TW200910967A - Low complexity macroblock mode decision and motion search method for scalable video coding with combined coarse granular scalability (CGS) and temporal scalability - Google Patents

Abstract

In the present invention, to speed up the encoder while minimizing the loss in coding efficiency, the computational redundancy between the coding layers is considered. Depending on the macroblock (MB) coding modes and the quantization parameters (Qp) of the reference/base layer, a look-up table is recursively used to determine the MB modes to be tested at the enhancement layers. In addition, to avoid exhaustive motion estimation, the reference frame indices of the base layer are adaptively reused, and according to the MB partition at the enhancement layer, the initial search point for motion estimation is properly selected from the motion vector at the base layer or the motion vector predictor at the enhancement layer. The proposed schemes are tested with standard sequences in CIF and 4CIF resolutions using 1 base layer, 3 CGS layers, 3 reference frames, and GOP sizes of 8 and 16. As compared with the mode decision algorithm in JSVM 8, the proposed schemes averagely provide 76% improvement in overall encoding time with an average increase of bit-rate below 1%, and an average Y-PSNR loss below 0.01dB.

Description

200910967 IX. Description of the Invention: [Technical Field of the Invention] The present invention relates to an algorithm suitable for communication, video image and multimedia coding, and relates to a low complexity macroblock mode suitable for coarse and time adjustable video coding. Decide on a fast algorithm for dynamic vector estimation. [Prior Art] In the prior art to which the present invention pertains, H. Li, Z.-G. Li, and CWen η '

Fast Mode Decision for Coarse Grain SNR Scalable Video Coding,” IEEE ICASSP, 2006 (hereafter referred to as prior art 1) 'H. Li, Z.-G. Li, and C. Wen, Fast Mode Decision for Spatial Scalable Video Coding,&quot IEEE IS CAS, 2006 (hereafter referred to as prior art 2), H.L i,Z . - G · L i, and C. Wen, Fast Mode Decision Algorithm for Inter-Frame Coding in Fully Scalable Video Coding, " IEEE Trans. Circuits Syst. Video Technol., vol. 16, no. 7, pp. 889-895, 2006 (hereafter referred to as prior art 3) has proposed a fast mode decision algorithm for calculating the relationship between different coding layers. Accelerating the operation of the encoder. In [Prior Art 1], the design of the mode judgment of the coarsely adjustable enhancement layer is such that the macroblock in the enhancement layer has a finer cut than the counterpart of the base layer. The internal prediction of the tunable enhancement layer is forcibly set to one of the Intra4x4 mode or the IntraBL mode. The similar burial is further used to additionally multiplex the mobile information and rate loss loss of each base layer macroblock to achieve spatial tunability' Such as [first ffii technology 2 And in [Prior Art 3], the movement activity and time level of the base layer are used as the background for different temporal layer mode selection. However, all 6 200910967 are only for a coarse adjustability/space enhancement layer. Designed with a reference picture, but not ideal. Since the average percentage of the enhancement layer IntraBL and Intra4x4 modes is over 90%,

