JP2001526850A - Method and apparatus for optimizing a quantization value in an image encoder - Google Patents

Abstract

(57) Abstract The quantizer controller specifies a target bit value equal to the total number of bits available for encoding a frame. The total distortion in the frame is modeled according to the predicted quantization value of each block. The predicted quantization value is characterized according to the amount of energy in each block and the number of bits available to encode each block. The quantization value is optimized for each block by minimizing the modeled distortion in the frame, subject to the constraint that the total number of bits in the frame encoding is equal to the target bit value. Thereafter, each block is encoded with the optimal quantization value.

Description

DETAILED DESCRIPTION OF THE INVENTION             Method and apparatus for optimizing a quantization value in an image encoder Background of the Invention   The present invention relates to quantization used for coding coefficients of digital image and moving image frames. Regarding the calculation of the value, more particularly when the encoding is performed with a limited number of bits Calculated quantization to reduce distortion in digital image and video frames Relating to optimizing values.   In many modern image and video encoders, the quality of the encoded image is one or more. It is controlled by selecting the above quality parameters. Block based Image and video encoders use a quantization scheme for each block of pixels in the image. A parameter known as the rule or quantization step is used. Quantization The step scales pixel values for the same value within the same step range. Used to Image blocks coded on the same quantization scale They have the same quality. The number of bits required to encode the image It depends on the image quality (quantization scale) and the intrinsic statistics of the image. As a result, Separate images encoded on one scale (same image quality) have different bit counts Occupy.   In many applications, encoding one or several frames Generates the target number of bits because the number of available bits is fixed in advance A quantization scale that encodes video frames with the highest possible quality Some technology is needed to do that. For example, in digital video recording In order to improve fast-forward and fast-reverse performance, the number of frame groups (GOP) Must occupy a bit. For videophone, channel rate, communication delay and For one or more frames, depending on the The number of usable bits is determined.   Existing quantizer or buffer control methods are mainly classified into three types. In the first type of quantizer control method, each image block is set with a set of quantization scales. Encode the mark several times. The number of bits generated in each case is measured and Scale is well selected, so the sum of the bits of all combined blocks The number will be the desired number of target bits. The first type of quantization control technology is rear Cannot be used for real-time encoding. To encode each image many times, This is because a high degree of complexity is required.   The first type of quantizer control is described in the following publications:   A second type of quantization control technique is used on previously encoded image blocks. The number of bits measured and other parameters such as buffer fullness and block activity Measure the activity. These measurements determine the quantization scale in the current block. Used to select. The second type of quantizer control is of computational complexity. It is common for real-time coding, to a lesser extent. However, the second The two types of quantizer control are inaccurate when achieving the target number of bits To avoid bit and buffer overflows and underflows, Coding techniques must be added and combined.   The second method is described in the following publication   In the third type of quantizer control, the number of bits required to encode an image block is determined. Use a predictive model. This quantizer model contains the quantization scale of the block. And other parameters such as block variance. The quantization scale is It is determined by some mathematical optimization of the encoder model. The third type of quantizer control is , It is computationally simple and can be used in real time, but it is extremely High sensitivity often results in inaccurate results.   A third type of quantizer control is described in the following publications:   Thus, the time and computational complexity required to generate the optimal quantization value is reduced. However, there is still a need to improve the image quality of quantized images or video frames. Remaining. Summary of the Invention   The quantizer controller uses a new block-adaptive Lagrangian optimization method, Generate a quantized value. The quantizer controller has been updated to determine whether previously quantized blocks It is improved using these information. The computational complexity of type 2 quantizer control technology is While mitigated, this quantizer controller is coarse to model errors. Therefore, the result is generated with the same level of accuracy as the type 1 quantizer control technique.   