JP2007074337A - Coding device and method therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To resolve the issue that coding amount estimation cannot be precisely performed because probability distributions of symbols are different by quantization parameters, when a generated code amount after arithmetic coding is acquired by temporary coding with a certain quantization parameter and then coding is performed with another set quantization parameter, with respect to coding amount control using arithmetic coding. <P>SOLUTION: This coding device includes a context selection circuit 131 and an adaptive quantization control circuit 130. When a proper quantization parameter input to a quantization circuit 120 for coding is obtained from a generated code amount outputted from an arithmetic coding circuit 140 by entering an arbitrary quantization parameter to the quantization circuit 120 for temporary coding, the circuit 131 selects contexts of arithmetic coding per quantization parameter in the case of temporary coding and selects the same context independently of quantization parameters in the coding, and the circuit 130 generates a quantization parameter having a value periodically changed, in the temporary coding and generates a proper quantization parameter in the coding. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an encoding apparatus that improves the accuracy of code amount control of an encoder to which arithmetic encoding is applied, and performs encoding with high efficiency and high image quality.

  Arithmetic coding is a technique that can reversibly compress the amount of information to the theoretical limit in accordance with the occurrence probability of the information source symbol. In the field of image coding, the JPEG2000 standard (ISO / IEC15444), H.264, etc. H.264 / MPEG4-AVC standard (see Non-Patent Document 1) and the like. H. In H.264, high-efficiency coding according to the probability characteristics of the syntax is realized as context adaptive arithmetic coding (CABAC).

  H. The context adaptive arithmetic coding in H.264 will be described with reference to FIG. FIG. 9 is a diagram for explaining the arithmetic coding circuit. In FIG. 9, the arithmetic coding circuit 140 includes a binarization circuit 510, a context calculation circuit 520, and a binary arithmetic coding circuit 530. The operation of this circuit will be described below.

  A binarization circuit 510 binarizes multi-value input data 501 such as encoding information such as flags and conversion coefficient data. In the binarization, a syntax element (syntax element) which is a type of input data is determined from the control information 502, and unary binarization or fixed-length binarization (fixed-length binarization) is performed according to the probability characteristics of the data. ). The binary arithmetic coding circuit 530 arithmetically codes the binarized binary string 503 and outputs a binary string 509.

  Based on the control information 502 indicating the syntax element, the context calculation circuit 520 determines the value of the context index (ctxIdx) used for encoding the current 1 bit in the binary string 503 from Table 1 defined by the standard. Ask.

  From Table 1, ctxIdx can take 460 values from 0 to 459. For one syntax element, the selection of ctxIdx differs depending on the slice type. In addition, for a syntax element in which a plurality of ctxIdx values exist, for example, ctxIdx of mb_field_decoding_flag in slice type I is 70-72, a rule determined for each syntax element (non-patent literature) 1), the value of ctxIdx is uniquely determined from a plurality of values.

  The context calculation circuit 520 initializes and stores occurrence probability information for each of 460 ctxIdx values. The occurrence probability information is a set of an MPS indicating a symbol having a higher occurrence probability of binary symbols 0 or 1 and an occurrence probability pState. The occurrence probability information corresponding to the obtained ctxIdx value will be referred to as context information 504.

  The context calculation circuit 520 generates context information 504 and inputs it to the binary arithmetic encoding circuit 530. As described above, the occurrence probability of a binary symbol is dynamically changed by adaptively switching the occurrence probability of a symbol input to the binary arithmetic encoding circuit 530 by the syntax element for performing arithmetic encoding, that is, the context information 504. Optimal arithmetic coding can be performed on the binary string 503.

  The operation of the binary arithmetic coding circuit 530 will be described with reference to FIG. In FIG. 10, a binary string {0, 0, 0, 1} is input from an information source having a symbol 0 occurrence probability of 0.75 (binary 0.11) and a symbol 1 occurrence probability of 0.25. Consider the case. In this case, the current context information 504 is represented by MPS = 0 and occurrence probability pState = 0.11 indicating the symbol having the higher occurrence probability. pState is expressed as a normalized integer value, but here it is assumed to be a binary probability value for simplicity. The context information 504 is updated by the current binary arithmetic coding and returned to the context calculation circuit 520. The context calculation circuit 520 updates the occurrence probability information of the corresponding ctxIdx. This value is reused the next time the same context is encoded.

In FIG. 10, when the first input binary value 0 is input, the section of [0, 1] is narrowed to the 0 side divided by the probability of 0.11, and becomes [0, 0.11]. . When the second input 0 is input, the interval [0, 0.11] is narrowed to [0, 0.1001]. Here, 0.1001 = 0.11 * 0.11. Similarly, the section becomes [0, 0.011011] by the third input 0. Since the fourth input 1 is narrowed to a section on one side, the final section is narrowed to [0.01010001, 0.011011]. Here, among the values included in the final section, the one having the shortest word length is the code word. Since 0.011 is included in this section, 011 after the decimal point becomes the output binary string, and the 4-bit input value is compressed to 3 bits. Actually, a process called renormalization is performed to shift the probability value to the left when the output bit (0 or 1) is determined.
ISO / IEC 14496-10 Advanced video coding for generic audiovisual services

  In moving picture coding, in order to maximize the image quality under a specific bit rate condition, appropriately control the quantization parameter for determining the quantization width when coding each macroblock, It is necessary to perform code amount control. For this purpose, it is essential to perform code amount prediction such as how much code amount can be obtained by encoding with the current macroblock or frame with various quantization parameters. It is used.