L. Xiong, “Reducing Enhancement Layer Directional Intra Prediction Modes," ISO/IEC JTC1/SC29/WG11 and ITU-T SG16 Q.6, JVT-P041, 2005 [hereafter referred to as prior art 4] and Y. Libo, C Ying, Z. Jiefu, and Z. Feng, "Low Complexity Intra Prediction for Enhancement Layer," ISO/IEC JTC1 /SC29/WG11 and ITU-T SG16 Q.6, JVT-Q0 8 4, 2005 [hereinafter Prior Art 5] The Intral 6xl6 and Intra8x8 prediction modes proposed by the two schemes did not provide considerable coding gain. IntraBL and Intra4x4 modes are ideal for saving internal prediction accuracy without the Intra8x8 or Intra 16x1 6 modes. Therefore, removing Intra8x8 and Intral6xl6 modes reduces the high computational load of internal coding without compromising performance. In addition, the Intra 4x4 type of prediction mode contains 9 different directions, which makes the computational complexity still too high. Therefore, in [Prior Art 4], only the prediction modes (DC, horizontal and vertical) of the three sets of directions are used for the intra-hardening layer prediction. It helps to reduce computational complexity while maintaining a certain coding performance. However, this solution fails to properly close some of the predictive patterns defined in the SVC coding standard and test only the fixed prediction directions of several enhancement layers. Therefore, it would be more desirable if the candidate mode could adapt to changes in the video sequence. In D. Alfonso, SVC Low-complexity Macroblock Mode Decision, " ISO/IEC JTC1 /SC29/WG11 and IT U - TS G1 6 Q. 6, 0Ί 9 , 2Q01 [flying prior art 6], ISVM encoder Lagrangian rate-distortion optimization is used in both motion estimation and macroblock mode judgment to achieve the best compression performance and height calculation at the same time. By including the same low complexity macroblock mode judgment method used in the reference JMH.264/AVC encoder in the JSVM encoder, the average value of the non-mobile estimation work can be obtained in the encoding process of the base layer. 4 4% to 54% time savings' At this time, the average compression loss is between 4.5% and 6.5%, while the average Y-PSNR loss is below 0.5 dB. However, such schemes are limited to the implementation of the base layer of the SVC encoder. HHI's Adjustable Video Coding (SVC) Technical Standard, T. Wiegand, G. Sullivan, J. Reichel, H. Schwarz, and M. Wien, “Joint Draft 9 of SVC Amendment,” ISO/IEC JTC1/ SC29/WG11 and ITU-T SG16 Q.6, JVT-V201, 2007 [hereafter referred to as prior art 7] are now nearing completion at the joint committee of ISO and ITU. The basic design commemoration is to extend the most advanced compression standard, H.264/AVC, and reuse most of its innovative components. Joint Scalable Video Model (JSVM), J. Reichel, H_Schwarz, and M. Wien, Joint Scalable Video Model JSVM-9," ISO/IEC JTCUSC29/WG11 and ΤΤίΖ-Γ CM JVT-V202, 2007 [hereinafter referred to as prior art 8], it refers to the bit stream of the encoder architecture (Bit-stream) 'can provide space, time and SNR adjustability at the same time. Time adjustability can be predicted by B-level in closed loop architecture The space and SNR adjustability is achieved through the use of the layering method. In this draft, the IS VM encoder uses the rate-distortion optimization (RD0)-based exhaustive search technique for each macroblock (MB). Select the best coding mode. For each macroblock, the calculation of RDcost in SVC must perform integer conversion, quantization, inverse quantization, inverse integer conversion and entropy coding of forward and reverse programs. Optimization (RD0) technology can achieve the maximum possible coding performance'] The complexity of the SVM encoder's encoding-decoding operation is too high to be effective in practical applications. Therefore, a kind of 8200910967 can reduce the computational complexity of SVC, but Not The algorithm that causes considerable coding loss is what I am guilty of.

In the Adjustable Video Coding (SVC) technical standard, the available prediction modes for each macroblock (MB) located in the base layer and the enhancement layer include three internal prediction modes (Intra4x4, Intra8x8, and Intral6xl6), and seven interframes. Inter prediction mode and S KI P mode. For the enhancement layer, additional evaluation uses new inter-layer prediction techniques to eliminate inter-layer correlation. Furthermore, when the corresponding macroblock of the base layer has been intra coded, the use of the additional smooth reference prediction (M0DE_SR) mode reduces the degradation of the base layer prediction. Accordingly, the present invention encodes a plurality of test video sequences and finds that the computational load of a single enhancement layer is 1.3 to 2.8 times that of a base layer. This means that JSVM encoders are too widely used due to their H.264 codec and hierarchical structure. Therefore, how to reduce the required coding time is the subject of the present invention. SUMMARY OF THE INVENTION In view of the above-mentioned prior art, the improvements proposed by the present invention are as follows: (i) retaining information obtained when compressing a base layer, such as: different partition mode (Partition mode) The best reference frame index, the motion vector for different cutting modes, the best cutting mode for each macroblock, and the best intra prediction mode for Intra4x4/Intra8x8. . By collecting the information of these basic layers, it can be used as a reference for compressing the enhancement layer to reduce the number of test modes, thereby achieving a large reduction in the computational complexity of the encoder. (ii) inheritance (i), not only that 'the best cutting mode for each macroblock (Macroblock) 200910967

