim 9, it is characterized in that, changing described described step through the frequency filtering coefficient comprises with described orthogonal transform basis matrix and is transformed into step through filtering vision-data item with described through filter factor, and with described through filtering vision-data item convert to a series of through filtered electrical signal so that be configured to one by bidimensional H * V array of forming through the filtering vision element through filtered electrical signal with described, thereby form the step of the digital image that described noise reduces, each described through filtered electrical signal corresponding to a described feature through the filtering vision element.
17. 17. method as claimed in claim 16, it is characterized in that, conversion through the described step of filtering vision-data item comprise extract a part described through filtering vision-data item step and merge described part to form the step of one group of bidimensional H * V low noise image-data, the image that the digital image that described noise reduces is reduced by bidimensional H * V noise-data set constitutes.
18. 18. one kind be used for thereby bidimensional is reduced because of quantizing to cause the method for block artefact influence through the decoded picture signal filtering, the block artefact that produces in the decoded picture signal is the result who carries out conversion picture coding, data compression and previous picture intelligence is quantized, previous picture intelligence provides as series of electrical signals, each signal of telecommunication is corresponding to the feature of previous visual a certain element, the compaction algorithms quantification item that used scale factor κ and from a quantization table, derived wherein, and be configured to one by image-data item s through the picture intelligence of decoding _R(z, y) the bidimensional H of Gou Chenging * V group is characterized in that, said method comprising the steps of:

    Define an overlap coefficient of representing with ω;

    According to following relational expression, will constitute a plurality of v that comprise through the picture intelligence group of decoding _{S, r}(j, i) Xiang N * N is through superimposed image-data matrix:

    v _s，r(j，i)＝s _R(j+ωs，i+ωr)；

    According to following matrix form, be transformed into coefficient matrix with described through superimposed image-data matrix through revising with an orthogonal transform basis matrix C:

    Sv _s，r(ν，μ)＝C×v _s，r(j，i)×C ^T；

    Estimate scale factor κ;

    Filtering parameter is defined as the function of described scale factor, and described filtering parameter is represented with α (κ) and β (κ);

    Select one and use E ₀The quantization error matrix of expression;

    Convert described coefficient matrix through revising to through filtering coefficient matrix with described quantization error matrix and described filtering parameter, described coefficient matrix through revising comprises uses Sf _{S, r}(ν, μ) Biao Shi item are changed described described step through the correction factor matrix and are carried out according to following equation: ${\mathrm{Sf}}_{s,r}(\mathrm{\ν},\mathrm{\μ})=\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ})\]\×\[\frac{\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2+\mathrm{\α}\[E_0(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\]}^{\mathrm{\β}};$ $S{f}_{s,r}(\mathrm{\ν},\mathrm{\μ})=\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ})\]\×\[\frac{\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\[S{v}_{s,r}(\mathrm{\ν},\mathrm{\μ}){\]}^2+\mathrm{\α}\[E_0(\mathrm{\ν},\mathrm{\μ}){\]}^2}{\]}^{\mathrm{\β}};$

    According to following matrix form, described coefficient matrix through filtering is transformed into uses rf _{S, r}(j, i) Biao Shi image-data matrix through filtering:

    Rf _{S, r}(j, i)=C ^T* Sf _{S, r}(ν, μ) * C; And

    Convert described image-data matrix to a series of signals of telecommunication through filtering through filtering, so that be configured to one by bidimensional H * V array of forming through the filtering vision element through filtered electrical signal with described, thereby form the digital image that noise reduces, described each through filtered electrical signal corresponding to a described feature through the filtering vision element.
19. 19. method as claimed in claim 18 is characterized in that, described orthogonal transform basis matrix is a dct basis plinth matrix.
20. 20. method as claimed in claim 18, it is characterized in that, change described described step through filtering vision-data matrix comprise from described each through filtering vision-data matrix, extract the step of a core submatrix, and merging a plurality of described core submatrixs to form the step of one group of bidimensional H * V low noise image-data, the digital image that described noise reduces is made of described bidimensional H * V low noise image-data set.
21. 21. method as claimed in claim 20, it is characterized in that, the step of described extraction core submatrix comprise that deletion is described and in filtering vision-data matrix, the 0th walk to the (image-data item of (N-1) row is walked in the row of ω/2-1) and (N-ω/2), and delete described in filtering vision-data matrix the 0th row (row of ω/2-1) and (N-ω/2) row are to the image-data item of (N-1) row to the.
22. 22. method as claimed in claim 18 is characterized in that ω is set to N/2.
23. 23. method as claimed in claim 18 is characterized in that, the step of described estimation scale factor κ comprises the step that the κ setting is equaled the pro rata quantification item of DC in the described quantization table.
24. 24. method as claimed in claim 18 is characterized in that, the step of described definite filtering parameter comprises the step that obtains a described filtering parameter from a numerical tabular at least.
25. 25. method as claimed in claim 18 is characterized in that, κ≤8, and α (κ)=β (κ)=0.
26. 26. method as claimed in claim 18 is characterized in that, 8＜κ≤24, and α (κ) 〉=1, β (κ) 〉=1.
27. 27. method as claimed in claim 18 is characterized in that, 24＜κ≤32, and α (κ) 〉=2, β (κ) 〉=2.
28. 28. method as claimed in claim 18 is characterized in that, κ＞32, and α (κ) 〉=3, β (κ) 〉=3.
29. 29. method as claimed in claim 18 is characterized in that, the step that described selection quantizes error matrix comprises the step that obtains a quantization error item from a numerical tabular at least.
30. 30. one kind be used for thereby bidimensional is reduced because of quantizing to cause the method for block artefact influence through the decoded picture signal filtering, the block artefact that produces in the decoded picture signal is the result who carries out conversion picture coding, data compression and previous picture intelligence is quantized, previous picture intelligence provides as series of electrical signals, each signal of telecommunication is corresponding to the feature of previous visual a certain element, the compaction algorithms quantification item that used scale factor κ and from a quantization table, obtained wherein, and be configured to one by image-data item s through the picture intelligence of decoding _R(z, y) the bidimensional H of Gou Chenging * V group is characterized in that, said method comprising the steps of:

    With an orthogonal transform basis matrix a described picture group is resembled-data item converts a plurality of coefficient of frequency items to;

    By filtering calculating described a plurality of coefficient of frequency items are carried out filtering, described filtering calculating comprises multiply by a coefficient filtering item to form a plurality of steps through the frequency filtering coefficient entry with described coefficient of frequency item; And

    Convert described coefficient of frequency to digital image that a noise reduces through filtering.
31. 31. method as claimed in claim 30, it is characterized in that, change a described picture group to resemble-the described step of data item comprises described image-data item constituted the step through superimposed image-data matrix array, and convert the coefficient matrix array that comprise coefficient of frequency item with described through superimposed image-data matrix array with described orthogonal transform basis matrix.
32. 32. method as claimed in claim 30 is characterized in that, also comprises the step that derives described coefficient filtering item.
33. 33. method as claimed in claim 32 is characterized in that, the described step that derives described coefficient filtering item comprises the step that derives a filtering parameter at least.
34. 34. method as claimed in claim 33 is characterized in that, described filtering parameter is the function of scale factor.
35. 35. method as claimed in claim 32 is characterized in that, the described step that derives described coefficient filtering item comprises the step that derives a quantization error matrix.
36. 36. method as claimed in claim 30, it is characterized in that, changing described described step through the frequency filtering coefficient comprises with described orthogonal transform basis matrix and is transformed into step through filtering vision-data item with described through filter factor, thereby and with described through filtering vision-data item convert to a series of through filtered electrical signal so that be configured to one through filtered electrical signal and form the step of the digital image that described noise reduces by bidimensional H * V array of forming through the filtering vision element with described, described each through filtered electrical signal corresponding to a described feature through the filtering vision element.
37. 37. one kind in the conversion picture coding and compression process that the graphics picture signals are converted to digital image, be used to reduce because of quantizing to cause the image processing facility of block artefact influence, wherein compression degree is determined by a scale factor κ and a quantization table, picture intelligence provides as series of electrical signals, each signal of telecommunication is corresponding with the feature of certain element of bidimensional image, and picture element is configured to a bidimensional H * V array, it is characterized in that, described equipment comprises:

    Thereby be used for that an orthogonal transform basis matrix is acted on the bidimensional H * V array that is made of picture element and form one group of device that comprises the encoded image-data that quantize item;

    Thereby be used for that described orthogonal transform basis matrix is acted on described encoded image-data set and form one group of device through the image-data of decoding;

    Be used for described image-data set conversion picture coding through decoding is formed the device of a plurality of coefficient of frequency items;

    Filter, it acts on described coefficient of frequency item, so that remove to take advantage of described coefficient of frequency item with a coefficient filtering item, thereby forms a plurality of coefficient of frequency items through filtering; And