  As a basic method for that purpose, there is a method for calculating an optimal quantization parameter for encoding after performing temporary encoding with a quantization parameter once in units such as a macroblock and a frame and measuring a generated code amount. is there. For example, in a real-time encoder, there is a case where encoding is performed after a delay of one frame and a provisional encoding of one frame. Also, provisional encoding and encoding can be repeated in units such as slices smaller than a frame. If it is not a real-time encoder, encoding called 2-pass encoding performed in larger units such as GOP (Group Of Pictures) or the entire sequence can be performed.

  However, in code amount control when arithmetic coding is used, provisional coding is performed with a certain quantization parameter to obtain a generated code amount after arithmetic coding, and then coding is performed with a newly set quantization parameter. In this case, since the probability distribution of the symbol differs from that in the temporary encoding due to the change of the quantization parameter, there is a problem that the code amount prediction cannot be performed with high accuracy.

  In view of such a problem, an object of the present invention is to improve the accuracy of code amount control of an encoding device to which arithmetic encoding is applied.

  In order to solve the above-described problems, an encoding apparatus according to the present invention is an encoding apparatus for encoding a signal, an orthogonal transform circuit that orthogonally transforms an input signal and outputs coefficient data, and coefficient data A quantization circuit for quantizing the data, a quantization control circuit for generating a quantization parameter for quantization by the quantization circuit, and a data quantized by the quantization circuit based on the quantization parameter An arithmetic encoding circuit that performs arithmetic encoding, and at the time of provisional encoding, the quantization control circuit generates the arithmetic by selecting the provisional quantization parameter as a quantization parameter from a plurality of representative quantization parameters, The encoding circuit selects context information including information indicating the probability of occurrence of coefficient data based on the temporary quantization parameter, performs arithmetic encoding based on the context information, and performs arithmetic The generated code amount of the encoded data is calculated, and at the time of encoding, the quantization control circuit determines the optimum quantization parameter based on the temporary quantization parameter, the generated code amount and a predetermined code amount to be given at the time of encoding. The quantization parameter is generated, and the arithmetic coding circuit is configured to perform arithmetic coding based on the optimum quantization parameter and predetermined context information.

  According to the present invention, it is possible to improve the accuracy of the prediction code amount for various quantization parameters and improve the accuracy of the code amount control of a real-time encoder to which arithmetic coding is applied. An encoding apparatus that performs encoding can be realized.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
An example of a video encoding apparatus according to Embodiment 1 of the present invention is shown in FIG. In FIG. 1, a video encoding apparatus 100 includes a DCT circuit 110 as an orthogonal transform circuit, an inverse DCT circuit 111, a quantization circuit 120, an inverse quantization circuit 121, an adaptive quantization control circuit 130, a context selection circuit 131, an arithmetic code. And the intra prediction circuit 150, the inter prediction circuit 160, the motion prediction circuit 161, and the deblock filter circuit 170. The operation of the video encoding apparatus configured as described above will be described below.

  First, the flow of video encoding will be described with reference to FIG. First, encoding of an intra macroblock that can be decoded only with data in a frame will be described. The macro block is a block of 16 × 16 pixels in general, and includes a plurality of blocks serving as units of DCT conversion and intra prediction described later. For each pixel value of the macroblock to be encoded in the input digital video signal 101, the pixel prediction value of the macroblock to be encoded created in the intra prediction circuit 150 and the difference value for each pixel are calculated, and 16 × 16 A block of pixel difference values is generated.

  Next, DCT conversion is performed on the difference value block by the DCT circuit 110. This processing is normally performed in units of blocks of 4 × 4 pixels or 8 × 8 pixels, and coefficient data of frequency components is output. The coefficient data is input to the quantization circuit 120 and quantized. The quantized coefficient data is arithmetically encoded by the arithmetic encoding circuit 140. Further, the quantized coefficient data is decoded by the inverse quantization circuit 121 and the inverse DCT circuit 111, and is added for each pixel with the predicted pixel value of the macroblock, and converted into a decoded pixel of 16 × 16 pixels. . The obtained decoded pixel is input to the intra prediction circuit 150. In the intra prediction circuit 150, pixel prediction values for peripheral macroblocks to be encoded later are created.