Its quantizer controller equals the total number of bits available for encoding the frame Specify the target bit value. The total distortion in the frame is divided into blocks. Modeled according to the applied prediction quantization value. The predicted quantization value is The amount of energy in the lock and the number of bits available to encode each block Therefore it has been characterized. The optimal quantization value is the total number of bits for encoding the frame. Subject to the constraint of being equal to the target bit value, Apply to each block by minimizing the modeled distortion. And , Each block is encoded with the optimal quantization value.   By reducing the quantization value for blocks with low energy, By increasing the quantization value for blocks with large Can be applied to each block. The quantization value assigned to the block also The number of image blocks to be encoded and the remaining It is optimized according to the number of available bits.   Some weighting factors can change the accuracy of the encoded block Arbitrarily applied to One weight according to the position of the block in the frame A weighting factor is applied to the quantized value. Each frame in multiple frame groups The optimal quantization value is applied to the blocks in Applies to any area inside.   The quantizer controller encodes the image only once and calculates the quantization value of each block with accuracy. Produce well. The quantization value is the target bit for the encoded image or video frame. Generate a number. Therefore, this quantizer controller uses a quantizer control technology of similar accuracy. Requires less computational consumption.   In the general configuration of the quantizer controller, various quantization / rate control measures are used. Can be used. For example, the quantizer controller may use the current video coding standard MP EG1-2, 4, H.E. 261, H .; 263, H .; 263+ Quantization for DCT (Discrete Cosine Transform) based coding of It can be used to select the value of the scale in real time. One frame System, some frames or some macroblocks within a frame It is encoded by the number of bits.   The above and other features, advantages, and objects are described below for preferred embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS The detailed description will be more apparent with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE FIGURES   FIG. 1 shows a plurality of blocks to which an optimal quantization value is assigned according to the present invention. FIG. 3 is a schematic diagram of a plurality of image frames included in the image frame.   FIG. 2 is a block diagram of an image encoder according to one embodiment of the present invention.   FIG. 3 is a flowchart for generating an optimal quantization value.   FIGS. 4 and 5 are diagrams showing the results of applying the optimal quantization value to the image data.   FIG. 6 is a block diagram of a quantizer controller according to one embodiment of the present invention. Detailed description   The present invention will be described with reference to a block-based image encoder 12. Is used to control the quantizer of any image or video encoder. Can be.   Referring to FIG. 1, an image encoding method based on a block 1 5 are transmitted in a plurality of frames. Each frame 26 is of the same size (typically (16 × 16 pixels per block). Number of bits B generated after encoding the i-th image block 14_iIs quantized Parameter Q_iAnd the block's statistics. For example, image bro The block i = 9 has more image information (energy) σ than the image block i = 17._i Contains. This is because the image of block i = 9 is a part of the face image together with the background information. Because it contains minutes. Conversely image blocks_i= 17 is image information energy σ_iIs reduced. This has a substantially identical background image at each pixel location. This is because that.   Referring to FIG. 2, the pixel value of each image block 14 is, for example, a block transform 1 At 6, a discrete cosine transform can be used to convert to a set of coefficients. these Are quantized by the block quantization 18 and encoded by the encoder 20. Mark Bit B of the encoded and quantized image block 14_iIs a telephone line, micro Transmission to a receiver (not shown) via a communication channel via a wave channel or the like Is done. The receiver consists of a decoder that decodes the quantized bits and an inverse discrete cosine And an inverse transform block for performing transform (IDCT). Decrypted bit B_i Is displayed on the visual display screen to the user.   Quantization of the coefficients transformed by the quantization block 18 is a key process. To Because this determines the quality at which the image block 14 is encoded It is. The quantization of the i-th block 14 is based on the parameter Q_iIs controlled by H. 261 and H.E. In the H.263 video coding standard, Q_iIs the amount of the i-th block Known as the quantization step, its value is used to quantize the transformed coefficients. Is equivalent to half of the step size used for. MPEG-1 and MPEG-2 standards Then, Q_iIs called a quantization scale, and the j-th coefficient of a certain block is Step size Q_iw_jIs quantized using Where w_jIs an MPEG code The j-th value of the quantization matrix chosen by the deceased person of the deck.   Let N be the number of 16 × 16 image blocks in one image frame 26. One The total number B of bits available to encode the image frame 26 of   Where the value of B is the selected quantization parameter Q₁, Q_Two,, Q_NAnd the above Depends on lock statistics. The present invention comprises a quantizer controller 22 (of FIG. 2). Which is the limited total number of bits B available to encode frame 26. For Q_iChoose the optimal value of. The quantizer controller 22 can have many different maps. Software (programmable processing unit software with dedicated hardware) ).   