TW569625B - Bit rate control method of low delay video signal by using dynamic rate table - Google Patents

Abstract

The present invention provides a bit rate control method of low delay video signal by using dynamic rate table. The method is to design a dynamic rate table according to the parameters such as the degree of complication of the video image block, quantization parameters and encoding bit numbers. For each inputted image block, by using the summation of absolute difference and the allocated number of bits to search the dynamic rate table, the best quantization parameters can be found under the requirement of minimum difference between the estimated number of encoding bits and allocated number of bits, so as to quantize and encode the block, and use the encoded practical number of bits to update the dynamic rate table. Thus, the dynamic rate table is updated automatically after each image block is updated, thereby the statistical characteristics of the video image can be quickly captured, and the variation of image content can be tracked correctly, so as to accurately control the output bit rate of the video encoder and obtain a better image reconstruction quality. In addition, since the bit rate control processing of the present invention has only integer computation program, the computational function of bit rate control is greatly enhanced, and the cost of hardware circuit is greatly reduced, so that the whole computation can be completed easily.

Description

569625 A7 ___ B7 V. Description of the invention (1) Background of the invention: (Please read the notes on the back before filling out this page) The present invention is a method for controlling the bit rate of the resolution / editing of video images, especially a method The bit rate control technology developed in the macroblock layer is aimed at low-latency video communication applications by using a dynamic bit rate table to achieve precise control of the output bits of the video encoder. Yuan rate. Prior art: Press, decode / encoding bit rate (hereinafter referred to as bit rate) control, play on video encoders, such as Η 26χ and MPEG encoders, playing The consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs has played an important role in printing. The bit rate control is used to control the bit stream of the video image to be encoded, so that it conforms to the bit rate of the channel to maintain good Image quality. Generally speaking, in order to successfully complete the control of the bit data stream, an encoder buffer is used to temporarily store the encoded bit data. However, this method will cause the bit data to be transmitted. A delay in time occurs during the process. For a real-time video image communication system, the delay incurred by the end-to-end transmission of the video image must be controlled and maintained within a very small range. Therefore, this encoding The capacity of the buffer must be small. In this case, if the number of bits generated for a frame is too large, the buffer will have a data overflow phenomenon. To avoid this phenomenon, the buffer is usually designed to The images of subsequent pictures are skipped and not coded. However, this method will cause discontinuities when reconstructing the image sequence. Conversely, if the number of bits generated by a certain screen is too small, the buffer will underflow bit data, causing no bit data to be transmitted to the channel, causing a waste of channel bandwidth. . Therefore, the goal of bit rate control is to control the bit data generated by the encoder in order to avoid data overflow (ie, frame skipping) or insufficient phenomena in the buffer. Generally speaking, the bit rate control of low bit rate video images can be completed at the frame layer and the macroblock layer (hereinafter referred to as the MB layer). The bit rate control of the picture layer is controlled by each encoding system. It is necessary, but the coding system often cannot accurately adjust the bit rate of the picture by controlling the picture layer alone. In many low-time-delay application systems, such as: video phone and video conference paper rule (CNS) A4 specification (210 X 297 male 1f) 569625 A7 B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention) In the video conference system, precise buffer adjustment functions are required. Therefore, the bit rate control of the block layer must be used in these application systems. Generally speaking, the control process of the block layer is more difficult. Here is a practical example to explain the control process of the block layer as follows: Standard Video coding systems, such as H.263 and MPEG, are augmented by motion compensation and discrete cosine transform (DCT) operations. The image compression ratio, in which motion compensation is completed on a 16x16 block basis, and after completing motion compensation, a motion comPensation difference frame (referred to as error frame) is generated, and then an 8x8 DCT is Apply it to the error screen and generate the corresponding DCT coefficient (coefficient), and then quantize