  Next, encoding of an inter macroblock using inter-frame prediction will be described. The motion prediction circuit 161 performs a matching process between a block to be encoded (a block serving as a unit for motion compensation) and a pixel of a reference frame, and calculates a vector value that minimizes an error. A difference value between the pixel prediction value of the encoding target macroblock output from the inter prediction circuit 160 based on the motion prediction result and the pixel value of the encoding target macroblock in the video signal 101 is calculated, and the DCT circuit 110 is input. The subsequent processing is basically the same as the encoding of the intra macroblock. The coefficient decoded by the inverse DCT circuit 111 is added for each pixel with the predicted pixel value of the macroblock. In many cases, the obtained decoded pixels are input to the inter prediction circuit 160 via the deblock filter circuit 170 in order to reduce visual block distortion.

  The present invention includes two types of phases, “provisional encoding” for performing code amount prediction and “encoding” for performing encoding for final output. A method of realizing these two types of phases with the video encoding device of FIG. 1 will be described with reference to FIG. In FIG. 2, the minimum number of units that can provide a quantization parameter for controlling the quantization width is N, and a set of N minimum units is called an operation unit. Usually, the minimum unit is a macroblock. N can be any natural number. N can be set to, for example, the number of macroblocks constituting one slice, the number of macroblocks constituting one frame, the number of macroblocks constituting M frame (M is a natural number of 2 or more), and the like. As will be described later, in this embodiment, the operation unit is described as one slice. That is, as N, the number of macroblocks constituting one slice is employed. As shown in FIG. 2, after performing N times of temporary encoding from the 0th to the (N-1) th macroblock for each operation unit, N times of encoding from the 0th to the (N-1) th macroblock are performed. I do. Thereafter, this operation is repeated.

The operation of provisional encoding #n (0 ≦ n ≦ N−1) in FIG. 2 will be described with reference to FIG. FIG. 3 shows the main part of the video encoding apparatus shown in FIG. When the coefficient data 180 for one macroblock output from the DCT circuit 110 is input to the quantization circuit 120, the temporary quantization parameter QP ₁ (n) 181 determined by the adaptive quantization control circuit 130 is converted into the quantization circuit 120. To enter. A method for determining the temporary quantization parameter QP ₁ (n) 181 will be described later. The quantization circuit 120 performs quantization using the given temporary quantization parameter QP ₁ (n) 181. The quantized coefficient data 184 is input to the arithmetic coding circuit 140b. The arithmetic coding circuit 140b receives the context selection information 183 corresponding to the value of the temporary quantization parameter QP ₁ (n) 181 one-to-one from the context selection circuit 131, performs arithmetic coding, and outputs a binary string 191. . When Q types (Q is a natural number) of temporary quantization parameters QP ₁ (n) 181 are used, the context selection information 183 can be an integer value from 0 to Q-1. The generated code amount R ₁ (n) 182 indicating the length (number of bits) of the binary sequence 191 output by arithmetic encoding for one macroblock is input to the adaptive quantization control circuit 130.

  An arithmetic coding circuit 140b according to the present invention is shown in FIG. The arithmetic encoding circuit 140b is obtained by replacing the context calculation circuit 520 of the arithmetic encoding circuit 140 of FIG. 9 with a quantization parameter adaptive context calculation circuit 521. The quantization parameter adaptive context calculation circuit 521 stores a total of 460 × Q occurrence probability information for each combination of 460 context indexes ctxIdx and Q types (Q is a natural number) of quantization parameters. The quantization parameter adaptive context calculation circuit 521 receives the ctxIdx calculated from the control information 502 and the context selection information 183, and sets the values of ctxIdx and the context selection information from the stored 460 × Q occurrence probability information. Corresponding context information 504 is output.

  The context information 504 output as described above is referred to in the binary arithmetic encoding circuit 530 when the coefficient data 184 input as the multi-value input data 501 is encoded in the arithmetic encoding circuit 140b. That is, in the arithmetic encoding of the coefficient data 184, different occurrence probability information is referred to depending on the quantization parameter.

Next, the operation of encoding #n (0 ≦ n ≦ N−1) in FIG. 2 will be described with reference to FIG. FIG. 4 shows the main part of the video encoding apparatus shown in FIG. The adaptive quantization control circuit 130 to which the generated code amount R ₁ (n) 182 after provisional encoding is input has the provisional quantization parameter QP ₁ (n) 181, the generated code amount R ₁ (n) 182, N pieces The optimum quantization parameter QP ₂ (n) 185 is calculated based on the code amount T to be given to the macroblock and input to the quantization circuit 120. The value of QP ₂ (n) 185 may be corrected before being input to the quantization circuit 120 depending on the control situation or the like. The quantization circuit 120 performs quantization with the given optimum quantization parameter QP ₂ (n) 185. The quantized coefficient data 187 is input to the arithmetic coding circuit 140b. In encoding #n, the context selection circuit 131 does not operate. Unlike temporary encoding, the context selection information 183 input to the arithmetic encoding circuit 140b is always fixed, and the arithmetic encoding circuit 140b performs the same operation as the arithmetic encoding circuit 140. The value of the generated code amount R ₂ (n) 186 in arithmetic coding is used as the accumulated generated code amount up to the present time to correct the value of the optimal quantization parameter QP ₂ (n). Entered.