The information with the Intra prediction mode of the Intra4x4/Intra8x8 is updated as each enhancement layer is compressed, meaning that the enhancement layer will refer to each macroblock of the previous layer (Mac rob). The optimal cutting mode of lock) and the Intra prediction mode of Intra4x4/Intra8x8 are used as the reference basis, instead of always using the information of the basic layer as the reference basis, so that the reference accuracy of the reference is more and more The more precise. In addition, the present invention also proposes a test mode lookup table and a layered adaptation Intra4X4/Intra8x8 prediction direction judgment mode, so that each macroblock (Macroblock) in the enhancement layer can refer to the optimal cutting mode of the corresponding macroblock of the previous layer. Compared with the best internal prediction mode, the query mode and the internal prediction direction judgment mode are used to inform the prediction mode of the required test, and the calculation amount is reduced to reduce the huge calculation amount when determining the optimal prediction mode. (iii) In coding the enhancement layer, the present invention proposes a preferred action vector setting of the initial search point for the cutting mode of the large square (eg, 16x16, 16x8, 8x16, and 8x8). It is appropriate to use the motion vector obtained by the base layer as the motion vector prediction point (Μ 〇U ο nvect 〇rpredict 〇r ); and for the small square cutting mode (eg 8x4, 4x8 and 4x4) The motion vector prediction point can be used by the original SVC algorithm. Also because the enhancement layer has a better motion vector prediction point, the motion vector can be found in a smaller range than the base layer to save computational complexity. Furthermore, the reference picture index used by the enhancement layer macroblock in performing the seek motion vector is likely to have the same reference picture index as its counterpart of the base layer, so only the same reference picture index needs to be in the enhancement layer. The reference picture can be used to search for the best motion vector. However, when the base layer is encoded at a lower bit rate and the macroblock selects the cut size 10 200910967 to be 16x16, an exhaustive search is used. (IV) The fast algorithm proposed in the present invention implements only a non-key picture in the enhancement layer, so the loss of coding efficiency only exists in each group of pictures (GOP). In the mean, the loss of coding performance in each group picture does not affect other group pictures, so the bit rate increase caused by the slight loss of coding efficiency can be controlled very stably. According to the experimental simulation results, the average code rate is increased by less than 1%, and the average decrease of PSNR is within O.OldB.

(v) The algorithm proposed by the present invention can randomly adjust the Candidate mode set to be referred to according to the quality of the base layer. In addition, when the picture quality of the base layer is too poor, the proposed algorithm does not perform the mechanism of residual prediction, and achieves the purpose of reducing the amount of calculation. In short, the fast algorithm proposed by the present invention provides a hierarchical adaptive mode judgment algorithm and a mobile search for adjustable video coding (SVC) operation with coarse adjustability (CGS) and time adjustability. Program. Another object of the present invention is to speed up the encoder speed and minimize the loss of coding performance, and to consider the computational complexity between coding layers, according to the macroblock (MB) coding mode and the reference/base layer quantization parameter (Qp), A look-up table is used recursively to determine the macroblock mode to be tested on the enhancement layer. Furthermore, in order to avoid full motion estimation, the method proposed by the present invention adaptively multiplexes the reference picture index of the base layer, and according to the macroblock cut of the enhancement layer, is properly predicted by the motion vector of the base layer or the motion vector of the enhancement layer. Selecting the search starting point of the mobile estimation successfully improves the calculation efficiency. 11 200910967 [Embodiment] In order to enable a person skilled in the art to understand the specific practices and effects of the present invention, a preferred embodiment is proposed, which is described as follows: 1. Between the base layer and the reinforcement layer Description of the Association Here, the present invention focuses on the analysis of the correlation between coding layers in combination with the configuration of coarse tunability (CGS) and time tunability. The statistical analysis is based on the coding of a basic layer and a coarsely adjustable enhancement layer in the CIF resolution. In short, the following QpB, QpE and Qp. Representing the quantization parameters of the base layer, the enhancement layer, and the reference layer, respectively, according to the prediction of the enhancement layer. Table 1. The intra-predictive mode of the conditional enhancement layer of the intra-coded macroblock mode of the enhancement layer (QPb) -QPe) =(39.29) (QPb-QPe) = (27.17) 16x16 | 16x8 8x16 8x8 16x16 16x8 8x16 8x8 16x16 0.44 ! 0.25 0.28 0.04 0.19 0.1S 0.15 0.10 16x8 0.19 1 0.25 0.12 0.04 0.16 0.18 0.07 0.05 8x16 0.20 j 0.12 0.29 0.04 0.17 0.08 0.17 0.05 8x8 0.17 ! 0.38 0.31 0.88 0.4S 0.56 0.61 0.80 The macroblock cutting of the enhancement layer has similar video characteristics to the macroblock cutting of the base layer and is therefore highly correlated. As shown in Table 1, the cutting of the base layer is an ideal predictive reference for the reinforcement layer cutting. For example, if the macroblock of the base layer is coded with 1 6 X 8 / 8 X 1 6 , its counterpart at the enhancement layer will not have an aspect ratio of 8x16/16x8. In addition, when the base layer uses 8 X 8 cuts to achieve better predictions, the reinforcement layer also benefits from the use of the same cut. Also, an increase in the quality of the reinforcement layer will result in an increase in the percentage of 8 x 8 cuts. The distribution of the internal prediction mode is the same as the macroblock cut, and it also depends on the quality of the base layer and the enhancement layer. See the first figure, which is the distribution map of the internal prediction mode of the enhancement layer. As can be seen from the first figure, when the basic layer uses a smaller quantization coefficient and is a better quality coding, most of the internal prediction is based on the basic layer. Internal coded macroblocks are based on IntraBL. On the other hand, with the gradual improvement of the quality of the reinforcement layer, internal pre-

