MXPA97007106A - Method and apparatus for coding a video signal using a calculation of motion based on unpunto caracterist - Google Patents

Abstract

The present invention relates to a method for detecting a set of motion vectors between a current frame and a reference frame of video signals, using a movement calculation approach based on the characteristic point, where the reference frame includes a reconstructed reference frame and an original reference frame, comprising the steps of: (a) selecting a series of characteristic points of the pixels contained in the reconstructed reference frame, where the series of characteristic points forms a polygonal grid that has a plurality of superimposed polygons, (b) determine a series of almost characteristic points in the current frame, based on the series of characteristic points, (c) assign a series of initial motion vectors for the series of quasi-characteristic points , where each of the initial movement vectors is set to (0,0); (d) point to one of the quasi-characteristic points as an almost-characteristic point object, where the quasi-characteristic point object has a number N of quasi-characteristic points object has a number N of adjacent quasi-characteristic points that form a current polygon object defined by line segments connecting the point almost-characteristic object and the number N of adjacent quasi-characteristic points, N being a positive integer, (e) sequentially adding the initial motion vector of the quasi-characteristic point object to the N number of candidate motion vectors, to produce a number M of updated initial motion vectors, where M is a positive integer, where the number M of candidate motion vectors covers a predetermined region in the current object polygon and the initial motion vectors of the adjacent characteristic points are fixed; f) determine a predicted position in the original reference frame for each pixel contained in the polygon non-current object based on each of the number M of initialized motion vectors for the quasi-characteristic object point and the N number of initial motion vectors of adjacent quasi-characteristic points; (g) provide a predicted pixel value for each pixel, based on the predicted position of the original reference frame, to form a number M of polygons actulaes object predictable, (h) calculate the difference between the current polygon and each of the polygons actulaes object predicted to produce a number M of maximum signal to noise ratios; (i) select one of the updated motion vectors as a selected updated motion vector, which gives rise to a predicted current object polygon, which has a maximum signal to noise ratio, to update the initial motion vector of the quasi-characteristic point object with the selected updated motion vector; (j) repeat the steps (d) to (i) until all the initial motion vectors are updated, (k) repeat step (j) until the repetition is performed for a predetermined number of times, and (n) set the game of initial vectors such as the game of motion vectors, to determine in this way the set of motion vectors

Description

METHOD AND APPARATUS FOR CODING A VIDEO SIGNAL USING A POINT-BASED MOTION CALCULATION CHARACTERISTIC TECHNICAL FIELD OF THE INVENTION The present invention relates to a method and apparatus for encoding a video signal; and, more particularly, still method to encode a digital video signal using a calculation of movement based on a characteristic, improved point, whereby the transmission speed of the digital video signal with good image quality is effectively reduced .

BACKGROUND OF THE INVENTION As is well known, the transmission of digitized video signals can achieve visual images of a much higher quality than the transmission of analog signals. When an image signal comprising a sequence of "picture frames" is expressed in a digital form, a substantial amount of data is generated for transmission, especially in the case of a high definition television system. Since, however, the bandwidth of the available frequency of a conventional transmission channel is limited, in order to transmit the substantial amounts of digital data through it, it is inevitable to compress or reduce the volume of the transmission data. Among the various video compression techniques, the so-called mixed or hybrid coding technique is known as the most effective, combining spatial compression techniques together with a statistical coding technique. Most hybrid coding techniques employ a two-dimensional DCT (discrete cosine transform), a quantification of the DCT coefficients, and the VLC (variable length coding) in a compensated motion DPCM (differential pulse encoded modulation). . Compensated DPCM of movement is a process to calculate the movement of an object between a current frame and a previous frame or future frame, that is, a frame of reference, and to predict the current frame according to the flow of movement of the object for produce a differential signal that represents the difference between the current frame and its prediction. This method is described, for example, in Staffan Ericsson, "Fixed and Adaptive Predictors for Hybrid Predictive / Transform Coding" IEEE Transactions on Communications, COM-33, No. 12 (December 1985); and in Ninomiya and Ohtsuka, "A Motion-Compensated Interframe Coding Scheme for Television Pictures" IEEE Transactions on Communications, COM-30, No. 1 (January 1982). The two-dimensional DCT, which reduces or makes use of spatial redundancies between image data, converts a block of digital image data, for example, a block of 8 x 8 pixels, into a series of data of the transform coefficients. This technique is described in Chen and Pratt, "Scene Adaptive Coder" IEEE Transactions on Communications, COM-32, No. 3 (March 1984). By "processing this transform coefficient data with a quantizer, a zig zag scanner and a VLC, the amount of data to be transmitted can be effectively compressed, specifically, in the compensated CPSD in motion, the current frame data are predicted from the corresponding reference frame based on the calculation of the movement between the current frame and a reference frame.This calculated movement can be described in terms of two-dimensional motion vectors representing the displacement of the pixels between the reference frames There have been two basic methods to calculate the displacement of the pixels of an object: one is a block-by-block calculation and the other is a pixel-by-pixel method In block-by-block calculation, a block is compared in the current frame with the blocks in their reference frame until a better match is determined. It is possible to calculate a vector of displacement between frames (indicating how many blocks of pixels have been moved between frames) for the entire block for the current frame that is transmitted. This block matching technique can be used to predict the P and B frames included in the video sequences, as described in ITU Telecommunication Standardization Sector Study Group 15, Working Party 15/1 Expert's Group on Very Low Bit Rate Visual Telephony, "Video Code Test Model, TMN4 Rev 1", (October 25, 1994), where a P or predictive table determines a table that is forecast from its previous table (according to the reference table) while a B or bidirectionally predictive table is predicted from its previous and future tables (according to the reference table). In the coding of the frame called B, in particular, a bidirectional movement calculation technique is used to obtain forward and backward displacement vectors, in which the forward displacement vector is obtained by calculating the movement of an object between a frame B and its previous table intra (I) or predictive (P) (according to the reference table) and the vector of backward displacement is obtained based on table B and its future table I or P (according to the reference table) .