  A specific example of provisional encoding in FIG. 3 will be described using an example in which an optimum quantization parameter is calculated for each slice in encoding of one frame of 1920 × 1080 pixels as shown in FIG. A frame of 1920 × 1080 pixels is composed of 120 × 68 = 8160 macroblocks of 16 × 16 pixels. A slice that divides a frame can be defined as a set of arbitrary continuous macroblocks. For example, a case where one slice includes 2040 macroblocks and one frame includes four slices is considered. In this case, N = 2040 can be set.

First, the temporary quantization parameter QP ₁ (n) is determined. The temporary quantization parameter is set by selecting one of a plurality of candidates for each macroblock. That is, the provisional quantization parameter QP ₁ (n) for one macroblock is obtained by changing _{one of} arbitrary different Q kinds of representative quantization parameters qp ₁ (x) (x = 0, 1,..., Q−1). This can be determined by selecting and setting the provisional quantization parameter QP ₁ (n) of the macroblock. For example, consider a case where two types of representative quantization parameters qp ₁ (x) = {0, 20}, 0 and 20, are given. Here, the notation of F (x) = {A ₀ , A ₁ ,..., A _Q-1 } (A _n is an integer) is F (0) = A ₀ , F (1) = A ₁ ,. (Q-1) = A It shall mean _Q-1 . One of these two types is selected as the provisional quantization parameter for each macroblock. Also, the selection frequencies of the two types of representative quantization parameters are made equal. In this case, the quantization parameter 0 is given to half the macroblocks, and the quantization parameter 20 is given to the other half macroblocks. Further, in order to prevent the temporary quantization parameter from being biased in the slice, the temporary quantization parameter is arranged alternately or randomly in the slice. As in the example of FIG. 5, provisional quantization parameters can be given alternately in the order of normal raster scan in progressive mode, and alternately in pairs of macroblocks in interlace.

A method for determining the type Q of the representative quantization parameter qp ₁ (x) and the value qp ₁ (x) (x = 0, 1,..., Q−1) in provisional encoding will be described with reference to FIG. The representative quantization parameter type Q used by the adaptive quantization control circuit 130 for provisional encoding is a divisor of 2040, or a divisor of 1020 when pairing macroblocks for interlace encoding. be able to. By setting the representative quantization parameter type Q in this way, the number of times that an arbitrary representative quantization parameter appears in a slice can be made constant. For example, in the case of progressive, 5, 10, 20 which is a divisor of 2040 can be selected as Q. In these cases, the number of appearances of each representative quantization parameter in one slice is 408 times, 204 times, and 102 times, respectively.

Representative quantization parameter qp ₁ (x), so that a different value from a range of quantization parameters that can be selected by the encoder will be Q pieces selected, facilitated the code amount prediction for the optimum coding, the value It is preferable to disperse appropriately. For example, when the quantization parameter QP that can be selected by the encoder is in the range of 0 to 51, as an example, Q = 10, qp ₁ (x) = {0, 4, 8, 12, 16, 22, 28, 34, 42, 51} can be selected. In addition, when the quantization parameter QP ₂ (n) at the time of encoding can be limited to an arbitrary range from information on the encoding difficulty level of the frame and the target code amount in advance, at a finer interval within the limited range. It is possible to select the representative quantization parameter qp ₁ (x). For example, when it is determined that only 0 to 31 are used as the quantization parameter QP ₂ (n), as an example, Q = 10, qp ₁ (x) = {0, 3, 6, 9, 12 , 15, 19, 23, 27, 31} can be selected.

A method of selecting the temporary quantization parameter QP ₁ (n) (n = 0, 1,..., N−1) from qp ₁ (x) (x = 0, 1,..., Q−1) will be described. . This selection method can be realized by a formula of mapping from the value of n to the value of x set so that representative quantization parameters are distributed and arranged in the slice. As a preferable example of the mapping formula, Formula 1 can be used for progressive, and Formula 2 can be used for interlace.
x = ((n% mb_width) + d (n))% Q (Formula 1)
x = ((((n / 2)% mb_width) + d (n))% Q (Formula 2)
However, the operation A% B is an operation that gives a remainder obtained by dividing the integer A by the integer B. N is a macroblock number currently encoded. n is reset to 0 in slice units or frame units. The operation A / B is an operation that gives a quotient obtained by dividing the integer A by the integer B. mb_width is the number of macroblocks in the horizontal direction, and is 120 in the example of FIG. d (n) is a variable for performing the vertical dispersive arrangement, and can be defined by, for example, Expression 3 for progressive and Expression 4 for interlace.
d (n) = ((n / mb_width) * a)% Q (Formula 3)
d (n) = (((n / 2) / mb_width) * a)% Q (Formula 4)
However, a is an integer which is relatively prime to Q. When Q that satisfies mb_width% Q = 0 is selected by d (n), the sequence of representative quantization parameters is shifted to the right by (Q−a) each time one line advances. In the example of FIG. 6, a = 7, and it can be confirmed that the value of QP ₁ (n) is shifted by Qa = 10 −7 = 3 macroblocks in the horizontal direction every time one line is advanced.