the DCT coefficient using a quantization parameter (referred to as 2 households), and then use a variable length code (variable lenSth code 'referred to as VLC ) Encode it. In H.263 and MPEG systems, each MB allows the use of different quantization parameters. If /), and the bit rate and distortion of the kth MB of an error picture, respectively ) And quantization parameters, and M is the number of MBs in a picture 'is the available coding bit number of the picture (bit budget). The optimal bit rate control of the block layer is to find the best of all MBs in a picture. The quantization parameter is limited by the following bit rate R (Q): Μ R (Q) = Σ rk (q0 < BT K = 1), so that the following overall distortion must be minimized (overaUdistortion): Μ D (Q ) = Σ dM K = 1 The optimization problem can be solved by Lagrange multiplier (mu) ltiPlier) 'This solution must rely on the bit rate-distortion mode (rate-distortion lion' referred to as RD mode) 'Many RD modes have been exposed in many documents However, in the existing RD mode technology, there are the following disadvantages: (1) The RD function is obtained under the assumption of a certain source statistics. __3 ___ — i paper scale is applicable China National Standard (CNS) A4 Specification (210 X 297 mm) " " (Please read the precautions on the back before filling this page) # Order --------- Line_-· 1 569625 A7 B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention) For example: Assume that the source of the signal is Laplaciait distribution. Since these assumptions are only an approximation, the RD model cannot be maintained regularly and accurately; (2) The RD mode is often related to the variation (σ2) of each error MB. However, in a typical video coding system, the sum of absolute difference (SAD) is usually used for motion estimation. (Motion estimation), in order to reduce the computation required. Therefore, the RD model for calculation based on the variation ο2 does not need to be properly modified before SAD criteria can be used and such corrections are usually not accurate enough; (3) According to the RD model, the quantization parameter ρρ is obtained by optimization. However, in low bit rate coding standards, such as: Η.263, the change of quantization parameters between adjacent blocks in a block group (GOB) is limited to two. Between levels (ie, -2 to +2), this limitation reduces the optimization performance, so that the number of bits generated by the screen cannot be accurately controlled, resulting in deterioration of image quality; ⑷R-D mode involves floating point Computing (floating point computation) requires more expensive hardware equipment and more complex computing mechanisms. Summary of the Invention: In view of the many shortcomings in bit rate control in the existing RD modes described above, the inventor finally invented a dynamic bit rate comparison table for low Method for bit rate control of delayed video image. An object of the present invention is to develop an efficient and low-cost bit rate control technology in the block layer (MB). The technology is based on the complexity of MB (that is, SAD), quantization parameter ρρ, and encoding bit Coding (coding bit counts) and other parameters, a dynamic bit rate table (dynamic rate table) is designed, which contains a certain complexity of the MB of SAD, using the quantization parameter ρρ for quantization encoding, the resulting Estimation of the number of coded bits. For each input MB, the method uses its SAD 値 and the allocated number of bits to find the dynamic bit rate comparison table, and estimates the number of encoded bits and the allocated number of bits. Under the requirements of the smallest error, find the best quantization parameter ρρ〇4 This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) (Please read the precautions on the back before filling this page) 569625 A7 B7_ 15. Description of the invention (4) (Please read the precautions on the back before filling this page) Another object of the present invention is to include a bit-distortion mode in the dynamic bit rate comparison table, and the comparison The table will be dynamically updated with the characteristics of the processing block, so the local statistics of the image block can be quickly tracked to achieve the purpose of accurately controlling the bit rate. Yet another object of the present invention is that in this technology, there is only an arithmetic program for integer unitary. Therefore, only a low-cost hardware circuit can be used to complete the entire operation, effectively reducing the cost of the solution / encoding circuit device. Description of drawings and attached tables: Table 1 is a comparison table of bit rates achieved by TMN8 and the present invention; Table 2 is the number of skipped pictures and average signal-to-noise ratio achieved by TMN8 and the present invention in H.263 codec (Referred to as PSNR) comparison table. Figure 1 shows the schematic diagram of the PSNR of each picture when using TMN8 and the present invention to encode the image sequence "Salesman" at a bit