In image coding, image blocks 14 are intra-coded or intra-classified. Is said to be. In video coding, many of the blocks 14 in a frame 26 Is quite similar to the previous frame. The pixel values of block 14 were previously encoded Blocks are often predicted, and only the differences or prediction errors are encoded. this These blocks are said to be inter-coded or inter-class. The invention can be used in frames with both intra and inter blocks. it can. Encoder model   The next model given in equation (2) specifies the number of bits given to the i-th block I do. Q_iIs the quantization step size or quantization scale, and A is the Number of pixels (for example, A = 16 in MPEG and H.263)^TwoPixel) And C are constants, σ_iIs the empirical standard deviation of the pixels in the block and is given by . Value P_i(j) is the j-th pixel of the first block, and P_iIs the pixel value of the block. It is an average and is expressed by the following equation. P for color images_i(j) is the luminance and color component values of each pixel. Equation (2 The model in () is derived using the rate-distortion analysis of the block encoder. formula( The value of K in 2) is the statistical value of the image block 26 and the quantization used in the encoder. Depends on the matrix. For example, if the pixel values are almost uncorrelated and Gaussian If the distribution is followed, the quantization matrix will be flat with uniform weights (ie, all W for j_j= 1), and K = π / ln2. The constant C in equation (2) is Modeling the average number of bits per pixel used to encode the encoder addition To For example, C indicates header and syntax information, pixel color or color component, Sent to receiver for decoding image blocks such as Q value, motion vector, etc. This is to explain what is done. If the values of K and C are unknown, the following invention technology It is estimated in "update of parameters of encoder model".   Equation (5) models the distortion D of N coded blocks.Where α_iIs a weight chosen according to importance or cost of blockiness. For example, the larger α_iIs used for images with artifacts that are more visible to humans. On the other hand, select image blocks belonging to more important objects in the scene. Selected. If α₁= Α_Two= ... = α_N= 1, the distortion represented by equation (5) is The mean square error (MSE) between the original and the encoded block. optimisation   The quantizer controller 22 (FIG. 2) calculates the optimal quantum that minimizes the distortion model of equation (5). Chemical value Q₁ ^*, Q_Two ^*,…, Q_N ^*Select This is because the total number of bits is defined by equation (1). According to the constraint that it must be equal to B, mathematically Is expressed as follows. The next purpose is each Q_i ^*Is to find the formula. To do this, Lagrange (6a) is used to transform the minimization constrained by equation (6a). become. Here, λ is called a Lagrange multiplier. Next, B in equation (6b)_iExpression ( By using 2), the following equation is obtained. Finally, in Equation (6c), if the partial derivative is set to 0, the i-th image block For the optimal quantization step size, the following equation is derived. Further, if i-1 block 14 is already quantized and encoded For example, the optimal quantization parameter of the i-th block is represented by the following equation. Where N_i= N-i + 1 is the number of image blocks remaining to be encoded , Where B_i-1Is the optimal quantization value Q using equation (2)._i-1 ^*It was obtained in. Therefore, equations (6) and (7) can be expressed as follows when the number of available bits is limited. This is to generate an optimal quantization value that minimizes distortion. As a result, the same bits 1, the image of frame 26 is displayed on the receiver end of channel 21 in FIG. When displayed on the screen, the distortion is lower than in other quantization schemes. Quantizer control method   FIG. 3 illustrates the steps performed by the quantizer controller 22 (FIG. 1). , The quantization value used to encode N image blocks 14 with B bits is selected Is done. Note that N is an image block, image part, some images, or general May be the number of any region of the image. Step 1: Receiving and initializing the energy value   In step 1A, the pixel values of the N image blocks are compared to the digital image (FIG. 2). Are supplied to the quantizer controller 22. In step 1B, i = 1 (first block Then, initialization is executed. Set (eg, α₁= Α_Two= ... = α_N= 1 to minimize the mean square error. Make it small). As one example, the energy amount σ_iIs calculated by the conversion block 16 It is derived from the DCT coefficients of the generated pixel values.   