A method of calculating the optimum quantization parameter QP ₂ (n) in the encoding of FIG. 4 will be described using a specific example of FIG. First, one slice including 2040 macroblocks is provisionally encoded using provisional quantization parameters. Next, the sum of the generated code amount R ₁ (n) is calculated for each macroblock temporarily encoded using the same representative quantization parameter qp ₁ (x) to obtain a slice prediction code amount r ₁ (x). . qp ₁ (x) = {0, 4, 8, 12, 16, 22, 28, 34, 42, 51} (x = 0, 1,..., 9), the generated code amount R ₁ (n ) (n = 0,1, ..., n-1) is the result of taking the sum for each representative quantization parameter slice prediction code amount r ₁ (x) is r ₁ (x) = {171248,120058,79390 , 56516, 40760, 24026, 14864, 7966, 4742, 2572}. Here, for example, r ₁ (0) = 171248 can be obtained by adding code amounts for the following macroblocks. Assuming that mb_width = 1920 pixels / 16 pixels = 120 and a = 7, Equation 1 and Equation 3 are expressed as x = ((n% 120) + d (n))% 10, d (n) = ((n / 120) * 7)% 10. Therefore, a macroblock temporarily encoded using the representative quantization parameter qp ₁ (0) = 0, that is, n = 0, 10, 20,..., 110, 123, 133,..., 233, 246, 256,. , 356,..., 1928, 1938,..., 2038 is assumed to be 171248 as a result of summing the generated code amounts R ₁ (n) for the macroblocks. Further, it is assumed that the target code amount T = 300,000 bits of the slice allocated using information such as the bit rate, the frame rate, the frame complexity, and the coding type of the slice.

First, from qp ₁ (x) and r ₁ (x), the relationship between the quantization parameter QP and the slice prediction code amount can be plotted on the coordinate axes as shown in FIG. Since r ₁ (x) is the sum of generated code amounts of 204 macroblocks for each x = QP, the slice prediction code amount for each QP is 10 times r ₁ (x). Is required. In addition, after plotting the QP value, the QP value can be linearly interpolated to create a graph of the slice prediction code amount for all the QP values (0 ≦ QP ≦ 51).

By obtaining the target code amount T = 300,000 of the slice and the intersection of this graph, the optimum quantization parameter QP ₂ (n) = QP _opt (0 ≦ n ≦ 2039) for generating the target code amount can be obtained. it can. The calculation of QP _opt is obtained by Equation 5 by linear interpolation.
QP _opt = q + (QP ₁ (q + 1) −QP ₁ (q)) * (Q * r ₁ (q) −T) / (Q * r ₁ (q) −Q * r ₁ (q + 1)) (Formula 5 )
However, q is an integer of 0 ≦ q ≦ Q−1 and is a value satisfying Q * r ₁ (q + 1) ≦ T ≦ Q * r ₁ (q). Here, one optimal quantization parameter is used throughout one slice. Also, q = 0 when T> Q * r ₁ (q) for all q, and q = Q−1 when T <Q * r ₁ (q) for all q. For this reason, it is desirable to set QP ₁ (0) = min (QP) = 0 and QP ₁ (Q−1) = max (QP) = 51 in advance as temporary quantization parameters.

Using Equation 5, QP _opt is QP _opt = 16 + (22−16) * (407600−300000) / (407600−240260) ≈19.9. Since the quantization parameter is an integer, 20 is set as the optimum quantization parameter QP ₂ (n). Depending on the value after the decimal point, the smaller value 19 may be selected.

The reason why the code amount prediction accuracy is improved by the present invention will be described. In provisional encoding, it is possible to perform highly accurate code amount prediction by changing the provisional quantization parameter in a short cycle such as a macroblock unit. However, when the conventional arithmetic encoding circuit 140 is used for provisional encoding, there are the following problems. The probability distribution of the symbol of the coefficient data by DCT transform varies greatly depending on the quantization parameter. This can be seen from the fact that, when quantized with a large quantization parameter, the probability that zero appears in the coefficient data is high, but it is difficult for zero to appear with a small quantization parameter. Therefore, when N macroblocks are encoded with Q types of quantization parameters in provisional encoding, the amount of generated codes obtained by performing state transitions on quantization parameters of various probability distributions is image encoding. As a result, the error tends to increase compared to the case where encoding is always continued with one kind of desired quantization parameter. This is because in the case of one kind of quantization parameter, state transition continues with a certain probability. As a result, the difference in state transition between the temporary encoding and the encoding becomes large. On the other hand, in the arithmetic coding circuit 140b of the present invention, for each of Q kinds of representative quantization parameters qp ₁ (x) (x = 0, ₁ ,. By holding and updating the state value, it is possible to approach the state transition of arithmetic encoding in temporary encoding that is always encoded with one kind of quantization parameter, and as a result, code amount prediction accuracy can be improved. In addition, since the same context information 504 is used regardless of the quantization parameter at the time of encoding, encoding by normal CABAC is performed, so that a stream conforming to the encoding standard can be output.