    Be used for described coefficient of frequency through filtering is converted to the device of the digital image of a noise reduction.
38. 38. image processing facility as claimed in claim 37, it is characterized in that, be used for the described described device that carries out the conversion picture coding through decoded picture-data set comprised being used for constituting a device through decoded picture-data set, and be used for carrying out the conversion picture coding to form a device that comprises the matrix of frequency coefficients array of described coefficient of frequency item through superimposed image-data matrix array described through overlapping image-data matrix array with described.
39. 39. image processing facility as claimed in claim 37 is characterized in that, also comprises the device that is used to derive described coefficient filtering item.
40. 40. image processing facility as claimed in claim 39 is characterized in that, the described device that is used to derive described coefficient filtering item comprises the device that is used to provide a quantization error matrix.
41. 41. image processing facility as claimed in claim 40 is characterized in that, is used to provide the described device of quantization error matrix to comprise a numerical tabular.
42. 42. image processing facility as claimed in claim 37 is characterized in that, described orthogonal transform basis matrix is a dct basis plinth matrix.
43. 43. image processing facility as claimed in claim 37, it is characterized in that, being used to change described described device through the frequency filtering coefficient comprises and is used for carrying out the conversion picture coding to form the device through filtering vision-data item to described through the frequency filtering coefficient, thereby and be used for described through filtering vision-data item convert to a series of through filtered electrical signal so that be configured to one through filtered electrical signal and form the device of the digital image that described noise reduces by bidimensional H * V array of forming through the filtering vision element with described, described each through filtered electrical signal corresponding to a described feature through the filtering vision element.
44. 44. image processing facility as claimed in claim 43, it is characterized in that, be used to change described described device through filtering vision-data item comprise be used to extract a part described device through filtering vision-data item, and being used to merge described part to form the device of one group of bidimensional H * V low noise image-data, the digital image that described noise reduces is made of described bidimensional H * V low noise image-data set.
45. 45. one kind be used for thereby bidimensional is reduced because of quantizing to cause the image processing facility of block artefact influence through the decoded picture signal filtering, the block artefact that produces in the decoded picture signal is to carry out the conversion picture coding, data compression and the result that previous picture intelligence is quantized, previous picture intelligence provides as series of electrical signals, each signal of telecommunication is corresponding to the feature of previous visual a certain element, the compaction algorithms quantification item that used scale factor κ and from a quantization table, obtained wherein, and be configured to one by image-data item s through the picture intelligence of decoding _R(z, y) the bidimensional H of Gou Chenging * V group is characterized in that described equipment comprises:

    With an orthogonal transform basis matrix to a described picture group resemble-data item carries out the conversion picture coding to form the device of a plurality of coefficient of frequency items;

    Filter, it acts on described coefficient of frequency item, so that remove to take advantage of described coefficient of frequency item with a coefficient filtering item, thereby forms a plurality of coefficient of frequency items through filtering; And

    Be used for described coefficient of frequency through filtering is converted to the device of the digital image of a noise reduction.
46. 46. image processing facility as claimed in claim 45, it is characterized in that, be used for to a described picture group resemble-described device that data item is carried out the conversion picture coding comprises and is used for described image-data item is constituted the device through overlapping image-data matrix array, and be used for carrying out the conversion picture coding to form a device that comprises the matrix of frequency coefficients array of described coefficient of frequency item to described through superimposed image-data matrix array.
47. 47. image processing facility as claimed in claim 45 is characterized in that, also comprises the device that is used to derive described coefficient filtering item.
48. 48. image processing facility as claimed in claim 47 is characterized in that, the described device that is used to derive described coefficient filtering item comprises the device that is used to provide a quantization error matrix.
49. 49. image processing facility as claimed in claim 48 is characterized in that, is used to provide the described device of quantization error matrix to comprise a numerical tabular.
50. 50. image processing facility as claimed in claim 45, it is characterized in that, being used to change described described device through the frequency filtering coefficient comprises and is used for carrying out the conversion picture coding to form the device through filtering vision-data item to described through the frequency filtering coefficient, thereby and be used for described through filtering vision-data item convert to a series of through filtered electrical signal so that be configured to one through filtered electrical signal and form the device of the digital image that described noise reduces by bidimensional H * V array of forming through the filtering vision element with described, described each through filtered electrical signal corresponding to a described feature through the filtering vision element.
51. 51. image processing facility as claimed in claim 50, it is characterized in that, be used to change described described device through filtering vision-data item comprise be used to extract a part described device through filtering vision-data item, and being used to merge described part to form the device of one group of bidimensional H * V low noise image-data, the digital image that described noise reduces is made of described bidimensional H * V low noise image-data set.

Images