In digital television system such as video telephone, videoconference and high-definition television system, need a large amount of numerical datas to define each frame of video signal, because comprising a sequence, the frame of video signal is called the numerical data of pixel value.Yet, because conventional transmission channel available frequencies limited bandwidth, in order to transmit a large amount of numerical datas by it, just need by utilizing various data compression techniques to compress and reducing data volume, especially under situation such as the low-bit rate video signal coder of video telephone and TeleConference Bridge.
One of this technology that is used for the encoding video signal of a low rate encoding system is so-called object-oriented analysis composite coding technology, wherein an input video image is divided into target, and three groups of parameters that are used to define motion, profile and the pixel data of each target are processed through different encoding channels.
One of this object-oriented encoding scheme example is so-called MPEG (motion picture expert group) stage 4 (MPEG-4), and designing it is for such as low bitrate communication, mutual multimedia (for example recreation, interactive television and this type of) with monitor a kind of content-based mutual, the code efficiency of improvement and/or audio frequency-video encoding standard of universal access of allowing is provided in the application of occasion.(referring to, MPEG-4 Video Verification Model Version2.0 for example, International Standards Organization, ISO/IEC JTC1/SC2P/WG11N1260, in March, 1996).
According to MPEG-4, the input video image be divided into can access a user and the bit stream of manipulation in the corresponding a plurality of video object planes of entity (VOP).One VOP can be known as a target and representing in abutting connection with rectangle by its wide and high minimum multiple that is chosen as 16 pixels (macroblock size) around each target, so that encoder is to pursue VOP, promptly, handle this input video image based on by target.VOP comprises by luminance component (Y) and chromatic component (Cr, Cb) colour information of Zu Chenging and the profile information represented by for example binary mask.
In the profile of processing target, profile information is most important for decomposition and synthetic target shape.Be used to represent that a traditional coding method of this profile information is chain coding (Chain Coding) method.Yet this key shape compiling method needs a large amount of being used to represent its position, though also free of losses in the profile information.
Therefore,, proposed several method, approached with the B-batten such as polygon and approach, with this profile information of encoding for overcoming this defective.One of unfavorable during polygon approaches be profile represent coarse.On the other hand, the B-batten approaches can represent this profile more accurately, yet it needs higher order polynomial to reduce approximate error, and the result has increased the overall computation complexity of this video encoder.
For improving the profile approximation technique that one of technology that this type of problem relevant with the coarse and computational complexity increase of expression profile in above-mentioned approach method introduce is to use discrete sine transform (DST).
Adopt in the device that approaches based on polygon with the profile approximation technique of DST one, as total unsettled U.S.'s sequence number 08/423 at " ACONTOUR APPROXI MATION APPARATUS FOR REPRESENTING A CONTOUR OF AN OBJECT " by name, publicity in 604 the application is determined a plurality of summits and by using polygon to approach to approach the profile of a target with this profile of line-fitting.And select the approximate error at N sampled point and each place of calculating in N sampled point for each line segment and think that each line segment obtains one group of approximate error, wherein each approximate error is represented distance or the displacement between each and this profile in N the sampled point.After this, produce one group of DST coefficient by on every group of approximate error, carrying out one dimension DST computing.
Even aforementioned profile approximation technique based on DST can improve the coarse and complexity of calculation of expression and reduce data quantity transmitted, expect that still it can further reduce data quantity transmitted.
With reference to figure 1, shown block diagram according to the device 200 of the input contour image data that is used to encode of the present invention, this wheel for inputting contour image data representation is formed the position when the profile pixel of front profile of a target.This input contour image data is sent to the summit and selects piece 201, main sampling piece 210, summit mapping block 220 and estimation and motion compensation (ME﹠amp; Mc) piece 280.This summit selects piece 201 by utilizing with the traditional polygon approximation technique of many current line-fittings when front profile determining current summit on front profile, wherein, each current line segment is to form by connecting along two current summits of current profile adjacent distributions.Represent that the current vertex information of the position on current summit selected piece 201 to be delivered to a switch 226 from the summit.
Simultaneously, ME﹠amp; MC piece 280 finds the centre of form of the current and first front profile of this target by the coordinate of all pixel locations on the average current and first front profile, and calculates the motion vector of the space displacement between expression two centres of form, promptly total motion vector (GMV).Calculating the centre of form of working as front profile carries out according to the input contour image data, and the acquisition of the centre of form of first front profile is to carry out according to the previous contour image data from memory 270 retrievals, and wherein previous contour image data represents to form the profile pixel of first front profile and the position on summit.After, by being overlapped onto, first front profile on front profile, produces a prediction profile.In other words, at ME﹠amp; MC piece 280 is by moving GMV with all pixels on the first front profile, so that the centre of form of prediction profile and the consistent prediction profile that provides of the centre of form of working as front profile.Previous summit is also at ME﹠amp; MC piece 280 has been moved GMV and has provided as the motion compensation summit.Via line L20, GMV is by from ME﹠amp; MC piece 280 is delivered to multiplexer (MUX) 290; Via line L30, the prediction contour image data of the position of the profile pixel of expression prediction profile is by from ME﹠amp; MC piece 280 is delivered to sub-sampling piece 230; Via line L40, the motion compensation vertex information of the position on expression motion compensation summit is by from ME﹠amp; MC piece 280 is delivered to summit mapping block 220 and sub-sampling piece 230.