rate of 64Kbps; Figure 1 (b) shows the use of TMN8 and the Schematic diagram of the PSNR of each picture when the "Silent" video sequence is coded at 48Kbps bit rate; Figure 2 shows the video sequence of "Silent" using TMN8 and the present invention at 48Kbps bit rate Schematic diagram of the number of bits in the buffer for each picture when encoding at the rate; Figure 2 (b) shows the image sequence of "Mother & Daughter" using TMN8 and the present invention at 24Kbps bit rate When encoding, the schematic diagram of the number of bits in the buffer of each picture is shown in Figure 3. Figure 3 shows the image sequence of "Foreman" using TMN8 and the present invention at the rate of 112Kbps. Schematic diagram of the actual coded bit count of the picture; printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, Figure 3 (b). The video sequence of "Silent" is encoded using TMN8 and the present invention at a bit rate of 64Kbps. Every painting Schematic diagram of real encoding bit counts. Detailed explanation: In general, when deciding a β of a MB, the complexity of the MB should be considered, that is, the amount of variation (ο2) or the absolute difference SAD drawn, and the available bit budget, also That is, as follows: _____5_ ^ The paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm). Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. 569625 a7 ____ B7 5. Description of the invention: Pride, bit Accumulated number) In the present invention, the absolute difference sum SAD is used instead of the amount of variation (0V. This is because SAD can be obtained directly after motion compensation. As mentioned earlier, in the existing bit rate control mode, the function / (·) Is calculated based on the RD mathematical model. The present invention does not use the RD mode, but rather designs a bit rate-complexity comparison table, which is a two-dimensional matrix A [ SADMBk] [βΡ], the first parameter in the array, SADMBk, is used to represent the SAD of the kth MB. The SAD is an integer between (SADmin, SADmax), the array. The second parameter (parameter) ρ / > is the representative Quantization parameters ^ 1, 2, ..., 31, the elements of the array represent an MB (with coding complexity SADMM) in a certain QP (quantization of the estimated number of coding bits) (estimate) In the present invention, in the offline state, the comparison table is constructed according to the following steps: 1) First, a sequence of video images is sequentially input to a block MB unit. Video encoder (such as: H.263); 2) Calculate the SAD 値 of the input MB and set (2P = 1, 2, ...., 31, respectively, to quantize and encode them; 3) Record each 値Enter the actual number of encoding bits after quantization encoding of the MB; 4) Repeat the previous procedure to process all MBs, and the actual number of encoding bits generated by all MBs of the same SAD and QP, that is, each b (SAD, ρρ) elements are averaged, and then stored in the 6 [SADMBk] [g / >] array in order, and the comparison table is constructed. The present invention constructs the dynamic bit comparison table based on the absolute difference sum SAD and the quantization parameter βP. According to the construction principle of the comparison table, those who are familiar with the art can easily consider the equivalent change (such as the amount of variation ο7 Substitute SAD, or replace with other quantization parameters >), shall all belong to the protection scope of the present invention. After completing the construction of the bit rate-complexity-ρρ comparison table, the present invention can sequentially perform frame-layer rate control (MB_layer rate control) and MB-layer rate control (MB-layer rate) online. control) 〇 This paper size applies the Zhongguanjia Standard (CNS) A4 specification ⑵Q χ 297 (Please read the precautions on the back before filling this page) --- r ---- Order ---- I 1 ^ 569625 A7 B7 V. Description of the Invention (g) In the present invention, the bit rate control of the picture layer aims to estimate the target coded bit count of the current input picture. The control procedure is similar to the TMN8 bit rate control procedure. The current day-time encoding BU 'is to first calculate the number of bits in the encoder buffer according to the following formula, also known as buffer fullness: W = max (Wprev + D-R / Fy0) .... .......... (1), where D refers to the actual number of bits used when encoding the previous picture; refers to the fullness in the buffer of the previous picture; Λ refers to Channel bit rate; F refers to the frame rate. At this time, 'if W is greater than a predefined threshold' d) M = R / F, the encoder will skip the subsequent pictures to be encoded until the buffer fullness W is less than M, so each time a frame is skipped, the buffer fullness W will decrease by M bits. In the present invention, if the setting of the current picture is M =, the maximum delay time of the buffer will be M / i? = 1 / F ^. In the present invention, the target encoding bit count of the current picture is estimated according to the following formula:

R Βχ = ..... Δ .............. (2)

F, where △ is calculated according to the following formula:

2 * W

——, W > Z · M Δ = F W- ZZ · M, others, where 値 is set to 0.1. Generally speaking, the higher the complexity of a MB (SAD), the greater the number of bits required. Therefore, in order to improve the coding performance, the present invention first performs motion estimation on all MBs in the MB bit rate control. Test and compensate the initialization operation, and record the SAD and motion vector of each MB, and then classify the MB into a compensable type (SADS critical), or an uncompensable type (SAD > critical) (threshold), because compensable MB does not need to be quantified, only non-book 7 paper sizes are applicable to Chinese National Standard (CNS) A4 (210 X 297 mm) (Please read the precautions on the back before filling this page ) Order · -I-Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 569625 A7 B7 V. Description of Invention (卞) (Please read the precautions on the back before filling this page) Non-texture bits, such as indicator bits Indicator bits and / or motion-vector bits are inserted into the header of the bit stream. On the other hand, non-compensable MBs, because they must be quantified, contain text and non- Bit data for this article The compensated MB is further classified into an inter-coding or an intra-coding type. According to the specifications of H.263, the present invention calculates the number of partial nontexture bits in a picture according to the following formula before encoding according to the following formula: 5 ^^

Buncode = X (Bc〇D, Bc〇D + BmCBPC ^ BcBPY + BmVD, BmVD, 0) ... ⑶ m = 1, where M is the total number of MB in a picture; (X, Y, Z, 0 ) Represents one of X, Y, Z, and 0 according to the encoding mode, where X, Y corresponds to the compensable type, Z corresponds to the non-compensable inter-coding, and 0 corresponds to the Non-compensable intra-coding, such as the number of bits representing the COD (Coded macroblock indication) signal, 5MC sail: the number of bits representing the MCBPC (Macroblock type & Coded block pattern for chorminance) signal Is the number of bits of the CBPY (Coded block pattern for luminance) signal, and is the number of bits of the MVD (Motion Vector Data) signal. As for the estimated total number of bits available for these non-compensable MBs, Fan Xin can be calculated according to the following formula =

Bcode = BT _ Buncode-Bph-BGOBH ............ (4) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, where the office represents the target code bit count in a picture, which can be calculated by Calculated by formula (2); is a bit count representing a picture header, and is a bit count representing a picture GOB (Group of Block) header. In formula (4), 5_ contains the number of data bits and header bits of the text of the non-compensable MB in this text. However, before the quantization and encoding, the header bits of the non-compensable MB are unknown. Therefore, the present invention uses the following formula to estimate the application of the Zhongguanjia standard to the paper size of the non-compensable MB ( CNS) A4 size (21G x 2S7 male 1) 569625

V. Description of the invention

Bava = Bcode-Bh4ntr, Nintra—Bh inte, Ninter ......... ⑸, where 'represents the total number of bits allocated to the non-compensable MB; represents the internal coding = required for each MB The number of header bits w represents the number of header bits required for interactive encoding (excluding motion vector bits (___ _; represents the number of MB remaining for internal encoding); u represents the interaction required The number of remaining MBs encoded. Private and melon-can be obtained by iterative calculation with the following formula: —a ^ hAntra = — (Bh4ntra ^ (j ^ l) + bh.intra) ........ .... ⑹ j

1 1 buckle 1 J

Jih-inter = — (Bh.inter * Ql) + bh.inter) ...... (7) 'Where the melon / coffee represents the use of y · MB (the first MB to The j-th MB) average number of header bits obtained as the internal coding. It represents the number of header bits used by the j-th MB to use internal coding. Where -⑽, it is used on behalf of) MB (first MB) To the j-th MB) as the average number of header bits obtained by the interactive coding; it represents the number of header bits used by the j-th MB to use the interactive coding. After 俟 B «v« is determined, the following formula can be used to estimate the number of bits allocated to the uncompensable k-th MB: ^ ava * bk: ............. ............... ................................(8)

N Σ SAD MBk k = l 'Where SADMM is the SAD of the k-th MB 値, N is the non-compensable type in a picture (please read the precautions on the back before filling this page) Staff Consumption of Intellectual Property Bureau of the Ministry of Economic Affairs Cooperatives print this paper in a suitable paper size _ Home Standard (CNS) A4 Specification _ (21G χ 297 km-