In the fixed mode, the values of the parameters K and C in the encoder model of Equation (7) are already set. Known or pre-estimated. For example, by using linear regression,₁ = K and C₁= C. Since the model parameters are not known in adaptive mode, K₁And C₁Is set to some small non-negative value. For example, according to experiments, K₁= 0.5 and C₁= 0. In video encoding, K₁as well as C₁Is K from each of the previously encoded frames_{N + 1}And C_{N + 1}Set to a value Can be. Step 2: Calculate the optimal quantization parameter of the i-th block   When the value of the Q parameter is a fixed set (eg, in H.263, QP = Q_i / 2, take values of 1, 2, 3,..., 31)._i ^*Box's Rounded to the nearest value in the set. Next, using the index table, the square root operation is It is performed as follows. Step 3: Encode i-th block with block-based encoder   B_i'Is the number of bits used to encode the i-th block. You. Step 4: Update the quantization value   In step 4, the parameter K_{i + 1}And C_{i + 1}Is updated in the quantizer controller 22 Is done. In fixed mode, K_{i + 1}= K and C_i-1= C. In adaptive mode, new K_{i + 1}And C_{i + 1}Are found using model fitting techniques. Model For details on the fitting technique, refer to the following "Update of parameters in the quantizer controller". This will be explained in the section. Step 5: Generate Quantized Value for Next Block   If i = N in decision step 5, the quantized value is Has been derived. Then, the quantizer controller 22 is stopped. All images If block 14 is not quantized, quantizer controller 22 determines in step 6 Receiving the coefficient of the next image block i = i + 1, skipping and returning to step 2 . The quantized value for block i = i + 1 is derived as described above.   4 and 5, the quantizer controller 22 controls the video sequence The frame is encoded. Telenar H. 263 compared to the offline method However, this method is a quantizer control technique adapted to the MPEG-4 anchor. You.   The total number of bits per video frame obtained by the quantization technique described in FIG. 4 is shown by a solid line. H. The H.263 off-line encoding technique is indicated by the dashed line . The 133 frames of the famous video sequence "Foreman" were coded. Ta The target bit number B is 11,200 bits per frame. B = 6400 5 except for the 140 frames of the video sequence “Mother and Daughter” Same as 4.   The quantizer controller 22 generates a very accurate and stable number of bits per frame. To achieve. Similar results were obtained over a wide range of bit rates. In the experiment, There was little visible difference in quality between the two encoded video sequences. amount The S / N ratio of the image processed by the densifier controller 22 is slightly on average 0.1 to 0.3 d. B was only low. Therefore, even if an image is encoded only once, The controller 22 controls the target bit rate for every frame with high image quality. It can be achieved with high accuracy. Other execution methods   Some quantization variables are based on the basic quantization optimization framework described above. ing. In step 1B of FIG._kIf you can not calculate in advance Estimate, for example, S from the previous video frame 26₁The value is used.   To further reduce the computational complexity, a less complex S₁Using the prediction of Can be. For less complex estimation, replace equation (3) with equation (9). Here, abs (x) is the absolute value of x. In video coding, interblock Equation (6) instead of equation (7) can be simplified and for fixed optimization Is used to select a quantization parameter. To do this, in step 3 of FIG. , Are replaced in the following order.   In the case of a variable rate channel, the available video after encoding i blocks   The quantization model defined by the equation (2) is generalized to the following equation (10). be able to. Where A_j, J-th area. The area of quantization is a block. Need not be. Additional model parameters φ and γ are set beforehand for quantization. Or obtained using a parameter estimation technique described below. Equation (10) quantization If the model is used in step 2, the optimal quantization value Q_i ^*Uses equation (11) Is derived.In step 1, S_iIs replaced by: In step 3, S_{i + 1}Is the expression Is replaced by Intra and inter block coding   Some of the blocks to be coded are of the intra (in the same frame) class, If some are inter (between different frames), the performance of the quantizer controller 22 And so on. The factor β is as follows. K_IAnd K_PIs the average value of K measured in the intra and inter blocks, respectively. It is. The value of β is estimated and updated during encoding. During the experiment, using the constant β = 3 Then, it turned out that it works well. Frame-based quantization control   The quantization step is fixed for all the blocks 14 in the frame 26. The same quantizer controller 22 shown in FIG. Can be used to encode the system. Parameter N = number of frames, B = the number of bits available to encode N frames, i = frame number in video sequence, Q_i= Quantization step for all blocks in i-th frame, A = the number of pixels in the frame.   Parameter α_i, Σ_i, B_i'Is the weight, variance, and bit of the i-th frame, respectively. It is. Parameter K_iAnd C_iUpdated the encoder model for the above frame Things.   If the computational complexity is not an issue, each image block 14 is encoded several times, And fitting procedures for classical models (eg, least squares fit, linear regression , Etc.) and use K_iAnd C_iA good estimate of the You. Therefore, in step 2, the quantized value Q_i ^*Is determined by the following equation. Can be set to avoid calculating and updating this model parameter Can be. In such a case, Q_i ^*Can be easily expressed by the following equation. You. Or, equivalently, it can be expressed by the following expression. Any subset of some of the techniques described in "Other Implementation Methods" can be combined. It can be used together with or. Update encoder model parameters   The following is the K in the quantizer controller 22_{i + 1}And C_{i + 1}One to update the parameters of Technology. This update technique is used in the adaptive mode described in step 4. . Pond classical parameter estimation or model fitting techniques, such as least squares , Recursive least squares, Kalman prediction, etc. can be used instead. Model Parameters can be specified for any block, frame, block group, or frame group. Can also be updated.   