However, in block-by-block motion calculation, blocking effects on the limit of a block can occur in a motion compensation process; and erroneous calculations can result if all the pixels in the block do not move in the same direction, by means of which the quality of the image in general is diminished. Using a pixel by pixel approach, on the other hand, a displacement is determined for each and every pixel. This technique allows a more accurate calculation of the value of the pixel and has the ability to facilitate changes on a manual scale (for example, zoom or amplification, movement perpendicular to the plane of the image). However, in the pixel by pixel method, since a motion vector is determined in each and every pixel, it is almost impossible to transmit all the motion vectors to a receiver. One of the techniques introduced to reduce the problem of dealing with surplus or superfluous transmission data resulting from the pixel by pixel method is a method of calculating movement based on a characteristic point. In the movement calculation technique based on a characteristic point, the motion vectors for a series of selected pixels, ie characteristic points, are transmitted to a receiver, where each of the characteristic points is defined as a capable pixel. to represent their nearby pixels so that motion vectors for non-characteristic points can be recovered or approximated from those characteristic points in the receiver. In an encoder that adopts motion calculation based on characteristic points, as described in a co-pending application of the public domain US Series No. 08/367, 520, entitled "Method and apparatus for encoding a video signal using motion calculation pixel by pixel ", first select a number of characteristic points from all the pixels contained in the previous table. Then, the motion vectors for the selected characteristic points are determined, where each of the motion vectors represents a displacement in the space between a characteristic point in the previous frame and the corresponding coincident point, that is, a more similar pixel. , in the current box. Specifically, the corresponding point for each of the characteristic points is searched in a search region within the current frame, using a known block equalization algorithm, where a characteristic point block is defined as a block surrounding the selected characteristic point; and the search region is defined as a region within a predetermined area comprising the position of the corresponding characteristic point. In this case, it would be more desirable or convenient to find only one block of the characteristic point that best matches the complete search region, which corresponds to the selected characteristic point. However, in some occasions there may be a plurality of blocks of the characteristic point that best matches, equivalents, found during the equalization of the characteristic point. As a result, it is difficult to correctly detect a motion vector for the characteristic point with this correlation between the characteristic point block and the corresponding search region. Moreover, bad calculations can result if the search region is not determined according to the displacement in the space between the characteristic point found in the reference frame and a corresponding coincident point, that is, a very similar pixel, in the current frame, to deteriorate through this the quality of the image in general.

DISCLOSURE OF THE INVENTION Therefore, an object of the invention is to provide a method for effectively calculating motion vectors for the characteristic points, by means of which the transmission speed of the digital video signals is effectively reduced with a good image quality. Another object of the invention is to provide an apparatus for use in a video signal encoder system, to calculate, effectively, motion vectors using a movement calculation based on a characteristic point, whereby the transmission speed of the digital video signal with a good image quality is effectively reduced. Another object of the invention is to provide a video signal coding system using selectively a movement calculation based on a characteristic point and a movement calculation based on a block, whereby the overall quality of the image is effectively improved. According to another aspect of the present invention, there is provided a method for detecting a series of motion vectors between a current frame and a video signal reference frame using a motion calculation method based on a characteristic point, wherein The reference table includes a reconstructed reference frame and an original reference frame, the method comprises the steps of: (a) selecting a series of characteristic points from the pixels contained in the reconstructed reference frame, where the series of characteristic points forms a polygonal grid with a plurality of overlapping polygons; (b) determine a series of quasi-characteristic points in the current frame based on the series of characteristic points; (c) assign a series of initial motion vectors for the series of quasi-characteristic points, where each of the initial motion vectors is determined in (0,0); (d) point to one of the quasi-characteristic points as an almost-characteristic point object, where the quasi-characteristic point object has a number N of adjacent quasi-characteristic points that form a current polygon object defined by connecting line segments the quasi-characteristic point object and the N number of adjacent quasi-characteristic points, where N is a positive integer; (e) sequentially add the initial motion vector of the quasi-characteristic point object to the number M of candidate motion vectors to produce a number M of updated initial motion vectors, M being a positive integer, where the number M of Candidate motion vectors cover a predetermined region in the current object polygon and the initial motion vectors of the adjacent characteristic points are fixed; (f) determining a predicted position in the original reference frame for each pixel contained in the current object polygon based on each of the number M of initial movement vectors updated for the near-characteristic object point and the number N of vectors of initial movement of adjacent characteristic points; (g) providing a predicted pixel value for each pixel based on the predicted position of the original reference frame to form an M number of predictive object current polygons; (h) calculate the difference between the current polygon and each of the predicted current object polygons to produce an M number of quotients of the maximum signal to noise (PSNR) (i) select one of the updated motion vectors as a vector of selected updated movement, that of origin to a current polygon predicted object with a maximum PSNR, to update the initial motion vector of the characteristic point-characteristic object with the selected updated movement vector; (j) repeating steps (d) to (i) until all the initial movement vectors are updated, (k) repeating step (j) until the repetition is carried out a predetermined number of times; and (n) establish the series of initial vectors as the series of motion vectors, so that, by means of this, the series of motion vectors is determined. In accordance with another aspect of the present invention, an apparatus for use in a video encoder system is provided for determining a series of motion vectors between a current frame and a frame of video signal reference using a motion calculation based at a characteristic point, wherein the reference frame includes a reconstructed reference frame and an original reference frame, the apparatus comprises: the first selection means for selecting a series of pixels from the reconstructed reference frame as a series of characteristic points, wherein the series of characteristic points forms a polygonal grid having a plurality of superposed polygons; the means to determine a series of quasi-characteristic points in the current frame corresponding to the series of characteristic points; the memory means for storing a series of initial motion vectors for the series of near-characteristic points, wherein each of the initial motion vectors is set to (0,0); the second means of selection to select a number L of quasi-characteristic points object from the series of quasi-characteristic points, where each of the quasi-characteristic points object has a number N of adjacent quasi-characteristic points forming a current non-superimposed object polygon defined by the line segments that connect to the quasi-characteristic point object and the N number of adjacent quasi-characteristic points, where L and N are positive integers; the adding means to add the initial motion vector of each of the quasi-characteristic points object to the number M of candidate motion vectors to generate the number M of initial movement vectors updated for each of the quasi-characteristic points object, M being a positive integer, where the number M of candidate motion vectors covers a predetermined area in each of the current non-superimposed object polygons and the initial motion vectors of the adjacent characteristic points for each of the near-characteristic points Object are fixed: the means to determine a predicted position on the original reference frame for each pixel contained in each of the current non-overlapping object polygons based on each of the updated initial motion vectors and the initial motion vectors of the corresponding adjacent quasi-characteristic points; means for obtaining a predicted pixel value from the original reference frame based on the predicted position for, by means of this, forming the M number of predicted current object polygons for each of the current non-superimposed object polygons; the means to calculate the differences between each of the current non-superimposed object polygons and the corresponding M number of the current polygons predicted objects to produce the number M of quotients of the maximum signal to noise (PSNR) for each of the current polygons object not overlapping; the third selection means to select one of the updated initial vectors for each of the quasi-characteristic object points, as a selected updated initial motion vector that gives origin to the current polygon predicted object with a maximum PSNR to produce L number of vectors of initial movement selected selected; The means for updating the initial motion vector for each of the quasi-characteristic objects stored in the memory medium with the corresponding selected initial movement vector corresponding; and the means for recovering the series of initial motion vectors from the memory means as the series of motion vectors when all the initial motion vectors are updated a predetermined number of times. According to another aspect of the present invention, "provides an apparatus for encoding a digital video signal to reduce the transmission speed of the digital video signal, this digital video signal has a plurality of frames that includes a current frame and a reference frame, the apparatus comprises: a first memory means for storing a reconstructed reference frame of the digital video signal; the second memory means for storing an original reference frame of the digital video signal; of motion compensation to detect a number of motion vectors between the current frame and the reconstructed frame of reference using a movement calculation based on blocks and to generate a first predicted current frame based on the number of motion vectors and the frame of reconstructed reference; the second movement compensation means for selecting a series of characteristic points from the reconstructed reference frame to detect a series of motion vectors between the current frame and the original reference frame corresponding to the series of characteristic points using a calculation of movement based on the characteristic point, and to generate a predicted second table based on the series of motion vectors and the reconstructed reference frame; the means for selectively providing the number of motion vectors and the first predicted current frame or the series of motion vectors and the second current frame predicted as the selected motion vectors and the predicted current frame; means for transforming-encoding an error signal representing the difference between the current predicted frame and the current frame, to produce a transformed encoded error signal; and means for statistically encoding the transformed encoded error signal and the selected motion vectors to produce a coded video signal to be transmitted.