ϋ ϋ | _1 一-口, ϋ ϋ ϋ ^ 1 ^ 1 —a— I ϋ ϋ ϋ n ϋ ϋ rn ϋ ϋ ϋ H ϋ > ^ 1 ^ 1. ^ 1 ^ 1 ^ 1 ^ 1 I 569625 A7 B7___ 5. Description of the invention (0) The total number of MB. At this time, for the k-th MB with SADMm, the best quantization parameter ρρ 値 can be found from the dynamic bit rate comparison table by the allocated bit number center and SAD 値 'according to the following formula: ( (Please read the notes on the back before filling out this page) -1

Qp * = 0 ^ 1,2, .bfr b [SAD ^ \ [QP] |} ............. (9), where the inverse min is the left side Conversely, the difference between the estimated number of allocated bits and the target bit count (targetbitcount) 6 [& 4i) MM] [2 / >] can be minimized. According to the specification of Η263, the difference in gp between horizontally adjacent blocks 値 is limited to the range (-2, _1, +1, +2). Therefore, the best quantification calculated by formula (9) The parameter ρp, 尙 needs to be readjusted so that the difference between ρρ 値 and the previous mb is not greater than 2. The difference between adjacent blocks, that is, D (2i7y47Vr (Quantizer Information) »DQUANTk- QPk -QPk.i ° However, if the foregoing processing occurs at the beginning of a GOB or a picture (that is, the first GOB), The invention will use the calculated 2P to decide to calculate the picture quantization parameters (Quantizer Information 烕 block group quantization parameters instead of D0t / A / Vr, the method is the same as /) 2 £ / Α / ν: Γ is calculated differently , Its processing procedure is: (1) if there is no non-compensable MB in GOB, it is set to any integer in the range of 1 to 31; (2) if there is an non-compensable MB in GOB, it is set; (3) If there are at least two non-compensable MBs in the GOB, it will be set according to the following formula = QPi + 2 printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs if QP2-QPi > h DQUANT: Pj if -h < QP2-QPi < h ....... (10)

QP] -2 if QP2-QP1 <-L, where L is a positive integer, the default value is L = 5, and the modification (formation) of formula (10) will make the first two in a GOB uncompensable (QP 値 difference) of the type MB becomes smaller, and further reduces the encoding distortion. If the first GOB of a picture is processed, GQUANT can be calculated by using formula (10), and PQUANT = GQUANT 〇 ________J10__ This paper size applies to Chinese National Standard (CNS) A4 specification (21 × 297 mm) A7 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 569625 B7 V. Description of the invention (10) In the present invention, The actual encoding bit count is used to update the number of encoding bits in the lookup table. Therefore, in consideration of reducing the cost of hardware implementation, the present invention does not really change each element in the lookup table, but two The dimension comparison table is fixed, and a one-dimensional shift array 56 [SADMBk] is changed. The size of this array is 1 x SADmax, that is, each SAD corresponds to a shift array element. [SADMBk] will be as follows Formula, updated: State SADMBk] [(2P] = ^ [SADMBk] + (h, -6 [SADMBk] [(2P] '[SADMBk]) / 2. ". (Ll) When you want to update the encoding bit prediction of each quantization parameter in the dynamic bit rate comparison table When estimating, you only need to add the lookup table to the shift array, which means: the number of updated estimated coding bits = 6 [SADMBk] [QP] + Shenzhen SADMBk] ........ ....... (12) It should be noted here that each element of the shift array [SAD] is initially set to 0, so that after the comparison table is constructed, there is no need to Those who need to change are only s6 [SADMBk]. The memory space required for this array is only 1/31 of the comparison table. In the present invention, the update principle of the comparison table is based on the actual encoding of the current MB. The bit count is used to update the estimated number of coded bits in the lookup table, but in a preferred embodiment, the lookup table is fixed and implemented by changing only a one-dimensional shift array. Those skilled in the art, according to the principle of updating the comparison table of the present invention, can easily consider the equivalent changes, which should not depart from the protection scope of the present invention. Based on the traditional RD mode In the basic bit rate control technology, many complex floating-point operations are involved, such as: square root operation, multiplication operation, and division operation. In contrast, the main technology of the present invention only includes the lookup of the lookup table and fixed points. (fixed-point) multiplication and division operations. Such operations can be performed in a shift operation. Therefore, in terms of hardware equipment, the technology used in the present invention is significantly higher than the bit rate of the existing RD mode. Control technology has the advantage of lower cost. The cost of the present invention is mainly in the additional memory added by the bit rate comparison table. The size of the memory depends on the range of SADCS ^ D ^, like /) ^). In the following embodiments of the present invention, if the range of QP is 31, and the range of SAD is 1660, the required memory positions are 31x1660 in total. The maximum value of the element (bit count) of the lookup table will be less than 65535, that is, each memory location is in the two standard paper sizes. Applicable to China National Standard (CNS) A4 (210 X 297 public love) (Please read the precautions on the back before filling this page)