The model parameters in one embodiment of the present invention use the following weighted average: And is updated or estimated on a block-by-block basis. Using the values of K and C, B_i'Is expected. Instead, some coordinates Tsu Where B '_{DCT, i}Are the bits used for the DCT coefficients of the i-th image block Is a number. is there. If the general model is used in equation (10), various estimators may be added with additional parameters φ And γ can be used to estimate These parameters are also based on each block. Updated in the order, and the i-th update φ_iAnd γ_iIs K_I, C_iAveraging technique similar to that of Can be found using Weight α_ISelection of   α_iCan be chosen in relation to the weight or importance of the blockiness. Α as default value₁= Α_Two= ... = α_N= 1, the MSE distortion is Minimum between the null and the coded block. Otherwise, the MSE distortion Miha α_iDecreases when is large, α_iIncreases when is small. Weight α_iTwo examples of choosing This will be described below. In an area like a rectangular window in a videophone image, Large α_iThe value is assigned the next smaller quantization value. Small quantization scale Weighted regions are encoded with better quality because the pixels are used to quantize pixel values. Be transformed into   Typically, humans pay more attention to the central region of the picture. Therefore, a larger α_i Is assigned to an area near the center of the picture. Pyramid formula is used, Larger α for blocks closer to the center of the frame_iCan be assigned. In particular B_XAnd B_YIs the number of blocks along the horizontal and vertical coordinate axes, respectively. . The weight of the i-th block is calculated by the following equation. Here, (a₁+ A_Two) And a_TwoIs the height and side branches of the pyramid, b_XAnd b_Y Are the horizontal and vertical positions of the block in the frame, respectively. An example For example, a₁= 15 and a_TwoIf you choose = 1, α of the central block_iValue is the bounding block 16 times of Block join   In the codec, Q₁ ^*, ... Q_N ^*The quantized value using (see step 2) Needs to be sent to the decoder. For example, H. At 263, the quantization value is Are encoded in the order of the pixel scan, and a 5-bit penalty is required to change the quantization value. Ah You. At high bit rates, the addition of a change in quantization value is negligible, The technology is effective. However, at very low bit rates, this addition is Some techniques are needed to limit the number of times the quantization value changes Become. Unfortunately, existing optimization methods that take account of additions are mathematically inaccurate or Highly costly.   Another feature of the present invention is that a block of similar standard deviation Combined into one set, so the quantization values remain constant within that set . This technique is called block combination, and changes the quantization value at a low bit rate. Is to reduce. Weight α_iIf you choose the value of It is. Where B / (AN) is the bit rate of the current frame (bits per pixel) It is. The values of B, A, and N are the number of available bits and the This is already defined as the number of pixels and the number of blocks. If the bit rate is 0. If greater than 5, α_iAre all equal to 1 and thus have no effect. Low bit At the rate, α_iIs linear with each σ_iApproaching, gradually Q^*Reduced range Let Actually, when the bit rate is 0, α_i= Σ_iAnd all quantization values Are the same, so all blocks are combined into one set.   FIG. 6 is a detailed block diagram of the quantizer controller 22 of FIG. One implementation State quantizer controller 22 executes on a general purpose programmable processor. It is. The functional blocks in FIG. 6 represent the main operations performed by this processor. You. The initialization parameters in block 31 may be from pre-processing of the current image or from previous From the parameters previously derived from the frame or from the processor memory (Fig. (Not shown) in any of the previously stored values. Initialization parameters Has N₁, S₁, B₁, K₁And C₁(Or K from the previous frame_{N + 1}And C_{N + 1}) Is included.   The image is divided into N image blocks 14 (FIG. 2) of the pixel A in the block 30. Divided. The energy of the pixels in each block is calculated in block 32. each The weighting factor assigned to the block is calculated at block. image The amount of energy remaining is updated in block 40 to encode the image Are updated in block 38. Encoder model parameters Is updated in block 42 and the number of blocks remaining for encoding is Be organized. The processor of block 36 comprises blocks 32, 34, 40, 38, According to the values derived in steps 42 and 44, the above-mentioned optimization step size is calculated.   Having described and illustrated the principles of the present invention in a preferred embodiment, It is clear that changes can be made to the equipment and details without leaving It is easy. All changes and changes are within the scope and spirit of the following claims. It is a thing to lay.