In the present embodiment, an example in which a fixed QP _opt is given as the optimum quantization parameter QP ₂ (n) has been shown. For example, QP _opt is used as an initial value for encoding N macroblocks, and the subsequent quantum The quantization parameter is subjected to code amount control by feedback control, and the optimal quantization parameter QP ₂ (n) can be dynamically changed. In the feedback control, in a certain macroblock number m, the generated code amount R ₂ (n) at the time of encoding is added within a range of 0 ≦ n ≦ m, and at a certain time (for example, the macroblock number N− The remaining code amount TB, which is the difference from the target code amount T until 1), is set to the target code amount S of the remaining macroblock n (m <n ≦ N−1). it can. In addition, it is possible to dynamically change the quantization parameter QP ₂ (n) based on the luminance value and color difference value of the macroblock, the characteristics in the frequency domain, and the like.

In the present embodiment, N is described as a slice unit. However, the generated encoding amount R ₂ (n) is obtained by performing temporary encoding and encoding in smaller units, and the result is selected as a temporary quantization parameter. In the case where it is reflected in the above, or when the feedback control is performed in a small unit, it is effective to implement it in a small unit such as several macroblocks. Although N is described as being fixed, it is also possible to dynamically change N in frame units or other units. Moreover, although the example which calculates | requires by performing the linear interpolation of the graph of a slice prediction code amount was shown, advanced numerical interpolations, such as spline interpolation, are also possible.

  In this embodiment, an example in which provisional encoding is performed only once has been shown. However, when real-time encoding is not necessary, or when provisional encoding is pipelined or parallelized, many processes can be performed with a certain delay. If possible, the provisional encoding may be performed any number of times.

  In the present embodiment, H.264 for video coding is used. The context adaptive arithmetic coding in the H.264 standard has been described as an example, but any coding apparatus or coding method that can apply arithmetic coding to quantized coefficient data and adaptively change the quantization parameter can be used. However, the present invention is not limited to this.

  The bit stream encoded using the present embodiment can be recorded on a recording medium such as a tape, an optical disk, a magnetic disk, or a semiconductor memory, so that a high-efficiency and high-quality video can be redistributed.

(Embodiment 2)
This embodiment shows a configuration in which a circuit that performs provisional encoding and encoding is independent of the first embodiment.

  FIG. 8 shows the configuration of the video encoding apparatus according to the present embodiment. The video encoding device 600 includes a temporary encoding circuit 610, an encoding circuit 620, an adaptive quantization control circuit 630, and a FIFO buffer 640. The temporary encoding circuit 610 and the encoding circuit 620 are assumed to have the same configuration as that of the video encoding apparatus 100 described in the first embodiment. The operation of the video encoding apparatus 600 configured as described above will be described below.

When the video signal 601 is input, the provisional encoding circuit 610 performs provisional encoding on N macroblocks. The generated code amount R ₁ (n) is passed to the adaptive quantization control circuit 630. The video signal 601 is input to the FIFO buffer 640 and held as data for N macroblocks. When the adaptive quantization control circuit 630 finishes calculating the optimal quantization parameter QP ₂ (n), the delayed video signal 601 is input from the FIFO buffer 640 to the encoding circuit 620 and is optimal for N macroblocks. Encoding with the quantization parameter QP ₂ (n) is performed. As a result, an output bit stream 602 is obtained.

  The advantage of this embodiment is that the effect of FIG. 11 can be obtained by separating the provisional encoding and encoding circuits. As shown in FIG. 11, at the time when the encoding circuit 620 encodes the macroblocks # 0 to # N-1, the temporary encoding circuit 610 performs provisional processing for the next N macroblocks #N to # 2N-1. Can be encoded. Therefore, by connecting two encoding circuits capable of real-time operation and implementing the adaptive quantization control circuit 130 that gives the temporary quantization parameter only inside the temporary encoding circuit 610, and the context selection circuit 131, This embodiment can be easily realized.

(Embodiment 3)
The present embodiment shows a configuration for realizing the configuration of the first embodiment with software.

  The video encoding process according to the present invention will be described with reference to the flowchart of FIG. When the video encoding process 200 is started, an operation unit is determined (210). N is as described in the first embodiment.

In the determination step 220 of the temporary quantization parameter QP ₁ (n), a temporary quantization parameter for quantization in the temporary encoding process 230 is calculated. Specifically, the temporary quantization parameter QP ₁ (n) (n = 0, 1,..., N−1) for N macroblocks is replaced with Q types of representative quantization parameters qp ₁ (x) (x = 0, 1, ..., Q-1). Based on the determined temporary quantization parameter QP ₁ (n), the macroblock temporary encoding process 230 is repeated N times (261).