569625 A7 B7

V. Description of the Invention (IV byte indicates that it is more than enough, so the required memory size is only about 1⑽K bytes. For the current semiconductor industry, It is a very low cost. (Please read the notes on the back before filling in this page} In a preferred embodiment of the present invention, a basic version of the H.263 codec is used to evaluate the bits of the present invention. The price of the logic of the rate control technology is compared with the bit rate control of TWN8 as follows. In this codec, its motion estimation uses a full search algorithm (hereinafter referred to as FSA) 'and X and Y In the direction, subsampling is performed at a ratio of 2: 1, and the MB of 16x16 is first reduced to 8x8 to reduce the search operation amount, and then the search is performed between -15 and +15 with FSA. In this embodiment, H.263 optional advanced tools such as advanced prediction (AP mode) and unrestricted motion vector are not introduced. The test video includes six QCIF image sequences, and each image sequence is With a frame rate of 10Hz The rate (fame rate). In this embodiment, Table I represents the actual bit rate achieved by the TWN8 and the bit rate control technology of the present invention. The table shows that the bit rate control of the present invention is closer to the target than the TWN8 method Bit rate, Table II represents the comparison of reconstructed pictures (PSNR) quality, which shows that the bit rate control technology of the present invention has better PSNR quality (which averages about 0.8dB more than TWN8). Printed by the Industrial and Commercial Bureau of Industrial and Commercial Cooperation, please refer to Figures 1 to 1 (b), which are the PSNR of each picture of different image sequences. These figures clearly point out that the bit rate control technology of the present invention In the first few pictures, although there is a lower PSNR, however, after several pictures, it quickly surpassed TMN8, and it remained in this state until the end of the image sequence. This phenomenon shows that the present invention Although the bit rate control technology is not ideal at the beginning, it can quickly grasp the statistical characteristics of the image and correctly track the changes in the image content. Therefore, the dynamic bit rate comparison table of the present invention can indeed be more accurate. Accurately reflect the shadow Refer to the figures 2 to 2 (b) for the fullness of each picture in the buffer. In this embodiment, the overflow threshold of the buffer is set to R / F. If the buffer is fuller than the critical threshold, it is called overflow. When the "Mother & Daughter" movie is encoded at a bit rate of 24Kbps, TWN8 overflows 5 times and 5 pictures are displayed. Skip, but under different test conditions, the present invention does not have any overflow phenomenon at all. Because the paper size of this paper is skipped, 12 Chinese paper standard (CNS) A4 (210X297 mm) 569625 Μ B7 V. The number of invention description (12) is related to the continuity of video playback, which means that the invention The continuity of playback is better than TMN8. If the buffer fullness curve touches the X axis, an underflow problem will occur. TMN8 has many bit shortages in most image sequences. However, the present invention does not have this problem. It can also be seen from these figures that referring to the figures 2 to 2 (b), it is the fullness of each picture in the buffer. In this embodiment, the present invention can achieve a lower and more stable buffer full. degree. Refer to Figures 3 (a) to 3 (b), which are actual coded bit counts of each picture. These figures show that the number of bits generated by each picture controlled by the present invention is relatively stable. In summary, since the dynamic bit rate comparison table of the present invention is automatically updated after each image block is processed, it can quickly grasp the statistical characteristics of the video image, and correctly track the changes in the image content, accurately Control the output bit rate of the video encoder to obtain better image reconstruction quality. In addition, in the bit rate control process of the present invention, there is only an integer operation program, so not only the operation performance of bit rate control is greatly improved. , And can greatly reduce the cost of hardware circuits, and easily complete the entire operation. The above are only the preferred embodiments of the present invention. However, the scope of the rights claimed by the present invention is not limited to this. According to those skilled in the art, the technical content disclosed by the present invention can be easily considered. And equivalent changes should all belong to the protection scope of the present invention. (Please read the notes on the back before filling out this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 13 This paper size applies to China National Standard (CNS) Α4 specification (210X 297 mm)

