CN102611888A - Encoding method for screen content - Google Patents

Abstract

The invention discloses an encoding method for a screen content with higher encoding efficiency. The method comprises the following steps that: in a time domain, a plurality of base colors and an index table are used for representing the images or image blocks of the screen content, wherein the index table is disassembled into a matched table and a non-matched index table; each element in the matched table shows whether the elements of the index table are equal to a predicted index or not; if not, the value of the index is updated and stored in the non-matched index table; the predicted index is generated through two methods; one method comprises the following step that: the index table of the image blocks which are searched after interframe motion estimation serves as a predicted index table with the strong correlation of the time domain; and the other method comprises the following step that: a pixel-level predicted index serves as the predicted index according to the texture direction of pixels around a current pixel with the correlation of a space domain.

Description

A kind of coding method to screen content

Technical field

The invention belongs to the technical field of video coding, relate to a kind of coding method particularly screen content (screen content).

Background technology

In existing technology, the coding of vision-mix (to the coding method of screen content) roughly can be divided into following three types.

1. based on the method for image encoding:

This method still adopts the conventional coding framework to come vision-mix is encoded; But adopt the different quantized step-length to different zones; For the piece that comprises literal and chart content, coding back rebuilds effect can be very fuzzy, so reduce quantization step when quantizing these pieces; Be its overabsorption bit, make the quality of its reconstruction as well as possible.But for fixing bit number, if be literal and the bit of having schemed the table block overabsorption, the bit number that distributed of natural image piece will correspondingly reduce so, so on the whole, this method still can not reach good effect.

2. based on the method for hierarchical coding:

This method adopts the mixed light bar phantom to compress.It is divided into piece image or image block the plane layer of foreground layer, background layer and a binary system operator.The plane layer of this binary system operator shows that each pixel of image or image block belongs to foreground layer or background layer; And JBIG (Joint Bi-level Image Experts Group can be used in this binary system operator plane; Associating bianry image expert group) standard or JBIG2 (Joint Bi-level Image Experts Group2, associating bianry image expert group two generations) standard is compressed.And the foreground layer and the background layer that pass through after the data filling algorithm blurs can compress with the traditional image coding framework.Paper " Compressing compound documents " (The Document and Image Compression Handbook; M.Barni; Ed.New York:Marcel-Dekker; 2005.) this method has been proposed, and compare with traditional encoding and decoding framework through this method of experiment proof, very big gain can be obtained.But this method has some shortcomings: at first gain depends on promptly how image is divided into prospect and background by the algorithm of cutting apart to a great extent.The second, if the effect of coding can be damaged in the hole that splits without any processing.The 3rd, the textcolor of the background layer of encoding respectively and the text shape of binary system operator also can reduce the efficient of coding.So this method still can not satisfy the needs of application.

3. block-based coding method:

This method can be divided into two types:

A kind of is the character of utilizing the spatial domain, like statistics with histogram, and gradient, the numbers of color etc. can be divided into a kind of in several types with the present encoding piece, like text, chart, natural image etc.Different classes is compressed to be applicable to their statistical property with the different coding framework then.

Another kind is to consider that the histogram of literal and chart in the screen content is more sparse; With it is the form of base color and concordance list in Time Domain Decomposition; It is Base color and Index map (BCIM) pattern; As shown in Figure 1, wherein four kinds of colors are represented in " 1 ", " 0 ", " 253 ", " 255 ".Paper " Enable efficient compound image compression in is intra coding H.264/AVC " (Proc.Int.Conf.Image Processing; Oct.2007; Vol.2; Pp.337-340) at first H.264 this method has been applied in the standard, and has obtained good gain, but code efficiency is still waiting to improve.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiency of prior art, provide a kind of code efficiency the higher coding method to screen content.

Technical solution of the present invention is: this coding method to screen content screen content, represent image or the image block of screen content with several base color and concordance list in time-domain; Wherein to be disassembled be the concordance list of a matching list and a non-coupling to concordance list, and whether the element of each the element representation concordance list in the matching list equate with its prediction index, if do not wait, then will upgrade the value of this index and be stored in the concordance list of non-coupling; The prediction index produces through two kinds of methods: a kind of is the strong correlation that utilizes time-domain, and the concordance list of the image block that searches after interframe movement is estimated is used as the prediction concordance list; Another kind is a correlation of utilizing spatial domain, according to the grain direction of the surrounding pixel of current pixel, with the prediction index of Pixel-level as the prediction index.