BRIEF DESCRIPTION OF THE DRAWINGS The foregoing and other objects and features of the present invention will be apparent from the following description of the preferred embodiments given in conjunction with the accompanying drawings, in which: Figure 1 is a signal encoding apparatus of image having a motion compensation device based on a characteristic point according to the present invention; Figures 2A and 2B represent schematic diagrams exemplifying two frame sequences. Figure 3 shows a detailed block diagram of the motion compensation device shown in Figure 1; Figure 4 presents an exemplified block diagram of the motion vector finder block illustrated in Figure 3; Figures 5A and 5B provide an exemplary diagram of the current frame and the previous frame reconstructed; Figures 6A to 6E describe exemplary diagrams to show the characteristic point selection operation according to the present invention; and Figures 7A and 7B illustrate a diagram as an example describing the process of searching the motion vector according to the present invention.

MODES OF CARRYING OUT THE INVENTION Referring to Figure 1, a block diagram of an image coding system according to the present invention is shown. The image encoding system comprises a frame reordering circuit 101, a subtracter 102 and an image signal encoder 105, an image signal decoder 113, an add-on 115, a first memory device 120, a second memory device 130 , an entropy encoder 107 and a motion compensation device 150. A digital input video signal includes two sequences of frames (or images) as shown in FIGS. 2A and 2B: a first frame sequence is provided with a intra (I) II frame, three bidirectionally predictive frames Bl, B2, B3 and three predictive boxes Pl, P2, P3; and a second sequence of frames has an intra (I), II, three forward predictive frames, Fl, F2, F3 and three predictive frames Pl, P2, P3. Therefore, the image coding system includes two sequence coding modes: a first sequence coding mode and a second sequence coding mode. In the first sequence coding mode, a line L17 is coupled to line 11 by means of a first switch 103 and the first sequence of frames including II, Bl, Pl, B2, P2, B3, P3, is applied to through the first switch 103 to the frame resetting circuit 101 which is adapted to reorder it in a rearranged digital video signal of, for example, II, Pl, Bl, P2, B2, P3, B3, to obtain bidirectionally predicted frame signals for the frames B. The rearranged digital video signal is then provided to a second switch 104A, the first memory device 120 and the motion compensation device 150 through the lines L18, L12, Ll, respectively. In the second sequence coding mode, the line L17 is coupled to an IOL line by means of the first switch 103 and the second block sequence II, Fl, Pl, F2, P2, F3, P3, is coupled through the first switch 103 to first memory device 120, motion compensation device 150 and second switch 104A in lines L12, Ll, L18, respectively. The first switch 103 is activated by a control signal of the sequence mode CS1 from a conventional system controller, for example, a microprocessor (not shown). As can be seen from the above, since there is a delay in reordering in the first sequence coding mode, the second sequence coding mode can advantageously be used as a low delay mode in applications such as video and audio devices. teleconferences.

As shown in FIG. 1, the image coding system includes the second switch 104A and a third switch 104B which are used to selectively perform two frame coding modes: an intra frame coding mode and a frame mode. coding of inter frames The second and third switches 104A and 104B, as is well known in the art, are activated simultaneously by means of a frame mode control signal CS2 from the system controller. In the intra frame coding mode, the intra II frame is directly coupled as a current frame signal via a line L14 to an image signal encoder 105, where the signal of the current frame is encoded in a series of quantized transform coefficients, for example, using a discrete cosine transform (DCT) and any of the known quantization methods. The intra II frame is also stored as an original reference frame in a frame memory 121 of the first memory device 120, wherein the first memory device 120 includes three frame memories 121, 122 and 123 which are connected to the device of motion compensation 150 through lines L2, L3 and L4, respectively. Next, the quantized transform coefficients are transmitted to an entropy encoder 107 and an image signal decoder 113. In the entropy encoder 107 the quantized transform coefficients of the image signal encoder 105 are coded together using, for example , a variable length coding technique; and they are transmitted to a transmitter (not shown) for transmitting it. On the other hand, the decoder of the image signal 113 converts the quantized transform coefficients of the encoder of the image signal 105 back to an intra-reconstructed frame signal using inverse quantization and a discrete inverse cosine transform. The intra-reconstructed frame signal from the image signal decoder 113 is then stored as a reconstructed reference frame in a frame memory 131 of the second memory device 130, wherein the second memory device 130 includes three frame memories 131, 132, 133, which are connected to the motion compensation device 150 through the lines L'2, L'3, L'4, respectively. In the inter-coding mode, an inter-frame, for example, the predictive Pl, the bidirectionally predictive frame or the forward predictive frame Fl, are applied to a current signal of the subtracter 102 and the motion compensation device 150, and it is stored in the frame memory 131 of the first memory device 120, wherein the so-called inter frames include bidirectionally predictive frames Bl, B2, B3, the predictive frames Pl, P2, P3 and the forward predictive frames Fl, F2, F3 . The original reference frame previously stored in the frame memory 121 is then coupled via the line L2 to the motion compensation device 150, and is moved or stored in the frame memory 122. The motion compensation device 150 includes a channel-based motion compensation channel and a motion compensation channel based on the characteristic point, as described below. When the current frame is a predictive frame Pl, the signal of the current frame on the line Ll and a reference frame signal reconstructed on a line L '1 from the frame memory 131 of the second memory device 130 are processed by the use of the channel-based motion compensation channel to predict the current frame with view on the production of the current frame signal predicted on a L30 line and the series of motion vectors on line L20. When the current frame is the forward Fl (or bidirectionally predictive Bl) box, the current frame signal on the line Ll, the signal of the original reference frame on one of the lines L2, L3 and L4 from the first memory device 120 and the reconstructed reference frame signal on one of the lines L ' 2, L'3 and L'4 of the second frame memory 130 are processed by using the motion compensation channel based on the characteristic point to predict the current frame to generate a predicted current frame signal on an L30 line and a series of motion vectors on an L20 line. The motion compensation device 150 will be described in detail with reference to FIG. 3. The current frame signal predicted on the line L30 is subtracted from a current frame signal on the line L15 in the subtracter 102 and the resulting data is say, an error signal that determines the differential value of the pixel is sent to an image signal encoder 105, where the error signal is encoded in a series of quantized transform coefficients, for example, using a DCT and any of the known quantification methods. That is, the errors obtained by subtracting the current predicted table from the current table are encoded in the DCT. In this case, the size of the quantizing step is set to a large value, to compensate only for the severely deformed area caused by incorrectly calculated modification vectors. Subsequently, the quantized transform coefficients are transmitted to an entropy encoder 107 and an image signal encoder 113. In the entropy encoder 107, the coefficients of the quantized transforms of the image signal encoder 105 and the motion vectors transmitted via the line L20 from the motion compensation device 150 are coded together using, for example, a variable