The measurement is turned from the base layer to the reinforcement layer. In particular, for better predictive results, the percentage of Intra4x4 is significantly higher than the other two modes. In addition, in order to reduce the candidate direction mode of Intra4x4/Intra8x8, the similarity between the basic layer and the optimal direction of the enhancement layer is analyzed here. In the second (a) diagram, if the best mode of the enhancement layer is equal to the best mode of the base layer, its two adjacent modes, or the D C mode, the base layer is highly correlated with the best mode of the enhancement layer. For example, if the best mode of the base layer is vertical mode (mode 0), then the two adjacent modes are mode 5 and mode 7. As shown in the second (1)), more than 70% or 80% of the macroblocks have similar directions in the 11^^414/1111^818 prediction type, and are not affected by the difference between QpB and QpE. The movement vectors of the base layer and the enhancement layer are also highly correlated. For example, in the third U) diagram, when the cut size of the macroblock is larger than 8x8, the motion vector of the base layer can be used as a good prediction of the enhancement layer motion vector. Reference 値. On the other hand, the third (b) diagram shows that the motion vector prediction reference of the enhancement layer is a better choice for a sub-macroblock cut. By adaptively selecting one of the predicted reference frames as the starting point of the search, the range of the search in the enhancement layer can be reduced. Similarly, the statistical analysis also shows that the macroblocks of the enhancement layer are likely to have the same reference picture index as their counterparts to the base layer. However, one of the 13 200910967 exceptions is when the base layer is encoded at a lower bit rate and the macroblock selection cut size is 16 6 1 6 , then the reference picture index of the base layer may not be applied to the enhancement layer. Reliable information. For inter-layer residual prediction, the test is performed when the residual of the base layer is not quantized to zero. In the fourth (a) diagram, when the quality of the base layer is improved by reducing the quantization parameter, the probability of predicting residual error between layers is significantly increased. However, as can be seen from the conditional rate of the fourth (b) graph, only half of the macroblocks subjected to the inter-layer residual prediction test are indeed encoded with residual prediction. In addition, depending on the quality of the enhancement layer, a larger percentage of the macroblocks may be encoded in a BLSkip mode that multiplexes the base layer movement information. Therefore, the encoding time is expected to be greatly shortened by appropriately testing the inter-layer residual prediction. Table 2, comparison table of stratified adaptation mode judgment

Mf » with Qp〇's most tta test candidate type 1 area: Qp〇柃0-33 2 area: Ορο in 34^51 Intra-predictive chess internal internal prediction chess type internal direct 16x16 16x8 8x16 Sx£ Direct 16κ16 16xS 8x16 Sx8 directly Ο Ο Ο Ο ο ο ο Ο 16 ο ο 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 8x8 ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο l Btra4x4 ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο ο <<Measure·&gt; 14 200910967 2. The present invention mode judgment algorithm is based on the facts of the previous paragraph, and the present invention further proposes a layered adaptive mode judgment algorithm and a combination of coarse adjustability and time adjustable The mobile mobile search method. As shown in the fifth figure, in order to utilize the inter-layer correlation, the basic layer encodes with exhaustive search and retains all mobile information for each possible combination. The coding mode to be tested by a reinforcement layer can then be obtained by referring to the quantization parameters and the coding mode of the reference layer in Table 2. In addition, the algorithm in the sixth figure determines the movement based on the size of the cut.