In the calculation step 240 of the optimal quantization parameter QP ₂ (n), a quantization parameter for quantization in the encoding process 250 is calculated. Specifically, the temporary quantization parameter QP ₁ (n) and the generated code amount R ₁ (n) (n = 0, 1,..., N−1) and N macroblocks in the temporary encoding process 230 Based on the target code amount T to be allocated, optimum quantization parameters QP ₂ (n) (n = 0, 1,..., N−1) of N macroblocks are calculated. Based on the calculated optimal quantization parameter QP ₂ (n), the macroblock encoding process 250 is repeated N times (262). Finally, if all the video data has been encoded, the process ends (290). If not completed, the process returns to the temporary quantization parameter calculation step of step 220, and the same processing is performed on the next video data. Repeat (260).

The operation of the provisional encoding process 230 will be described with reference to the flowchart of FIG. First, DCT processing 270 is performed. After performing quantization (231) using the temporary quantization parameter QP ₁ (n), different arithmetic coding context selection (232) is performed according to QP ₁ (n). In this operation, in particular, in arithmetic coding of syntax corresponding to coefficient data by DCT transformation, a context, that is, a state value (MPS or pState) is converted into a quantization parameter QP ₁ (n) ∈qp ₁ (x) (x = It is used by switching every 0, 1, ..., Q-1). That is, syntax state values corresponding to coefficient data by DCT conversion used by CABAC are increased to Q types and held for each qp ₁ (x), and the state values to be used are divided by QP ₁ (n). . Arithmetic coding (233) in provisional coding is performed to obtain a generated code amount R ₁ (n). Then, the prediction process 280 including the inverse quantization process 272, the inverse DCT process 271, the intra prediction and the inter prediction is performed, and the process ends (239).

The operation of the encoding process 250 will be described using the flowchart of FIG. First, DCT processing 270 is performed. After performing quantization (251) using the quantization parameter QP ₂ (n), arithmetic context selection (252) is performed. This operation is different from using different context information depending on the representative quantization parameter qp ₁ (x) in the provisional encoding process 230, and does not use different context information depending on the quantization parameter. Arithmetic coding (253) is performed to obtain a generated code amount R ₂ (n). Then, the prediction process 280 including the inverse quantization process 272, the inverse DCT process 271, the intra prediction and the inter prediction is performed, and the process ends (259).

  According to the present embodiment, the first embodiment can be realized by software.

  The encoding apparatus of the present invention is effective for a recording apparatus that requires real-time operation, such as a camera recorder or a recording apparatus.

Configuration diagram of video encoding apparatus in Embodiment 1 of the present invention The figure explaining the scheduling of the phase in Embodiment 1 The figure explaining the circuit operation at the time of provisional encoding in the first embodiment The figure explaining the circuit operation | movement at the time of the encoding in Embodiment 1 The figure which shows the example of the temporary encoding of 1 frame in Embodiment 1 The figure which shows the example of selection of the quantization parameter in Embodiment 1 The figure which shows the example of calculation of the optimal quantization parameter in Embodiment 1 Configuration diagram of video encoding apparatus in Embodiment 2 of the present invention The figure explaining the arithmetic coding circuit The figure explaining the operation | movement of binary arithmetic coding The figure explaining the scheduling of the phase in Embodiment 2 of this invention The flowchart which shows the encoding process in Embodiment 3 of this invention. Flowchart showing provisional encoding processing in the third embodiment Flowchart showing encoding processing in the third embodiment The figure explaining the arithmetic coding circuit in Embodiment 1 of this invention

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Video coding apparatus 101 Video signal input 110 DCT circuit 111 Inverse DCT circuit 120 Quantization circuit 121 Inverse quantization circuit 130 Adaptive quantization control circuit 131 Context selection circuit 140 Arithmetic coding circuit 150 Intra prediction circuit 160 Inter prediction circuit 161 Motion Prediction circuit 200 Video encoding process 210 Step of determining operation unit N 220 Calculation step of temporary quantization parameter QP ₁ (n) 230 Temporary encoding process 231 Quantization step by temporary quantization parameter QP ₁ (n) 232 QP ₁ ( n) Context selection step of arithmetic coding that differs depending on n) 233 Arithmetic coding step in provisional coding 240 Calculation step of optimum quantization parameter QP ₂ (n) 250 Encoding process 251 Quantization by quantization parameter QP ₂ (n) Conversion 252 Context selection step of arithmetic coding not based on QP ₂ (n) 253 Arithmetic coding step

Claims (13)