length coding technique; and they are transmitted to a transmitter (not shown) for transmitting it. On the other hand, the image signal decoder 113 converts the coefficients of the quantized transforms of the image signal encoder 105 back to a reconstructed error signal using the inverse quantization and the inverse discrete cosine transform. The error signal reconstructed from the decoder of the image signal 113 and the signal of the current frame predicted on the line L16 from the motion compensation device 150 are combined through the switch 104B in the adder 115 for, by means of this, provide a reconstructed reference frame signal through the line L '1 to be stored as the previous frame in the second frame memory 130. The frame memory device 130 includes, for example, the three memories of frames 131, 132 and 133 which are connected in series as shown in Fig. 1. That is, the frame signal reconstructed from the adder 115 is first stored in, for example, the frame memory 131, and then sent to the motion compensation device 150 through line L2 and also is displaced in the second frame memory 132 frame by frame if the next frame signal reconstructed from the additive r 115 is entered into the first frame memory 131. This process is repeated sequentially as long as the image coding operation is performed. Referring to Figures 2A and 2B, exemplary diagrams showing the first and second frame sequences are provided as described above. As shown, when the current frame is the predictive frame Pl, a series of movement vectors SMV1 is obtained block by block using the intra-reconstructed frame II as the reference frame retrieved from the second frame memory 130. In a similar manner , the series of motion vectors SMV2 and SMV3 for the current frames b2 and B3 are obtained using the reference frames Pl and P2. When the current frame is the bidirectionally predictive frame Bl, a series of forward motion vectors FMV1 are obtained from the characteristic points using the reconstructed reference frame II recovered from the second frame memory 130 and the original reference frame II retrieved from the first memory 120. In a similar manner, the series of forward motion vectors BMVl for the current frame Bl is obtained using the original reference frame Bl and the reconstructed reference frame Pl. Subsequently, the image coding system chooses between the series of forward motion vectors FMV1 and the series of backward motion vectors BMV1 and transmits the corresponding motion vectors. When the current frame is the forward predictive frame Fl, a series of forward motion vectors FMV2 are obtained from the characteristic points using the original reference frame II recovered from the first memory device 120 and the reconstructed reference frame. recovered from the second memory 130. As can be seen from the above, for the calculation and compensation of the movement, the frames contained in the first and second frame sequences are arranged in the first and second frame devices 120 and 130 as is shown in Tables I and II, Table I First sequence of tables Ll II Pl Bl P2 B2 P3 B3 L3 XX (II) (Pl) Bl (P2) B2 Table II Second sequence of tables Ll 11 Fl Pl F2 P2 F3 P3 L2 X r Fl (P) F2 (P2) F3 L3 X X @ Fl @ F2 @ L4 X X X II Fl Pl P2 Where a (~) indicates a box that is used to calculate forward movement and / \ determines a box that is used for calculating backward movement. As can be seen in the above, the predictive tables Pl, P2, P3 are reconstructed using the predictive coding based on the DCT, called TMN4, using the movement calculation based on blocks; and the intermediate frames, that is, the bidirectionally predicted tables Bl, B2, B3 or the forecasted tables Fl, F2, F3 are reconstructed using a discrete cosine transform with motion compensation (MC-DCT) based on the point improved characteristic according to the present invention. Referring to Figure 3, details of the motion compensation device 150 are illustrated as shown in Figure 1. As seen in Figure 3, the motion compensation device 150 includes input selectors 154, 155 and 156, a motion compensation circuit based on blocks 151, a first motion compensation circuit based on characteristic point 152, a second motion compensation circuit based on characteristic point 153 and output selectors 157 and 158. The compensation circuit of block-based movement 151 employing a conventional block matching algorithm serves to detect a series of motion vectors for each of the predictive frames Pl, P2, P3; and to generate a current predicted table for the corresponding predictive chart. Therefore, when the predictive frame Pl, as described in Tables I and II, is applied as a current frame to the motion compensation circuit based on blocks 151, the selector 154 serves to couple the intra-reconstructed frame II on the line L'2 as the reference frame for the motion compensation circuit based on block 151. In the motion compensation circuit based on blocks 151 a series of motion vectors is calculated and the signal of the frame is calculated through it Current predicted. Subsequently, the series of motion vectors and the predicted current frame signal are respectively coupled through the output selectors 157 and 158 on lines L20 and L30. The first motion compensation circuit based on the characteristic point 152 employing an affine transform as described below serves to detect a series of forward-calculated motion vectors for each bidirectionally predictive frame Bl, B2, B3 or the predictive frame towards forward Fl, F2, F3 and to generate a predicted current frame for the bidirectionally predictive or forward corresponding frame. Therefore, when the bidirectionally predictive frame Bl on the line Ll is applied as the current frame to the motion compensation circuit based on the characteristic point 12, the selector 155, as shown in Table I, serves to couple the original intra frame II on line L2 as the original reference frame for the motion compensation circuit based on characteristic point 152. Selector 156 serves to couple the intra-reconstructed frame II on line L'2 as the reconstructed reference frame for the circuit of motion compensation based on characteristic point 152 to generate the predicted picture. In the first motion compensation circuit based on characteristic point 152, a series of forward-calculated motion vectors is calculated using the original and reconstructed reference frames and a predicted current frame signal is reconstructed using the reconstructed reference frame. Then, the series of forward-calculated motion vectors and the predicted current frame signal are respectively coupled via output selectors 157 and 158 on lines L20 and L30, wherein the output selectors 157 and 158 are controlled by control signal CS5 and CS6 from the system controller (not shown). The second motion compensation circuit based on characteristic point 153 employing an affine transform as described below serves to detect a series of backward-calculated motion vectors for each of the bidirectionally predictive frames Bl, B2, B3 and for generate a predicted current frame for the corresponding bidirectional predictive box. Therefore, when the bidirectionally predictive frame Bl is applied as the current frame to the second movement compensation circuit based on characteristic point 153, the original predictive frame Pl on line L2 is coupled as the original reference frame to the compensation circuit. of motion based on the characteristic point 153 and the reconstructed predictive frame Pl on the line L'2 is coupled as the reconstructed reference frame to the second motion compensation circuit based on the characteristic point 153. On the second motion compensation circuit Based on the characteristic point 153, a series of backward-calculated motion vectors is obtained using the reconstructed and original reference frames and a predicted current frame signal is constructed using the reconstructed reference frame. Then, the series of backward-calculated motion vectors and the predicted current frame signal are respectively coupled through the output selector 157 and 158 on lines L20 and L30. Referring to Figure 4, there are illustrated the details of the motion compensation circuit based on the characteristic point shown in Figure 3. A reference frame signal reconstructed in the line L'2 of the second frame memory 130 the characteristic point selection block 210 is entered to