Responsive movement compensation vertex information and input contour image data, summit mapping block 220 is determined a prediction summit for each motion compensation summit, and calculating displacement therebetween, wherein should represent the profile pixel of working as the most close each motion compensation summit on the front profile in the prediction summit.The prediction vertex information via line L50 of the position on representative prediction summit is sent to switch 226; And representative respectively predict the summit and and their corresponding motion compensation summits between the summit motion vector via line L60 pattern of being sent to of displacement determine piece 225 and MUX290.
With reference to figure 2, the summit mapping process of carrying out at summit mapping block 220 places has been described, wherein CC represents to work as front profile, and A to E is illustrated in the motion compensation summit on the prediction profile PC.As shown in the figure, motion compensation summit A to E is mapped to prediction summit A ' respectively to E ', and each prediction summit A ' is to compensate the immediate point in summit with its corresponding sports on front profile CC to E '.
Later with reference to figure 1, pattern determines that piece 225 is according to the coding mode of determining from the summit motion vector that summit mapping block 220 is sent here when front profile again.Particularly, in order to determine coding mode, pattern determines that size that piece 225 calculates the summit motion vectors is with thus with each size and a predetermined threshold TH relatively and count the number of its size greater than the summit motion vector of this threshold value TH.Differ greatly when front profile and previous profile if the number of counting more than or equal to presetting several P (P is a positive integer), is then thought, and coding mode is changed to internal schema, wherein when the coding of front profile not with reference to first front profile.On the other hand, if the number of counting less than number P, is then determined between front profile and first front profile very alike, and, therefore, when the coded system of front profile is confirmed as interactive mode, wherein is based on first front profile and carries out when the difference between front profile when the mutual coding of front profile.
If what determine is internal schema, then pattern determines that piece 225 provides first control signal to switch 226, sub-sampling piece 230 and MUX290; Otherwise provide second control signal.
In internal schema, in response to first control signal of determining piece 225 from pattern, switch 226 is elected to be summit when front profile with current summit, and the vertex data via line L10 of the position on the summit that representative is selected provides to main sampling piece 210, vertex encoding device 227 and contour reconstruction piece 260, and the vertex data in the internal schema is identical with current vertex information; And sub-sampling piece 230 provides all group second error of 0 value to subtracter 235 and adder 255.At vertex encoding device 227, by using any known vertex encoding technology, arithmetic coding scheme for example, the vertex data that the vertex data of encoding is encoded with generation, and the vertex data via line L70 that is encoded is sent to MUX290.
At main sampling piece 210, selecteed, promptly current summit will be worked as front profile and will be divided into a plurality of main outline sections.Each main outline segment table shows the part when front profile of the profile pixel that connects two adjacent selected summits and distribute therebetween; And the main section that each main outline Duan Youyi bar connects two adjacent selected summits is similar to.Then, main sampling piece 210 is got N sampled point and is being calculated a main error in each sample point with a predetermined way on each main section, thus one group of master's error of each main outline section is provided to subtracter 235, herein in a preferred embodiment of the present invention, sampled point on the main section is equidistant each other, and N is a positive integer.Main error represents from a sampled point to the main outline section and from the vertical displacement of drawing to the crosspoint the straight line of this main section of this sampled point, this main error comprises distance and the symbol of indicating this crosspoint with respect to the relative position of this main section between this sampled point and this crosspoint.
At subtracter 235, organizing the inferior error that deducts 0 value the main error from each provides to a transform and quantization (T﹠amp with one group of difference error with every group of master's error; Q) piece 240.In internal schema, all be 0 value from the inferior error of sub-sampling piece 230, and correspondingly, difference error is identical with main error respectively.
By utilizing predetermined transform method, for example discrete sine transform (DST) or discrete cosine transform (DCT), T﹠amp; Q piece 240 is with every group of difference error transform and quantization, to produce one group of quantized transform coefficients of every group of difference error thus.Every group of quantized transform coefficients is by from T﹠amp; Q piece 240 is transferred to statistical coding device 245 and inverse transformation and inverse quantization (IT﹠amp; IQ) piece 250.At statistical coding device 245, by utilizing traditional statistical coding method, VLC (Variable Length Code) technology for example, the conversion coefficient that each group is quantized is encoded.The error information via line L80 that is encoded of all group of conversion coefficient that is quantized is provided to MUX290 from statistical coding device 245.In internal schema, the first control signal via line L90 is sent to MUX290.In response to it, MUX290 selects the vertex data that is encoded in the error information of selecting on the circuit L80 to be encoded and on circuit L70; And with its as provide when the outline data that is encoded of front profile to a transmitter C (not shown) with its transmission.