Claims (1)

569625 A8 B8 C8 D8 6. Scope of patent application 1. A method for controlling bit rate of low-latency video image by using a dynamic bit rate comparison table. The method includes the following steps: In the offline state, design a dynamic bit Yuan rate comparison table; (Please read the precautions on the back before filling out this page) Estimate the target code bit count BT of the current input day and day; Perform the initialization operation on the current input screen. The initialization operation includes calculating and recording each MB. SAD 値 and motion vector, then classify MB into compensable and non-compensable, and then further classify non-compensable MB into interactive coding and internal coding, calculate the number of interactive coding and internal coding MB, and Recorded in and AU, respectively; Estimate the number of allocated bits allocated to the uncompensable k-th MB. For each input MB, use its SAD 値 and the number of allocated bits to find the dynamic bit rate comparison table. To get the best quantization parameter βP 値, and readjust βP 値 so that the difference between it and the previous MB's βP 値 is not greater than 2; use this quantization parameter to inverse M B performs quantization and encoding, and updates the dynamic bit rate comparison table by using the actual encoded bit count after encoding. 2. The method as described in item 1 of the scope of patent application, wherein the dynamic bit rate comparison table is a two-dimensional array: b [SAO] [QP], the first parameter SAD in the array, which is not a reference An integer between the (SADmin, SADmax) range. The second parameter (2P in the array, which represents the quantization parameter ρ / M, 2, ...., 31. The elements of the array represent a SAD.) The estimation of MB's coded bit count. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. 3. The method described in item 2 of the scope of patent application, where the dynamic bit rate comparison table is constructed according to the following steps. : First of all, input a sequence of video pictures in block MB as a unit, input them to a video encoder in sequence, calculate the SAD 値 of input MB, and set! GP = 1, 2, ..., 31, respectively. Quantized coding; Record the actual number of coding bits after each quantized coding of these input MBs; Repeat the previous steps to process all MBs, and generate 14 paper sizes from all MBs of the same SAD and QP Applicable to China National Standard (CNS) A4 (210 x 297 mm) ) 569625 A8 B8 C8 D8 6. The actual number of coded bits in the scope of the patent application, that is, each b (SAD, βP) element is averaged, and then stored in the 6 [SADMBk] [eP] array in order, that is, constructed The comparison table 4. The method as described in item 1 of the scope of patent application, wherein the target code bit count Bτ of the current picture is estimated according to the following formula: R «M ··· Λ F, where Δ is according to the following Formula calculation: 2 * W ——, W > Z · M Δ = F W- ZZ · M, others, where ζ is set to · 1, D-ΙΪ / Ρ, 〇), D means before When encoding a picture, the actual number of bits used refers to the fullness of the previous picture buffer, and refers to the channel bit rate, and F refers to the picture bit rate. 5. The method described in item i of the scope of patent application, wherein the number of bits allocated to the non-compensable k-th MB is calculated according to the following formula: bkz SAD prepared female N Σ SAD MBk k = l (Please read first Note on the back, please fill in this page again) Order --------- Line · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs, where the total number of bits allocated to non-compensable MB by the representative, SAD Satoshi It represents the SAD of the first MB, and N represents the total number of non-compensable MBs in the picture. 6. The method described in item 5 of the scope of patent application, which is estimated using the following formula: Bn Bcode-Bh-intra Nintra —Bh-inter * Ninter 丄 where represents the total number of bits allocated to the non-compensable MB; Represents the number of header bits for internal coding. Ma " is the number of header bits required to represent one mb of the parent's mutual encoding; Li · 7 ^ represents the remaining MB number for internal encoding; iv ^ r refers to the need to interactively edit 569625 A8 B8 C8 D8 6. Remainder of patent application scope code] MB _ stomach. , 7 The method as described in item 6 of the scope of the patent application, where Fan Ru, and I " Gamma system can be obtained by iterative calculation with the following formula: Bh-L · jl Q_1) + b h-intra. N Winter = — (Bh4nter * (j ^ l) + ^ inter. 