generate a series of characteristic points, and a movement compensation block 240. The series of characteristic points is then coupled to the motion vector search block 230 and the compensation block. of motion 240. The motion vector search block 230 receives the original reference frame and the current frame and serves to generate a series of motion vectors for the series of characteristic points. The series of motion vectors is coupled to the motion compensation block 240 which serves to generate a predicted current frame based on the series of motion vectors and the series of characteristic points. In the selection block of the characteristic point 210, the series of characteristic points is selected from a multiplicity of pixels contained in the reconstructed reference frame, each of the characteristic points being defined in terms of the position of a pixel. As an example, a current frame and a reconstructed reference frame are shown in FIGS. 5A and 5B. Referring to Figures 6A to 6E, explanatory diagrams are shown which represent a characteristic point selection process according to the present invention. As shown in Figure 6A, margins are detected in the reconstructed reference frame p "(x, y) shown in Figure 5B, using a margin detector known as Sobel (see, AK Jain," Fundamentals of Digital Image Processing ", 1989 Prentice-Hall International) The output from the operator Sobel is compared to a predetermined threshold T. The predetermined default threshold Te is selected as 6 according to the present invention. 'of the operator Sobel "is less than the threshold default, the output value 'is set to 0. Otherwise, the output value 'may remain unchanged. Therefore, a marginal image signal eg. { x, y) as shown in Figure 6A is defined as follows: Í if | y (x, y) \ < Te vp (x. YM) \, or «t.h..erwise In a preferred embodiment of the present invention, feature dots are predetermined using a grid technique using a hexagonal grid having a plurality of overlapping hexagons as shown in Figure 6B. As shown in Figure 6C, a hexagon 610 is defined by line segments connecting 7 grid points 611 to 617. The grid point 617 contained in a hexagon 610 comprises more adjacent grid points 611 to 616 than a tetragon, whereby the characteristic points are allowed to be organized more effectively. The hexagon 610 includes 6 non-superimposed triangles 621 to 626 and the grid points 611 to 617 are the vertices of the triangles 621 to 626. The resolution of the hexagon 610 is defined by the lines HH and HV, which according to the present invention, preferably they are set to 13 and 10, respectively. Referring to Figure 6D, for each of the grid points, for example, Gl to G4, the non-overlapping search ranges are established, for example, SR1 through SR4. A marginal point, for example, 7 located in the search range SRl becomes a characteristic point for the grid point, for example, Gl, if the value of its value of the 8 pixels surrounding the marginal point, by example, 7, is maximum. Therefore, the characteristic point Di can be represented as follows: where eg. { x, y) is a value of the marginal point contained in the search region SRi and i is a positive integer. The series of characteristic points is determined using equation 2, where the series of characteristic points includes a grid point that is superimposed on a marginal point, a marginal point located in the non-superimposed search region SRi and that has the summation value maximum of its adjacent pixel points, and the grid point with no marginal point contained in its non-superimposed search range. If there is more than one marginal point with the same maximum value in the summation, then the marginal point closest to the grid point is selected as a characteristic point. When the series of characteristic points is determined, the hexagonal grids shown in Figure 6B are deformed as a characteristic dot hexagonal grid as shown in Figure 6E. After the hexagonal grid of the characteristic point is determined, the series of characteristic points is coupled to the search block of the movement vector 230 as shown in Figure 4 which serves to detect a series of vectors of movement of this. In accordance with the present invention, a convergence process is used which employs an affine transform to search the series of motion vectors. Referring to Figures 7A and 7B, a diagram illustrating the process of searching the motion vector according to the present invention is exemplified. In the current frame a series of quasi-characteristic points is determined using the series of characteristic points, where each of the characteristic points in the reconstructed reference frame is mapped to the corresponding quasi-characteristic point in the current frame. For each of the quasi-characteristic points, for example, DI to D30, the initial motion vector is set to (0,0). When the quasi-characteristic point, for example, D7 is then assigned or set as an almost-characteristic point object to be processed to calculate its motion vector, a current polygon object 700 is used in the convergence process. The current polygon in question 700 is defined by the line segments connecting the quasi-characteristic points D7 object and its adjacent quasi-characteristic points, for example, DI to D6 surrounding the quasi-characteristic point D7. The current polygon 700 includes 6 non-superimposed triangles 701 to 706, where the quasi-characteristic point object is located at a common vertex of the triangles. A predetermined amount of candidate motion vectors are then sequentially added to the initial motion vector of the near-characteristic point D7, wherein the predetermined amount of candidate motion vectors is preferably selected in the range from 0 to ± 7, in the horizontal and vertical direction, and the candidate motion vector D7Y1 is not allowed since triangle 701 is inverted. A candidate motion vector D'xi is added to the initial vector of the quasi-characteristic point object D7 without changing the initial motion vectors of its adjacent characteristic points from DI to D6 to produce an updated initial motion vector D7D '7. , the updated initial movement vector D7D '7 represents a displacement between the quasi-characteristic point object D7 and a quasi-characteristic point candidate D'7. A predicted position for each of the pixels contained in the current object polygon 700 is determined in the original reference frame using the updated initial motion vector and the initial vectors of adjacent quasi-characteristic points. Then, each of the pixel positions contained in the current object polygon 700 is interpolated by a pixel value in the original reference frame corresponding to the predicted position to form a predicted current object polygon. According to a preferred embodiment of the present invention, this process is performed by a known affine transform in each of the triangles, for example, 701 which has the three characteristic points, for example, DI, D2, D7 as its vertices . The affine transform is defined as follows: Eg 3 where (x, y) represents the x and y coordinates of a pixel within the predicted current object polygon; (x ', y') defines the coordinates of a predicted position in the original reference frame; and from a to f are the coefficients of the affine transform. The six mapping parameters a, b, c, d, e, f are determined solely by using the motion vectors of the three quasi-characteristic points, for example, Di, D2, D7. Once the coefficients of the affine transform are known, each of the remaining pixels in the triangle 701 can be mapped at a position in the original reference frame. Since the obtained predicted position (x ', y') of the original reference frame is not a series of integers in most cases, a technique known as bilinear interpolation is used to calculate the gray level interpolated in the predicted position. { x ', y'). The afine mapping process is applied to triangles 701 to 706, independently. The current polygon is then obtained object predicted for the candidate motion vector. The predicted object current hexagon is then compared to the current hexagon 700 and verified if a maximum signal to noise ratio (PSNR) of the predicted object current hexagon and the current hexagon is increased. If this is the case, the initial motion vector (0,0) of the quasi-characteristic point object D7 is updated with the updated initial motion vector D7D'7.