Simultaneously, at IT﹠amp; IQ piece 250, every group of conversion coefficient that is quantized is converted into one group to be provided thus to the difference error of the reconstruction of adder 255, in adder 255, every group of rebuilt difference error is converted into one group of rebuilt main error, and the rebuilt main error of this group is provided to contour reconstruction piece 260.In internal schema, each rebuilt main sum of errors is identical with its corresponding rebuilt difference error, because the inferior error of importing so far from sub-sampling piece 230 is 0 value entirely.At this contour reconstruction piece 260, front profile is worked as in reconstruction according to the vertex data on the circuit L10, and the current contour image data that the main sum of errors that this group is rebuild is rebuild is sent to memory 270 and is stored in herein with the previous contour image data as next profile, wherein, rebuilt current contour image data comprises the rebuilt summit of working as front profile and the positional information of present picture element.
In interactive mode, pattern determines that piece 225 sends second control signal it is loaded into switch 226, sub-sampling piece 230 and MUX290 on circuit L90.In response to second control signal, switch 226 selects the prediction summit as the summit of working as front profile, and will represent the vertex data of the position on selected summit to provide to circuit L10.Under the interactive mode situation, vertex data is identical with the prediction vertex information.
At main sampling piece 210, when front profile by selected, promptly Yu Ce summit is divided into a plurality of main outline sections.Each main outline section representative in interactive mode connects the part and the profile pixel that is distributed in therebetween when front profile on two adjacent prediction summits, and each main outline section is approximate by a main section that connects two adjacent prediction summits.After this, get one group of master's error with method identical in a kind of and the internal schema.
Simultaneously, in response to second control signal of determining piece 225 from pattern, sub-sampling piece 230 with main sampling block 210 in identical method determine all grouping errors with respect to prediction profile and the motion compensation summit on it.In other words, the prediction profile is divided into contour segment a plurality of times, and the contour segment inferior line segment that is positioned at the two motion compensation summits at time contour segment two ends by a connection is similar to each time; And every group of time error representative is by N displacement between the above-mentioned inferior contour segment that calculates with reference to the mode of main sampling piece 210 and its corresponding the line segment.All group time error is provided to subtracter 235 and adder 255.
As shown in Figure 2, the motion compensation summit A to E on prediction profile PC and the prediction summit A ' on front profile CC to E ' be mutually one to one; And, correspondingly, each main outline section, for example contour segment C ' D ' is corresponding to a contour segment, for example contour segment CD.At subtracter 235, deduct one group error of each corresponding contour segment from one group of master's error of main outline section.For example, describe, if the main error of main outline section C ' D ' is confirmed as d as Fig. 3 A and 3B
1', d
2' and d
3', and the inferior error of inferior contour segment CD is calculated as d
1, d
2, d
3, d then
i' deduct a d
i(i=1,2,3) are to provide a difference error c at subtracter 235
i(=d
i'-d
i).In Fig. 3 A and 3B, suppose that N is 3, and I
1' to I
3' and I
1To I
3Represent main outline section C ' D ' and time profile CD and the sampled point P on main section C ' D ' respectively
1' to P
3' locate and the P on inferior line segment CD
1To P
3The crosspoint of the straight line of place's picture.One group of difference error of each main outline section is provided to T﹠amp; Q piece 240 is transformed and quantizes so that one group of quantized transform coefficients to be provided in these piece 240 these group difference error.
Simultaneously, every group of quantized transform coefficients is at IT﹠amp; IQ piece 250 is changed into one group of difference error of rebuilding and is provided for adder 255, and the difference error of rebuilding in these adder 255 these groups is provided one group of main error of rebuilding to contour reconstruction piece 260 with its corresponding error group mutually.At contour reconstruction piece 260, based on the front profile of working as of vertex data on the circuit L10 and all group of main error generation one reconstruction of rebuilding.To represent from profile reconstructed block 260 that the current contour image data when the reconstruction of the summit of front profile and the positional information on the profile pixel of this reconstruction provides to memory 270 and is stored in this sentences and is used to handle next profile.
At statistical coding device 245, every group of quantized transform coefficients handled in the mode identical with internal schema, and the error information that is encoded of all group of conversion coefficient that is quantized is provided to MUX290 with via line L80.At interactive mode, second control signal on the circuit L90 is provided to MUX290.In response to it, MUX290 is at circuit L20, L60, and selective sequential in the signal on L70 and the L80: the error information that is encoded on circuit L80, at GMV on the circuit L20 and the summit motion vector on circuit L60; And with them as offering transmitter when the outline data that is encoded of front profile with its transmission.At the decoder of receiving terminal, can and be stored in the vertex information that previous vertex information in the memory in the decoder obtains prediction, i.e. vertex data in the interactive mode by GMV, summit motion vector from switch 226; Work as front profile and can rebuild with the method identical with the method for in contour reconstruction piece 260, using.
Though with reference to specific embodiment, it will be apparent for a person skilled in the art that under the prerequisite of defined scope of the present invention in not deviating from following claim, description of the invention can do various modifications and change to it.