The reference picture of the cable; the starting point is adaptively selected by the motion vector of the base layer or the motion vector prediction reference of the enhancement layer. Where ^ = 0 indicates that the prediction from reference list 0 (Li stO) is off; likewise, ^ = 0 indicates that the prediction from reference list 1 (List 1) is off. 2.1 Hierarchical adaptation mode judgment In order to greatly shorten the coding time while minimizing the coding performance loss, the design of the hierarchical adaptive mode selection in Table 2 is based on the inter-layer correlation. For example, when the reference layer is coded using quantization parameters ranging from 〇 4 to 5 1 ), since the reference layer is of low quality, the inter-layer residual prediction will be skipped. In addition, as indicated by the analysis results, when a macroblock is coded at 16x8/8x16, its counterpart at the enhancement layer will not be estimated with a cut of 8x16/16x8. Also, when a macroblock is coded at 8 x 8 , its counterpart on the enhancement layer will not be tested in any mode having a cut size greater than 8x8. For further improvement, the internal prediction of the enhancement layer is forced to Intra4x4, Intra8x8 or IntraBL. In addition, as shown in the seventh figure, except for the DC mode being the best of the reference layer 15 200910967 mode, only the four modes of the I n t r a 4 χ 4 and the I n t r a 8 X 8 prediction type are tested at the enhancement layer. On the other hand, when the reference layer is encoded with better quality, the main changes compared to the foregoing include: (1) all modes with inter-layer residual prediction are used for testing, and;

(2) When a reference layer macroblock is encoded in 1 6 X 8, 8 X 1 6 or 8 X 8 , only the 8 X 8 prediction mode and the mode with the same cut are tested. In the latter case, although Table 1 indicates that these designs may not be optimal in terms of mode distribution, the experimental results show that the effect of replacing 16x16 cuts with 16x8 or 8x16 cuts on coding performance is negligible, especially when the enhancement layer is coded at a higher quality. Time 2.2 The layered adaptive mobile search analog to mode selection, the sixth layer of the layered adaptive mobile search system by using the basic layer of mobile information design, thereby avoiding the exhaustive mobile search. When one of the enhancement layer macroblocks is to be tested in a prediction mode other than 16 6 16 , the associated reference index located in the base layer is multiplexed. However, in the 16x16 prediction mode, when the coding quality of the reference layer is worse than I, it means that the reference picture index located at the base layer cannot be reliably used for the enhancement layer, and therefore it is necessary to perform an exhaustive search at this time. The present invention is tested in a standard sequence using CIF and 4CIF parsing using a base layer, three coarsely adjustable layers, three reference pictures, and 8 and 16 group picture sizes. Compared with the mode judgment algorithm of ISVM8, the present invention provides an improvement of 76% on the overall coding time, and only increases the bit transmission rate below 1% and the Y-PSNR loss below O.OldB. In summary, the present invention has the following advantages: 1. The proposed algorithm can be used in the characteristics of the encoder to include multiple coarsely tunable 16 200910967 code layers (Multiple CGS layers) and its reference picture can be multiple pictures ( Multiple reference frames), make full use of the information obtained by the base layer, such as MB blockition mode (MBpartitionmode), motion vector (Motion vector), the best reference frame index (Reference frame index) and the best internal Intra prediction mode, etc., to achieve a large reduction in the computational complexity of the encoding end and close to the original encoding efficiency. 2. The proposed algorithm still retains the prediction module of In tra8x 8, and changes the prediction mode in different directions according to the characteristics of the local image. 3. The proposed algorithm, both in terms of coding time savings and coding efficiency loss, is better than the method proposed in [Prior Art 6]. In addition, the algorithm proposed in [Prior Art 6] is limited to the implementation of the base layer, and the method of multi-layer coding is not proposed, and in some test images, significant coding efficiency loss is also generated. The present invention has been described in terms of the preferred embodiments described above, and is not intended to limit the scope of the invention. Equivalent changes and modifications of the technical features and spirits described in the claims are intended to be included in the scope of the invention. [Simple description of the diagram] Figure 1 is a distribution diagram of the internal prediction mode of the enhancement layer. Fig. 2 shows (a) Intra4x4/Intra8x8 prediction direction (for reference) (b) Correlation between the basic layer and the internal prediction direction between the reinforcement layers. Figure 3 is the difference of the motion vector with (a) the motion vector of the base layer (b) the enhancement layer movement 17 200910967 The vector prediction reference 値 is used as a reference. Figure 4 shows (a) the rate of test residual prediction (b) the conditional rate of prediction using residuals. Figure 5 is a flow chart for determining the layered adaptation mode. Figure 6 is a flow chart of hierarchical adaptation selection with reference to the picture index. Figure 7 is a hierarchical adaptive Intra4x4/Intra8x8 prediction direction judgment diagram. [Main component symbol description]