────────────────────────────────────────────────── ─── Continuation of front page    (81) Designated countries EP (AT, BE, CH, DE, DK, ES, FI, FR, GB, GR, IE, IT, L U, MC, NL, PT, SE), CN, JP, KR, S G

Claims (1)

[Claims] 1. A method of encoding a plurality of blocks in a frame of image data, comprising: specifying a target bit value equal to a total number of bits available for encoding the frame; and assigning a frame with a quantized value assigned to each of the blocks. Predicting the total distortion in, the quantized value characterized by the amount of energy in each block and the number of bits allocated to each block, the total number of bits available for coding the frame being equal to the target bit value Applying the optimal quantization value to each of the plurality of blocks by minimizing the total predicted distortion in the frame according to the constraint that the block is encoded with the predicted optimal quantization value. An encoding method for a plurality of blocks. 2. The encoding method according to claim 1, wherein the optimal quantization value is generated using a Lagrangian optimization method for the predicted total distortion. 3. The encoding method according to claim 1, wherein the optimal quantization value is represented by a formula: Here, Q _i ^* is the optimal quantization value of each block i, N is the number of blocks in the frame, B is the total number of bits available to encode the frame, and A is the number of bits in each of the plurality of blocks. The number of pixels, K and C are constants associated with the image block, σ _i is the empirical standard deviation of the pixel values in the block, α _i is a weight incorporating the importance of the block, Characteristic encoding method. 4. 2. The encoding method according to claim 1, wherein the optimal quantization value is adjusted according to the number of image blocks remaining to be encoded and the number of bits still available for encoding the remaining image blocks. An encoding method comprising: 5. The encoding method according to claim 3, wherein a K parameter for modeling a correlation statistic of a pixel in an image block for each block to adjust the optimal quantization value of each of the plurality of blocks; and additional data. And C parameters for modeling the bits required to encode. 6. The encoding method according to claim 5, wherein the K and C parameters are known in advance, or an adaptive mode in which the K and C parameters are derived according to the K and C parameters of a previously encoded block. An encoding method, which comprises deriving the optimal quantization value. 7. 7. The encoding method according to claim 6, wherein the adaptive mode comprises: deriving values of the K and C parameters that accurately predict the number B of bits used for encoding a previous block; Deriving the average of the K and C parameters derived for the transformed video block; and linearly weighting the average of the K and C parameters according to the initial estimation of the K and C parameters, Estimating K and C parameters for a video block. 8. The encoding method according to claim 7, wherein the values of K and C for predicting B are: Or Where B ′ _DCT , _i is the number of bits used for the DCT coefficients of the current image block, and the average of K and C is The linearly weighted average of K and C is An encoding method characterized in that: 9. 4. The encoding method according to claim 3, wherein the amount of energy of the frame is not determined beforehand, here A coding method, characterized in that α _K and σ _K are estimated according to: previously obtained video frames obtained for the block. 10. In the coding method according to claim 9, and several coding the image blocks, the parameter K ₁ for each image block, K _2, ... K _n and C _1, C _2, the ... C _n Estimate and then the equation By deriving a super-optimized quantization value. 11. 2. The encoding method according to claim 1, wherein a plurality of frame bit values equal to the total number of bits usable for encoding the plurality of frames are specified; and a quantization value is assigned to an energy amount of each frame and each frame. Characterizing according to an assigned number of bits and modeling a total distortion amount of the plurality of frames according to the quantization value assigned to each of the frames; minimizing a total distortion amount modeled in the plurality of frames. Estimating the optimal quantization value for each frame, and encoding each frame with the predicted optimal quantization value. Characteristic encoding method. 12. The encoding method according to claim 1, further comprising: applying a weighting factor to each of the optimal quantization values according to a position of the block in the frame. 13. 