The reason that code efficiency is higher: because screen content screen content has the sparse character of histogram distribution, so the image block of representing to have screen content with several base color and concordance list is very effective.We so concordance list is decomposed into matching list and non-match index table.Because boolean's property of matching list, promptly each element have only 0 with 1 two kind of value, represent in the concordance list whether index with predict that index equates.When prediction is point-device; Be used for so the distribution meeting of code stream of codes match table very uneven with; 0 and 1 ratio great disparity is very big; If such code stream is encoded with CABAC (Context-based Adaptive Binary Arithmetic Coding is based on contextual binary arithmetic coding), the minimizing amplitude of code check can be very big so.And the non-match index table of encoding is a small probability time, and it only occurs in the index of concordance list and predicts when index is unequal, at this time owing to got rid of a kind of possibility; In order to guarantee continuity; Can be with the value reach of a part, like this, the probability of 0 appearance is higher relatively.Equally CABAC is suitable for, can reduces code check.And above-mentioned two kinds of methods that produce the prediction index have utilized the correlation of room and time, and the accuracy that produces the prediction index is guaranteed, so the method that we propose can be improved the code efficiency of screen content significantly.

Description of drawings

Fig. 1 shows existing BCIM pattern;

Fig. 2 shows the flow chart according to screen content of the present invention coding method;

Fig. 3 is the key diagram that decomposes concordance list;

Fig. 4 shows how to predict index according to the correlation generation of time-domain;

Fig. 5 shows how to predict index according to the correlation generation of spatial domain;

Fig. 6 shows a current pixel and a left side thereof, and is upper left, on, the pixel of upper-right position.

Embodiment

Image or the image block of screen content are represented in this coding method to screen content screen content with several base color and concordance list in time-domain; Wherein to be disassembled be the concordance list of a matching list and a non-coupling to concordance list; Whether the element of each the element representation concordance list in the matching list equates with its prediction index; If do not wait, then will upgrade the value of this index and be stored in the concordance list of non-coupling (as shown in Figure 3); The prediction index produces through two kinds of methods: a kind of is the strong correlation that utilizes time-domain, and the concordance list of the image block that searches after interframe movement is estimated is used as the prediction concordance list; Another kind is a correlation of utilizing spatial domain, according to the grain direction of the surrounding pixel of current pixel, with the prediction index of Pixel-level as the prediction index.

Preferably, as shown in Figure 2, this method may further comprise the steps:

(1) each coding unit CU is divided into predicting unit PU according to the quaternary tree form;

(2) each predicting unit PU is divided into converter unit TU according to the quaternary tree form;

(3) for current TU YUV3 component arranged, carries out the processing of step (4)-(12) for each of each component of YUV:

(4) pixel in the TU is carried out the Dynamic Programming cluster, obtain N representative value, N is the integer greater than 0, and this N value promptly is a N base color, and by the incremental order ordering; Then each pixel among the current TU is mapped in N the base color according to the nearest principle of Euclidean distance, and obtains an index and show which base color it maps to, and calculate the distortion of each TU;

(5) produce the prediction index;

(6) produce matching list according to each index principle whether the prediction index corresponding with it equates;

(7) for those and the unequal index of prediction index, upgrade index, and be stored in the concordance list of non-coupling;

(8) code check that concordance list produced of calculation code base color, matching list and non-coupling;

(9) code check according to distortion that produces in the step (4) and step (8) generation comes the calculation rate distortion cost;

(10) value of from 1 to 4 traversal N, repeating step after the fix N (2)-(9) make step

(9) rate distortion costs of calculating in is minimum;

(11) information of the concordance list of the information of the base color that the rate distortion costs of selecting is minimum and matching list and non-coupling is sent to decoding end;

(12) decoding end is rebuild current TU according to base color and concordance list after obtaining corresponding information;

(13) rebuild PU and CU.

Preferably, the method for acquisition base color may further comprise the steps:

(a) with the line of pixels ordering in the present image; Statistics also writes down the pixel value and the frequency of occurrences thereof in each piece; And all pixels with different value are lined up one line up interval X according to the value incremental order; And the frequency of occurrences of each pixel value also forms an interval Y, and the element in these two intervals of X and Y is one to one;

(b) setting is divided into the N class with the pixel value in the piece, promptly will obtain N base color, and N is the integer greater than 0:

β (on behalf of the element in LowBound to the UpBound interval, lowBound, UpBound N) be assigned to the distortion that is produced in N type,