18

Claims (1)

200910967 X. Application for Patent Park: 1. A low-complexity macroblock mode decision and dynamic vector estimation fast algorithm for coarse and time-adjustable video coding, characterized by: Reserved in the compressed base layer (Base layer) The information obtained at the time, by collecting the information of these basic layers, is used as a reference for compressing the enhancement layer, reducing the number of test modes and reducing the range of motion vectors, thereby achieving a large reduction in coding. The computational complexity of the device. f 2 . The low complexity macroblock mode decision and dynamic vector estimation fast algorithm applicable to coarse and time adjustable video coding according to item 1 of the patent application scope, wherein the compressed base layer is obtained The information is: the best reference frame for different partition modes (Reference frame index), the motion vector for different cutting modes, the best cutting mode for each macroblock (Macroblock), Intra4x4/ Intra prediction mode of Intra8x8. 3. For the low-complexity macroblock mode decision and dynamic vector estimation fast algorithm for coarse and time-adjustable video coding, as in the second paragraph of the patent application, each macroblock (M acr 〇b 1 〇ck The optimal inter prediction mode/inte prediction mode and the Intra prediction mode of the Intra4x4/Intra8x8 are updated as each enhancement layer is compressed, ie This layer of enhancement layer will refer to the best cutting mode of each macro block (Macroblock) in the previous layer and the Intra prediction mode of Intra4x4nntra8x8 as the reference basis, so as to avoid using the information of the basic layer forever. And then make its reference based on the accuracy of 19 200910967 accurate. 4. A test pattern lookup table for low complexity macroblock mode decision and dynamic vector estimation fast algorithm for coarse and time adjustable video coding, characterized in that the table uses the patent application scopes 1 to 3 The method described in the item, for each macroblock (Macroblock) in the enhancement layer, can refer to the optimal cutting mode of the corresponding macroblock of the previous layer and its optimal internal prediction mode, and inform the prediction mode of the required test by using a lookup table. To perform the operation. 5. As applied in the third paragraph of the patent application, r and R, which are suitable for coarse and time-adjustable video coding; low-complexity macroblock mode decision and dynamic vector estimation fast algorithm, where the motion vector is encoded when the enhancement layer is coded The 寻找nitial search point is set as follows: For the large square cut mode (eg, 16x16, 16x8, 8x16, and 8x8), the motion vector obtained by the base layer is the motion vector prediction. Motion vector predictor; for small square cutting modes (eg 8x4' 4x8 and 4x4), the motion vector prediction point uses the SVC algorithm; thus the enhancement layer has a better motion vector prediction point, so The range of the motion vector can be smaller than the base layer to save computational complexity. In addition, the reference layer index used by the enhancement layer macroblock in performing the seek motion vector can directly use the corresponding base layer. The object has a reference picture index, but when the base layer is encoded at a lower bit rate and the macroblock selects a cut size of 16x16, an exhaustive search is used. 6. The low-complexity macroblock mode decision and the dynamic vector estimation fast algorithm applicable to coarse and time-adjustable video coding according to item 3 of the patent application scope, wherein the fast algorithm only implements the non-key in the enhancement layer Non-key picture, because 20 200910967 This loss of coding efficiency will only exist in each group of pictures (GOP), and the loss of coding efficiency in each group picture will not affect other groups. The picture, so that the bit rate increase due to slight loss of coding efficiency can be stably controlled. 7. A low-complexity macroblock mode decision and a fast motion vector estimation fast algorithm for rough and time-adjustable video coding, as in the sixth application scope of the patent application, wherein the average code due to a slight loss of coding performance The rate of increase (Bit-rate) is within 1%, while the average decrease of PSNR is within O.OldB. 8. The low-complexity macroblock mode decision and dynamic vector estimation fast algorithm applicable to coarse and time-adjustable video coding according to item 5 of the patent application scope, wherein the quality of the base layer can be good or bad. And adaptively adjust the Candidate mode set to be referenced. 9. A low-complexity macroblock mode decision and a fast motion vector estimation fast algorithm for rough and time-adjustable video coding, as in claim 8 of the patent application, wherein when the quality of the base layer is too poor That is, the mechanism of Residual Prediction is not implemented, and the purpose of reducing the amount of calculation is achieved. twenty one