2. The encoding method according to claim 1, further comprising: controlling a variety of optimal quantization values assigned to the blocks by assigning the same quantization value to blocks having similar standard deviation values. Encoding method to be used. 14． The encoding method according to claim 13, wherein the optimal quantization value is a weight value. Where B / (AN) is the bit rate in bits per pixel of the current frame, B is the number of available bits, A is the number of pixels in the block, N is the total number of blocks in the frame, σ _i Is controlled by allocating the standard deviation of the pixels in the block. 15. In a method for quantizing a region of a moving image, receiving image information for different regions, predicting a distortion amount generated in the moving image based on a quantization value assigned to the region, and predicting distortion in the region. Characterizing the amount of information in the region and the number of bits available to encode the information in the region using the quantized value, optimizing the quantized value assigned to the region, A quantization method comprising: minimizing the amount of prediction distortion with respect to the number of usable bits; and encoding the region with an optimal quantization value. 16. The quantization method according to claim 15, wherein the optimal quantization value is: here, Where γ, φ, K, and C are constants, A _i is the number of pixels in the i-th area, σ _i is the energy of the pixel value in the i-th area, B _i is the number of available bits, and α _i is the area A quantization method, which is derived by a weighting factor incorporating the importance of distortion. 17． 16. The quantization method according to claim 15, wherein the optimal quantization value of a selected region is: summing the energy in each region to determine the total energy of the moving image; and calculating the total energy in the selected region. Multiplying the multiplied energy by a scaling factor, scaling the multiplied energy according to a scaling factor, squaring the scaled energy, and thereby deriving the optimal quantization value for the selected region. Method. 18. 18. The method of claim 17, wherein the scaling of the multiplied energy comprises applying a first scaling factor that is proportional to the number of regions in the frame remaining to be quantized. Applying a varying second scaling factor to each region according to the total number of bits available for encoding and the total number of bits already used to encode a previous region in the frame. Quantization method. 19. 20. The method of claim 18, comprising applying third and fourth scaling factors, wherein the third scaling factor is modeled with a correlation statistic within the region, and wherein the fourth scaling factor is Is a quantization method characterized by being represented by additional data in an encoded frame. 20. 20. The method according to claim 19, comprising applying a fifth scaling factor proportional to the number of pixels in each region. 21. 16. The method according to claim 15, wherein the energy in each region is proportional to a standard deviation of pixel values or an average of all pixels in the same region and a sum of absolute values of associated pixels. Quantization method. 22. 16. The quantization method according to claim 15, further comprising: predicting a total energy in a moving image by acquiring a total energy of a previous moving image. 23. 16. The quantization method according to claim 15, wherein the predicted optimum quantization value is reduced when the block has a small energy, and is increased when the block has a large energy. . 24. 22. The quantization method according to claim 21, wherein the energy is adjusted by a scaling value of a pixel in the intra coding region, the scaling value according to a different K value representing a correlation statistic of a different type of coding region. A quantization method characterized by being characterized. 25. An encoder for quantizing a region of a moving image frame, a circuit for detecting an amount of moving image information of the one region, and a quantization for minimizing an amount of prediction distortion in the moving image frame with respect to a target bit value. Assigning a value to each region, predicting the amount of distortion generated in the video frame before the information in the region is actually quantized, and adjusting the quantization value assigned to each region , A quantizer controller that minimizes the prediction distortion, subject to a constraint that the total number of bits available for frame encoding is equal to the target bit, associated with the region generated from the quantizer controller. A quantizer that quantizes the moving image information in the area according to an adaptive quantization value. 26. 26. The encoder according to claim 25, further comprising: a transform circuit that receives the moving image at an input and generates a transform coefficient at an output, wherein the quantizer is configured to generate the transform coefficient according to an associated quantization value. An encoder characterized by quantizing transform coefficients.