'The center_ represents the average header bits of ^ obtained by using ... MBs (the first MB to the jth MB) as the internal encoding] The number represents the number of header bits used by the j-th MB to use internal encoding. Knowing * represents the average header bits obtained by using the number of MBs (the first MB to the j-th MB) as the interactive encoding. The number of bits represents the number of bits in the header of the j-th MB using interactive encoding. 8. The method described in item 6 of the scope of the patent application, in which the total number of bits that can be used for the non-compensable MB is estimated. Calculated according to the following formula: (Please read the notes on the back before filling out this page) GOBH Printed Bcode Βτ-Buncode-ΒρΗ-B by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs, where w is the position of the screen header Yuan count, tolerance is a bit count representing the head of a picture gob. 9. The method described in item 8 of the scope of patent application The number of data bits in one of the pictures that is not in this article can be calculated according to the following formula: Μ ^ uncode- Σ (Bq〇D y Bc〇D + BmCBPC ^ BcBPV ^ BmVD f ^ MVDy〇) m = l 16 This paper size applies to China National Standard (CNS) A4 (210 x 297 mm) 569625 A8 B8 C8 Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs, Consumer Cooperatives, ^, and Shen Jing Patent Scope, where 1 ^ represents the total number of MB in a picture; (X, Y, Z, 0) represents X, γ, B according to the coding mode. And 〇, where X and γ correspond to the compensable type, z corresponds to the non-compensable interactive code '0 corresponds to the non-compensable internal code, and ^ represents the bit of the COD signal The number is the number of bits representing the MCBPC signal, the number of bits representing the CBPY signal, and the number of bits representing the MVD signal. The method as described in item 1 of the scope of patent application, wherein, for each input MB, using its SAD 値 and the number of allocated bits ^, look up the dynamic bit rate comparison table so that it is similar to Fusinaiwei] The difference between 2 is the smallest. At this time, the corresponding quantization parameter ρρ is the best quantization parameter. Adjust OP * so that the difference from the previous MB is not greater than 2 and record it in Z) 2Κ4ΑΤ in. 11. The method described in item 1 of the scope of patent application, wherein the actual encoding bit count of the current MB ~ 'is used to update the number of encoding bits in the lookup table, and each quantization parameter in the dynamic bit rate lookup table The estimated number of coding bits is updated according to the following formula: The updated estimated number of coding bits = 6 [SADMBk] [βΡ] + Shenzhen SADMBk], where Shenzhen SADMBk] is a one-dimensional shift array ), The size of this array is 1 x SADmax, that is, each SAD corresponds to a shift array element, and SADMBk] is updated according to the following formula: s6 [SADMBk] [0 户] = zhenSAD] \ iBk] + ( ~ ^ [SADMBkn ^ 3] -ZhenSAD] MBk]) / 2 0 12. The method described in item 1 of the scope of patent application, wherein if the processing occurs at the beginning of a GOB, the picture quantization will be calculated according to the following steps If there is no non-compensable MB in the GOB, it is set to any integer in the range of 1 to 31. If there is an uncompensable MB in the GOB, it is set; if there are at least two non-compensable MBs in the GOB, Compensation MB will be set according to the following formula: QPi + 2 if QP ^ QP! ≫ L DQUANT = Pi if -h < QP2-QPi < l ^ 17 This paper size applies to China National Standard (CNS) A4 (210 x 297 mm) ^ -------- Order --------- Line · (Please read the precautions on the back before filling this page) 569625 A8 B8 C8 D8 VI. Patent application scope QP2 2 if QP2-QP1 <-L where L is a positive integer and the default value is L = 5. (Please read the precautions on the back before filling out this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 18 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)