An encoding device for encoding a signal,
An orthogonal transform circuit that orthogonally transforms the input signal and outputs coefficient data;
A quantization circuit for quantizing the coefficient data;
A quantization control circuit for generating a quantization parameter for quantization by the quantization circuit;
An arithmetic coding circuit that performs arithmetic coding on the data quantized by the quantization circuit based on the quantization parameter;
At the time of provisional encoding,
The quantization control circuit generates a temporary quantization parameter as the quantization parameter by selecting from a plurality of representative quantization parameters,
The arithmetic coding circuit selects the context information including information indicating the occurrence probability of the coefficient data based on the temporary quantization parameter, performs arithmetic coding based on the context information, and arithmetically encoded data The generated code amount of
When encoding,
The quantization control circuit generates an optimal quantization parameter as the quantization parameter based on the temporary quantization parameter, the generated code amount, and a predetermined code amount to be given at the time of encoding,
The coding apparatus according to claim 1, wherein the arithmetic coding circuit performs arithmetic coding based on the optimum quantization parameter and predetermined context information. The quantization control circuit generates Q quantization parameters (Q is a natural number of 2 or more) when generating the quantization parameter for quantizing the coefficient data of N (N is a natural number of 2 or more). The encoding apparatus according to claim 1, wherein a parameter is set, and any one quantization parameter is selected from the Q representative quantization parameters to generate the quantization parameter. N is N = L * Q (L is a natural number),
The generation of the quantization parameter in the quantization control circuit is performed by selecting different Q quantization parameters for consecutive Q pieces of coefficient data. Encoding device. An encoding device for encoding a signal,
An orthogonal transform circuit that orthogonally transforms the input signal and outputs coefficient data;
A first quantization control circuit that generates a first quantization parameter by selecting from a plurality of representative quantization parameters;
A first quantization circuit that quantizes the coefficient data based on the first quantization parameter to obtain first quantized data;
The context information including information indicating the occurrence probability of the coefficient data is selected based on the first quantization parameter, and arithmetic coding is performed on the first quantized data using the context information. A first arithmetic encoding circuit for calculating a generated code amount of the arithmetically encoded data;
A second quantization control circuit for generating a second quantization parameter based on the first quantization parameter, the generated code amount, and a predetermined code amount to be given at the time of encoding;
A second quantization circuit that quantizes the coefficient data based on the second quantization parameter to obtain second quantized data;
A coding apparatus comprising: a second arithmetic coding circuit that performs arithmetic coding on the second quantized data based on the second quantization parameter and predetermined context information. The first quantization control circuit generates Q (Q is a natural number equal to or greater than 2) when generating the first quantization parameter for quantizing the N (N is a natural number equal to or greater than 2) the coefficient data. 5), the first quantization parameter is generated by selecting any one quantization parameter from the Q representative quantization parameters. The encoding device described in 1. N is N = L * Q (L is a natural number),
The generation of the first quantization parameter in the first quantization control circuit is performed by selecting Q representative quantization parameters different from each other for consecutive Q pieces of coefficient data. The encoding device according to claim 5. 7. The encoding apparatus according to claim 2, wherein the Q representative quantization parameters are configured to always include a maximum value and a minimum value of the quantization parameter. . The signal is a video signal;
The context information includes at least an MPS value indicating a symbol having a high probability of occurrence;
A pState value indicating the probability of occurrence of the symbol,
8. The encoding apparatus according to claim 1, wherein the coefficient data is a coefficient obtained by converting the signal into a frequency component. An encoding method for encoding a signal, comprising:
An orthogonal transform processing step of orthogonally transforming the input signal and outputting coefficient data;
A first quantum generated by selecting a first quantization parameter from a plurality of representative quantization parameters, and quantizing the coefficient data based on the first quantization parameter to obtain first quantized data Step,
The context information including information indicating the occurrence probability of the coefficient data is selected based on the quantization parameter, and arithmetic coding is performed on the first quantized data and arithmetic coding is performed based on the context information. A first arithmetic encoding step for calculating a generated code amount of the processed data;
When encoding,
A second quantization parameter is generated based on the first quantization parameter, the generated code amount and a predetermined code amount to be given at the time of encoding, and the coefficient data is quantized based on the second quantization parameter. A second quantization step that is converted into second quantized data;
And a second arithmetic encoding step for performing arithmetic encoding on the second quantized data based on the second quantization parameter and predetermined context information. The first quantization step is representative of Q (Q is a natural number of 2 or more) when generating the quantization parameter for quantizing the coefficient data of N (N is a natural number of 2 or more). Set the quantization parameter,
The encoding method according to claim 9, wherein the quantization parameter is generated by selecting any one quantization parameter from the Q representative quantization parameters. N is N = L * Q (L is a natural number),
The quantization parameter generation in the first quantization step is performed by selecting Q representative quantization parameters different from each other for consecutive Q pieces of coefficient data. The encoding method described in 1. The encoding method according to claim 10 or 11, wherein the Q representative quantization parameters are configured to necessarily include a maximum value and a minimum value of the quantization parameter. The signal is a video signal;
The context includes at least an MPS value indicating a symbol having a high probability of occurrence;
A pState value indicating the probability of occurrence of the symbol,
13. The encoding method according to claim 9, wherein the coefficient data is a coefficient obtained by converting the signal into a frequency component.

Images