The process is repeated for the remaining candidate motion vectors. The above process is also performed on all quasi-characteristic points contained in the current frame in an iteration. Referring to Figure 7B, assuming that the iteration is completed, the quasi-characteristic point D7 is set to an almost-characteristic object point; and the initial movement vectors updated for adjacent quasi-characteristic points DI to D6, are D1D'2, D2D'2, D3D'3, D4D'4, D5D '5, and D6D' 6; and, in a similar way, the predetermined candidate motion vectors are added sequentially to the initial vector of the quasi-characteristic point object D7D '7. For example, the candidate motion vector 6 ^ 7X2 is added to the initial vector of the near point - characteristic object D7D '7 without changing the initial motion vectors of its 6 adjacent characteristic points DlD'l, D2D'2, D3D'3, D4D'4, D5D' 5 D6D'6. Therefore, the updated initial motion vector becomes D7X2. The predetermined amount of candidate motion vectors, as described above, is preferably selected in the range of 0 to ± 7, in the horizontal and vertical direction, and the candidate motion vector D7Y2 is not allowed since the triangle 701 is inverse.

A predicted position for each of the pixels contained in the current object polygon 700 is determined by the original reference frame using the updated motion vector D7X2 and the initial vectors of the adjacent quasi-characteristic points DlD'l, D2D'2, D3D '3, D4D '4, D5D' 5, D6D '6. Next, each of the pixel positions contained in the current object polygon 700 is interpolated by a pixel value on the original reference frame corresponding to the predicted position to form a current polygon predicted object 700 '(represented by a dashed line as shown in Figure 7B). The current predicted object hexagon 700 'is then compared to the current hexagon and checked if the PSNR of the predicted current hexagon is increased and the current hexagon is increased. If this is the case, the initial motion vector of the quasi-characteristic point object D7D '7 is updated with the updated initial motion vector D7X2. The process is repeated for the remaining candidate motion vectors. The previous process is also performed in all the quasi-characteristic points contained in the current frame in a second iteration. This process is also performed with respect to all quasi-characteristic points several times until convergence is reached. Preferably, the iteration for the process is established for five times, since, in most cases, the motion vectors converge before the fifth iteration. As can be seen from the above, in the process of convergence, a displacement of each of the characteristic points is given to a vector of movement of these and the six triangles of each of the hexagons are affine-transformed independently using the displacements of its vertex characteristic points. And the offset provides a better PSNR, the motion vector of the characteristic point object is updated sequentially. Therefore, the convergence process is very efficient in the equalization process to determine the predicted image as similar as possible to the original image having objects with amplification, rotation or scale. According to a preferred embodiment of the present invention, for the hardware implementation, this process can be carried out in three stages. The quasi-characteristic points defined as DI, D3 and D5 as shown in Figure 7A, which form a non-superimposed current object polygon, are first processed simultaneously using each of the 6 adjacent characteristic points (D2, D7, D6, DIO, Dll, D17), (D2, D4, D7, D12, D13, D19), (D4, D6, D7, D8, D9, D15). The same process is repeated afterwards for points D2, D4 and D6. As a last step, the remaining points D7, D8 and D9 are processed last. Returning to Figure 4, the motion vectors obtained for all quasi-characteristic points are then coupled as the series of motion vectors for all characteristic points with the motion compensation block 240 which serves to generate a current frame signal predicted by using the reconstructed reference frame. That is to say, the signal of the predicted current frame is obtained by means of the related transformation using the reconstructed previous frame and the motion vectors obtained. From the above it can be seen that, this is the same mapping used by the affine transform that is used for the motion vector search process, except that the reconstructed reference frame is used instead of the original reference frame, since a decoder system (not shown) has only reconstructed reference frames. On the other hand, since the coding system that uses this movement compensation based on the characteristic point produces a considerably good image only with the motion vectors, the difference or error signal between the current frame and the predicted current frame can not be transmitted. As can be seen from the above, it is easily appreciated that the inventive coding system that uses motion compensation based on the characteristic point can obtain a reliable series of motion calculation, which improves the efficiency of the coding. The motion compensation algorithm based on the characteristic point is based on the characteristics of the image and the affine transformation is used to compensate for the rotation and amplification of the object. In most cases, compensated motion images have a superior PSNR with good subjective quality. If motion prediction fails in cases of large scale movement, the error image can be encoded and transmitted using DCT with a large quantization step, specifically, a good subjective quality is obtained using the inventive coding system 24 Kbps. Furthermore, since the positions of the characteristic points change from frame to frame, the inventive coding system uses, like the frame of reference, a reconstructed previous frame that exists both in the encoder and in the decoder so that it does not it is necessary to transmit the position information of the characteristic points. further, this motion compensation of the pixels used in the coding system of the present one produces a subjective quality better than block-based movement compensation, since the amplification, rotation and scale of the objects can be compensated using the related transformation only with the motion vectors. While the present invention has been shown and described in relation to particular embodiments, those skilled in the art will appreciate that multiple changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (17)