β(lowBound，UpBound，N)＝α(LowBound，p-1)+β(p，UpBound，N-1) N≥2

α(LowBound，UpBound) N＝1

If N >=2, the P point is tried to achieve through following formula

$p=\underset{\mathrm{lowBound}<P<\mathrm{UpBound}-N+1}{\arg \min }(\mathrm{\&alpha;}(\mathrm{LowBound},p-1)+\mathrm{\&beta;}(p,\mathrm{UpBound},N-1)$

$\mathrm{\&alpha;}(\mathrm{LowBound},p-1)=\underset{i=\mathrm{LowBound}}{\overset{p-1}{\mathrm{\&Sigma;}}}{(X_i-\mathrm{avg}(\mathrm{LowBound},p-1))}^2\&times;Y_i$

$\mathrm{avg}(\mathrm{LowBound},p-1)=\underset{i=\mathrm{LowBound}}{\overset{p-1}{\mathrm{\&Sigma;}}}\frac{X_i\&times;Y_i}{Y_i};$

(c)

β(lowBound，UpBound，4)＝α(LowBound，p1-1)+β(p1，UpBound，3)

β(p1，UpBound，3)＝α(p1，p2-1)+β(p2，UpBound，2)

β(p2，UpBound，2)＝α(p2，p3-1)+β(p3，UpBound，1)

β(p3，UpBound，1)＝α(p3，UpBound)

Wherein pi is the optimum point that finds in every layer line property search, and i is 1,2 or 3,

(p3, UpBound), (p2, p3-1), (p1, p2-1), (LowBound p1-1) is exactly the base color of being tried to achieve to α to α to α to α, and LowBound＜p1＜p2＜p3＜UpBound is arranged here.Quantification employing formula: C (S)=sign (S) * max in the step (2) (0, floor (| S|/Q+1-Z), wherein S is an input signal, and floor is following bracket function, and Q is a quantization step, and Z is the adjusting parameter of presetting.

Below specify the method that produces the prediction index, mainly comprise:

1. the correlation according to time-domain produces

According to the correlation of time domain, as shown in Figure 4, the matching image piece that estimation found is carried out the quantification with respect to the base color of current block, obtain a concordance list, as the predicted value of current block concordance list.

2. the correlation according to spatial domain produces

According to the correlation in spatial domain, the index of each pixel in the piece is predicted, weigh at first to four pixel travel direction property testings on every side, and with 2 norms.2 norms are more little, represent texture smooth more, and continuity is strong more, so after 4 pixels around the test, with the index of the respective pixel of the 2 norms minimum predicted value as current index.Fig. 5 has well explained this process.

When the codes match table, context is carried out corresponding design.Be total up to 16 kinds of contexts of its design, the matching of each contiguous on every side pixel have 2 kinds maybe, mate and do not match, represent with 1 and 0 respectively; Like this, obtain 4 neighborhood pixels, i.e. lefts here altogether; The top, upper left side and upper right side, corresponding context is 2 ⁴=16.

Under unmatched situation, get corresponding element in the non-match index table of encoding, at this moment need carry out carrying out contextual adaptively selected to reach better coding effect according to surrounding pixel.

Among Fig. 5 be a current pixel and a left side thereof, upper left, on, the pixel of upper-right position.

For it designs 15 kinds of context models altogether, for better explanation, every kind of context is with a quadruple notation, and expression is left, upper left successively respectively for four-tuple, on, the index that the pixel of upper-right position is represented.These 15 kinds of contexts are respectively { AAAA, AAAB, AABA, ABAA, BAAA, AABB, ABAB, ABBA, AABC, BAAC, BCAA, ABAC, ABCA, BACA, ABCD}.A wherein, B, C, D represent 4 different index values, and for example to represent the index of 4 positions be identical to AAAA, and AAAB represents left, the upper left side, 3 index values of top are the same, top-right index value is different with other three index values.Specific practice is following:

At first obtain 4 pixel corresponding index value up and down, by the frequency of occurrences of 4 index and position thereof see its belong to 15 kinds contextual any, and encode with the context model of correspondence.And before entropy coding, we need remap current index to reach better coding effect.The method that remaps is following: according to context model under it, be 0 with needing the highest index of frequency to remap, the order of successively decreasing by frequency then remap the index that might occur, form a mapping table.At last in mapping table, find the index after current index remaps, and the index after this mapping of encoding.

The above; It only is preferred embodiment of the present invention; Be not that the present invention is done any pro forma restriction, every foundation technical spirit of the present invention all still belongs to the protection range of technical scheme of the present invention to any simple modification, equivalent variations and modification that above embodiment did.