1. CLAIMS 1.
2. A method for detecting a series of motion vectors between a current frame and a video signal reference frame using a motion calculation method based on the characteristic point, where the reference frame includes a reference frame reconstructed and an original reference frame, the method comprises the steps of: (a) selecting a series of characteristic points from the pixels contained in the reconstructed reference frame where the series of characteristic points forms a polygonal grid with a plurality of overlapping polygons; (b) determine a series of quasi-characteristic points in the current frame based on the series of characteristic points; (c) assign a series of initial motion vectors for the series of quasi-characteristic points, where each of the initial motion vectors is terminated at (0,0); (d) point to one of the quasi-characteristic points as an almost-characteristic point object, where the quasi-characteristic point object has a number N of adjacent quasi-characteristic points that form a current polygon object defined by connecting line segments the quasi-characteristic point object and the N number of adjacent quasi-characteristic points, where N is a positive integer; (e) sequentially adding the initial motion vector of the quasi-characteristic point object to the number M of candidate motion vectors to produce a number M of updated initial motion vectors, where M is a positive integer, where the number M of candidate motion vectors covers a predetermined region in the current object polygon and the initial motion vectors of the adjacent characteristic points are fixed; (f) determining a predicted position in the "original reference frame for each pixel contained in the current object polygon based on each of the number M of initial movement vectors updated for the quasi-characteristic point object and the number N of vectors of initial movement of the adjacent characteristic features; (g) providing a predicted pixel value for each pixel based on the predicted position from the original reference frame to form an M number of current predictive object polygons; (h) calculate the difference between the current polygon and each of the predicted current object polygons to produce an M number of quotients of the maximum signal to noise (PSNR) (i) select one of the updated motion vectors as a vector of selected updated movement, that of origin to a current polygon predicted object with a maximum PSNR, to update the initial motion vector of the characteristic point-characteristic object with the selected updated movement vector; (j) repeating steps (d) to (i) until all the initial movement vectors are updated, (k) repeating step (j) until the repetition is carried out a predetermined number of times; and (n) establish the series of initial vectors as the series of motion vectors, so that by means of this the series of motion vectors is determined. .
3. The method as recited in claim 1, wherein step (a) includes the steps of: (a) detecting a marginal image of the reconstructed reference frame, wherein the marginal image eg (x, y) is defined as follows [\ vp (x, y) \, otherw? se where p. { x, y) represents the reference frame; | * _3 (x, y) I determines a known output of the Sobel operator; and Te is a predetermined threshold; (a2) establishing a polygonal grid in the marginal image wherein the polygon grid includes a number of grid points to form the plurality of superimposed polygons; (a3) assign a superimposed search range for each of the grid points; (a4) determine the series of characteristic points where the series of characteristic points includes a grid point that superimposes a marginal point, the marginal point is located in the non-superimposed search range and has a maximum sum value of its surrounding pixel points, and the grid point has no marginal point contained in its non-superimposed search range. .
4. The method, as mentioned in claim 3, wherein the series of characteristic points includes a marginal point closer to the polygonal grid when more than one marginal point having the same maximum sum value appears in the search range. .
5. The method, as mentioned in claim 3, wherein the polygon is a hexagon and N is 6.
6. The method, as mentioned in claim 4, wherein the current hexagon object includes 6 triangles defined by the line segments connecting the near-characteristic object point; and its adjacent quasi-characteristic points and steps (f) and (g) are made using a known affine transform.
7. The method, as mentioned in claim 5, wherein the number of surrounding pixel points is eight; the default repetition number is 5; and the predetermined threshold is 6.
8. The method, as mentioned in claim 6, wherein the predetermined region is in the range of 0 ± 7, in the horizontal and vertical direction, The method, as mentioned in claim 7, where the characteristic point Di is defined as follows. where EG. { x, y) is a value of the marginal point contained in the search region and i is a positive integer.
9. An apparatus for use in a video encoder system, for detecting a series of motion vectors between a current frame and a reference frame of the video signals, employing a motion calculation based on the characteristic point, wherein the frame of reference includes a reconstructed reference frame and an original reference frame, the apparatus comprises: the first selection means for selecting a series of pixels from the reconstructed reference frame as a series of characteristic points, wherein the series of characteristic points it forms a polygonal grid with a plurality of overlapping polygons; the means to determine a series of quasi-characteristic points on the current frame that correspond to the series of characteristic points; memory means for storing a series of initial motion vectors for the series of near-characteristic points, wherein each of the initial motion vectors is set to (0,0); the second means of selection to select the number L of quasi-characteristic points object from the series of quasi-characteristic points, where each of the quasi-characteristic points object has a number N of adjacent quasi-characteristic points forming a current non-superimposed object polygon defined by the line segments connecting the quasi-characteristic point object and the N number of adjacent quasi-characteristic points, where L and N are positive integers; the adding means to add the initial motion vector of each of the quasi-characteristic points object to the number M of candidate motion vectors to generate the number M of initial movement vectors updated for each of the quasi-characteristic points object, M being a positive integer, where the number M of candidate motion vectors covers a predetermined area in each of the current non-superimposed object polygons and the initial motion vectors of the adjacent characteristic points for each of the quasi-characteristic points objects are fixed; the means to determine a predicted position in the original reference frame for each pixel contained in each of the current non-superimposed object polygons based on each of the updated initial motion vectors and the initial motion vectors of the normal points corresponding adjacent features; the means for obtaining a predicted pixel value from the original reference frame based on the predicted position for, by means of this, forming the M number of predicted current object polygons for each of the current non-superimposed object polygons; the means to calculate the differences between each of the current non-superimposed object polygons and the corresponding M number of current polygons predicted objects to produce the number M of quotients of the maximum signal to noise (PSNR) for each of the current polygons object not overlapping; the third selection means to select one of the updated initial vectors, for each of the quasi-characteristic points object, as a selected updated initial motion vector which gives origin to the current polygon predicted object with a maximum PSNR to produce L number of selected initial movement vectors selected; means for updating the initial motion vector for each of the quasi-characteristic object points stored in the memory medium with the corresponding selected initial movement vector selected; and the means for recovering the series of initial motion vectors from the memory means as the series of motion vectors when all the initial motion vectors are updated a predetermined number of times.
10. The apparatus, as mentioned in claim 9, wherein the first selection means includes: means for detecting a marginal image of the reconstructed reference frame, wherein the marginal image eg. { xry) is defined as follows: [\ vp (x, y) \, otherwise where ~ p [x, y) represents the reference frame; \ p (x, y) I determines a known output of the Sobel operator; and Te is a predetermined threshold; means for providing a grid of polygons in the marginal image wherein the grid of polygons includes a number of grid points to form the plurality of superposed polygons; the means for establishing a non-superimposed search range for each of the grid points; and the means to determine the series of characteristic points, where the series of characteristic points includes a grid point that superimposes a marginal point, the marginal point is located in the search range and has a maximum value of summation of its points of surrounding pixels, the grid point has no marginal point contained in its non-superimposed search range.
11. The method, as mentioned in claim 10, wherein the series of characteristic points includes a marginal point closer to the polygonal grid when more than one marginal point with the same maximum sum value appears in the search range.
12. 12. The apparatus, as mentioned in claim 11, wherein the polygon is a hexagon and N is 6.
13. 13. The apparatus, as mentioned in claim 12, wherein the current hexagon object includes 6 triangles defined by line segments. which connect the quasi-characteristic point object and its adjacent characteristic features; and the means for determining the predicted position includes a known affine transformer.
14. The apparatus, as mentioned in claim 13, wherein the number of surrounding pixel points is 8; the default repetition number is 5; and the predetermined threshold is 6.
15. 15. The apparatus, as mentioned in claim 14, wherein the predetermined region is in a range from 0 to ± 7, in the horizontal and vertical directions.
16. 16. An apparatus for encoding a digital video signal to reduce the transmission speed of the digital video signal, the digital video signal having a plurality of frames including a current frame and a frame of reference comprising: the first means of memory for storing a reconstructed reference frame of the digital video signal; the second memory means for storing an original reference frame of the digital video signal; the first motion compensation means for detecting a number of motion vectors between the current frame and the reconstructed frame of reference using a movement calculation based on blocks and for generating a first predicted current frame based on the number of motion vectors and the reconstructed reference frame; the second movement compensation means for selecting a series of characteristic points from the reconstructed reference frame, to detect a series of motion vectors between the current frame and the original reference frame corresponding to the series of characteristic points using a calculation of movement based on the characteristic point, and to generate a second predicted table based on the series of motion vectors and the reconstructed reference frame; means for selectively providing the number of motion vectors and the first predicted current frame or series of motion vectors and the second current frame predicted as the selected motion vectors and the predicted current frame; the means to transform-encode an error signal representing the difference between the current predicted frame and the current frame to produce a transformed-encoded error signal; and means for statistically encoding the transformed-encoded error signal and the selected motion vectors to produce a digital encoded video signal to be transmitted. The apparatus, as mentioned in claim 16, wherein the second movement compensation means includes: the first selection means for selecting a series of pixels from the reconstructed reference frame as a series of characteristic points, in where the series of characteristic points forms a polygonal grid with a plurality of superposed polygons; the means to determine a series of quasi-characteristic points in the current frame corresponding to the series of characteristic points; memory means for storing a series of initial motion vectors for the series of near-characteristic points, wherein each of the initial motion vectors is set to (0,0); the second means of selection to select the number L of quasi-characteristic points object from the series of quasi-characteristic points, where each of the quasi-characteristic points object has a number N of adjacent quasi-characteristic points that form a current non-superimposed object polygon defined by the line segments connecting the quasi-characteristic point object and the N number of adjacent quasi-characteristic points, where L and N are positive integers; the adding means to add the initial motion vector of each of the quasi-characteristic points object to the number M of candidate motion vectors to generate the number M of initial movement vectors updated for each of the quasi-characteristic points object, M being a positive integer, where the number M of candidate motion vectors covers, a predetermined region in each of the current non-superimposed object polygons and the initial motion vectors of the adjacent characteristic points for each of the normal points. characteristic objects are fixed; the means to determine a predicted position in the original reference frame for each of the pixels contained in each of the current non-superimposed object polygons based on each of the updated initial motion vectors and the initial movement vectors of the corresponding adjacent quasi-characteristic points; the means to obtain a predicted pixel value, from the original reference frame, based on the predicted position to, by means of this, form the M number of predicted current object polygons for each of the current non-superimposed object polygons; the means to calculate the differences between each of the current non-superimposed object polygons and the corresponding M number of the current polygons predicted objects to produce the number M of quotients of the maximum signal to noise (PSNR) for each of the current polygons object not overlapping; the third selection means to select one of the updated initial vectors, for each of the quasi-characteristic object points, as an updated initial motion vector selected that originates the predicted current polygon object with a maximum PSNR to produce L number of selected updated initial motion vectors; means for updating the initial motion vector for each of the quasi-characteristic object points stored in the memory medium with the corresponding selected initial movement vector selected; and the means for recovering the series of initial motion vectors from the memory means as the series of motion vectors when all the initial motion vectors are updated a predetermined number of times. The apparatus, as mentioned in claim 17, wherein the first selection means includes: the means for detecting a marginal image of the reconstructed reference frame, wherein the marginal image of eg. { x, y) is defined as follows: vß (x, y), sc erwise where p. { x, y) represents the reference frame; \ f. { x, y) I determines an output of the known Sobel operator; and Te is a predetermined threshold; means for providing a polygonal grid in the marginal image, wherein the polygon grid includes a number of grid points to form the plurality of superposed polygons; the means for establishing a non-superimposed search range for each of the grid points; and the means to determine the series of characteristic points, where the series of characteristic points includes a grid point that superimposes a marginal point, the marginal point is located in the search range and has a maximum value of summation of its points of surrounding pixels, the grid point has no point marginal content in its search range superimposed. The apparatus, as mentioned in claim 18, wherein the polygon is a hexagon, N is 6, the current hexagon object includes 6 triangles defined by line segments connecting the near-characteristic object point and its adjacent quasi-characteristic points.; and the means for determining the predicted position includes a known transformer-affine. The apparatus, as mentioned in claim 19, wherein the number of surrounding pixel points is 8, the predetermined repetition number is 5; and the predetermined threshold is 6. The apparatus, as mentioned in claim 20, wherein the predetermined region is in the range from 0 to ± 7, in the horizontal and vertical directions. SUMMARY OF THE INVENTION This invention relates to an apparatus for encoding a digital video signal to reduce a transmission speed of the digital video signal, the apparatus consists of a motion compensation circuit based on a characteristic point to select a series of characteristic points to from the reconstructed reference frame to detect a series of motion vectors between a current frame and an original frame that corresponds to the series of characteristic points using a movement calculation based on the characteristic point, and to generate a predicted second frame based on the series of motion vectors and the reconstructed reference frame the calculation of movement based on the characteristic point employs a convergence process in which a displacement of each of the characteristic points is given for a vector of movement of these and the six triangles of each of the hexagons suffer a transformation Affinity analysis independently using the displacements of its characteristic vertex points. If the offset provides a better peak signal to noise ratio (PSNR), the motion vector of the object characteristic point is updated sequentially. Therefore, the process of convergence of the inventive is very efficient during the process of correlating to determine the predicted image as close as possible to the original image with approach, rotation or scale of the objects.