Claims (3)

1. coding method to screen content screen content is characterized in that: image or the image block of representing screen content in time-domain with several base color and concordance list; Wherein to be disassembled be the concordance list of a matching list and a non-coupling to concordance list, and whether the element of each the element representation concordance list in the matching list equate with its prediction index, if do not wait, then will upgrade the value of this index and be stored in the concordance list of non-coupling; The prediction index produces through two kinds of methods: a kind of is the strong correlation that utilizes time-domain, and the concordance list of the image block that searches after interframe movement is estimated is used as the prediction concordance list; Another kind is a correlation of utilizing spatial domain, according to the grain direction of the surrounding pixel of current pixel, with the prediction index of Pixel-level as the prediction index.

2. method according to claim 1 is characterized in that: may further comprise the steps:

(1) each coding unit CU is divided into predicting unit PU according to the quaternary tree form;

(2) each predicting unit PU is divided into converter unit TU according to the quaternary tree form;

(3) for current TU YUV3 component arranged, carries out the processing of step (4)-(12) for each component of YUV component:

(4) pixel in the TU is carried out the Dynamic Programming cluster, obtain N representative value, N is the integer greater than 0, and this N value promptly is a N base color, and by the incremental order ordering; Then each pixel among the current TU is mapped in N the base color according to the nearest principle of Euclidean distance, and obtains an index and show which base color it maps to, and calculate the distortion of each TU;

(5) produce the prediction index;

(6) produce matching list according to each index principle whether the prediction index corresponding with it equates;

(7) for those and the unequal index of prediction index, upgrade index, and be stored in the concordance list of non-coupling;

(8) code check that concordance list produced of calculation code base color, matching list and non-coupling;

(9) code check according to distortion that produces in the step (4) and step (8) generation comes the calculation rate distortion cost;

(10) value of from 1 to 4 traversal N, repeating step after the fix N (2)-(9) make the rate distortion costs of calculating in the step (9) minimum;

(11) information of the concordance list of the information of the base color that the rate distortion costs of selecting is minimum and matching list and non-coupling is sent to decoding end;

(12) decoding end is rebuild current TU according to base color and concordance list after obtaining corresponding information;

(13) rebuild PU and CU.

3. method according to claim 1 is characterized in that: the method that obtains base color may further comprise the steps:

(a) with the line of pixels ordering in the present image; Statistics also writes down the pixel value and the frequency of occurrences thereof in each piece; And all pixels with different value are lined up one line up interval X according to the value incremental order; And the frequency of occurrences of each pixel value also forms an interval Y, and the element in these two intervals of X and Y is one to one;

(b) set the pixel value in the piece is divided into the N class, i.e. the N that tries to achieve base color, N is the integer greater than 0:

β (on behalf of the element in LowBound to the UpBound interval, lowBound, UpBound N) be assigned to the distortion that is produced in N type,

β(lowBound，UpBound，N)＝α(LowBound，p-1)+β(p，UpBound，N-1) N≥2

α(LowBound，UpBound) N＝1

If N >=2, the P point is tried to achieve through following formula

$p=\underset{\mathrm{lowBound}<P<\mathrm{UpBound}-N+1}{\arg \min }(\mathrm{\&alpha;}(\mathrm{LowBound},p-1)+\mathrm{\&beta;}(p,\mathrm{UpBound},N-1)$

$\mathrm{\&alpha;}(\mathrm{LowBound},p-1)=\underset{i=\mathrm{LowBound}}{\overset{p-1}{\mathrm{\&Sigma;}}}{(X_i-\mathrm{avg}(\mathrm{LowBound},p-1))}^2\&times;Y_i$

$\mathrm{avg}(\mathrm{LowBound},p-1)=\underset{i=\mathrm{LowBound}}{\overset{p-1}{\mathrm{\&Sigma;}}}\frac{X_i\&times;Y_i}{Y_i};$

(c)

β(lowBound，UpBound，4)＝α(LowBound，p1-1)+β(p1，UpBound，3)

β(p1，UpBound，3)＝α(p1，p2-1)+β(p2，UpBound，2)

β(p2，UpBound，2)＝α(p2，p3-1)+β(p3，UpBound，1)

β(p3，UpBound，1)＝α(p3，UpBound)

Wherein pi is the optimum point that finds in every layer line property search, and i is 1,2 or 3,

(p3, UpBound), (p2, p3-1), (p1, p2-1), (LowBound p1-1) is exactly the base color of being tried to achieve to α to α to α to α, and LowBound＜p1＜p2＜p3＜UpBound is arranged here.

Images