JP4361836B2 - Moving picture coding method, moving picture coding apparatus, moving picture coding program, and computer-readable recording medium recording the program - Google Patents

Description

  The present invention relates to a moving image coding technique, and in particular, H.264 with a given code amount. In order to efficiently encode based on standards such as H.264, MPEG-4, MPEG-2, etc., the encoding is divided into a plurality of stages, and the encoding information in the previous stage is used at any stage. The present invention relates to a moving picture coding technique that achieves higher coding efficiency.

  For example, based on the MPEG-2 video encoding method described in Non-Patent Document 1 below, one of the methods for performing encoding in a plurality of stages in order to perform efficient encoding is the first method. In this stage, encoding is performed using the same quantization parameter value in all image areas, and the complexity of each area is predicted from the amount of generated codes. There is a method of determining the quantization parameter value based on the coding efficiency and increasing the coding efficiency.

A code amount distortion D and a code amount R are calculated to determine an encoding syntax. In the H.264 video encoding method, a method for improving encoding efficiency by performing encoding in a plurality of stages has not been proposed. The encoding syntax means an encoding syntax indicating how the encoding characters (“1” or “0”) are arranged.
MPEG-2 International Standard ISO / IEC 13818-2 1995.

  From the viewpoint of the Lagrange multiplier method, it is best to select a coding syntax that minimizes D + λR by calculating the code amount distortion D and the code amount R while keeping the Lagrange multiplier λ as constant as possible. This is an efficient encoding method.

  However, it is not guaranteed that the Lagrangian multiplier λ is close to a constant even if the encoding that gives a code amount close to the target code amount can be achieved by only one stage of encoding. In addition, in order to give a code amount close to the target code amount and keep the Lagrangian multiplier λ at a constant value, many encoding steps are required when encoding is performed in a plurality of steps. There is a problem that the amount of calculation required for encoding becomes enormous.

  An object of the present invention is to solve the above-described problems of the prior art and to realize moving picture encoding that suppresses an increase in the amount of calculation required for encoding when encoding is performed in a plurality of stages.

  In order to solve the above problems, the present invention is characterized in that, at each encoding stage, an encoding syntax that minimizes D + λR is selected for a plurality of Lagrange multipliers λ. Thereby, the λ dependency of the code amount R can be calculated, and the value of λ giving a code amount close to the target code amount can be determined in as few steps as possible.

  Further, the present invention is characterized in that at least two or more kinds of λ that give a code amount larger than a target code amount and a small code amount are selected, and the difference between the maximum value and the minimum value of λ is reduced as the number of steps increases. To do. As a result, the value of λ giving a code amount close to the target code amount can be determined without modeling the λ dependency of R.

  In the process of selecting an encoding syntax that minimizes D + λR for a plurality of Lagrangian multipliers λ, D + λR for a plurality of λs can be continuously calculated for each candidate encoding syntax. In addition, since it is only necessary to confirm whether or not the encoding syntax minimizes D + λR for a certain λ, the increase in the amount of calculation is not so large.

That is, according to the first aspect of the present invention, in the moving picture coding method for calculating the code amount distortion D and the code amount R and determining the coding syntax, L, M, and N are integers of 2 or more, and i is from 1. In the k-th stage, L Lagrange multipliers λ _1, for the jth block, where L is an integer, j is an integer from 1 to M, and k is an integer from 1 to (N−1) _{. j, k} , λ _{2, j, k} , λ _{3, j, k} ,..., λ _{L, j, k} are selected from among the encoding syntaxes that are candidates for encoding the j th block. , The coding syntax that minimizes D + λ _{i, j, k} R in _{the j-th} block is selected for each of i = 1, 2, 3,..., L, and the (j + 1) -th block is similarly selected. Is repeated until j becomes 1 to M, the encoded information at the k-th stage is held, and at the (k + 1) -th stage, the first With coded information which is held between the floor up to the stage of the k lambda _{i, j,} determining the _{k + 1,} at the stage of the N, are candidates for encoding a block of the j Coding that minimizes any one of D + λ1 _{, j, NR} , D + λ2 _{, j, NR} , D + λ3 _{, j, NR} ,..., D + λL _{, j, NR} from the encoding syntax. The output of the syntax as an encoded stream for the j-th block is repeated until j becomes 1 to M.

The second aspect of the present invention is the first aspect of the present invention, wherein the overall target code amount from the first block to the Mth block is _Rtarget, and D + λ _i in the jth block at the kth stage. _{, j, k When} the code amount in the encoding syntax that minimizes R is R _{i, j, k} , and the sum of R _{i, j, k} where j is 1 to M is R _{i, sum, k} , The absolute value of the difference between R _{i, sum, k} and R _target is smaller than a certain value, and λ _{i, j, k} when j = 1, 2, 3,. I and k are i ₀ and N, respectively, and in the Nth stage, D + λ _{i0, j, N} R is selected from the encoding syntaxes that are candidates for encoding the jth block. It is characterized in that the encoding syntax to be minimized is output as an encoded stream for the j-th block until j becomes 1 to M. The

The third of the present invention, in the first invention, the in k stages of the block in the D + lambda _{i, j} of the _{j, k} code amount in encoding syntax to minimize R R _{i, j , k} , j is the sum of R _{i, j, k} from 1 to M _{, where} R _{i, sum, k} is λ _{i, 1, k} = λ _{i, 2, k} = λ _{i, 3, k} = ,..., Λ _{i, M, k} is set so that Lagrange multipliers hold for all i and k, and λ _{i, 1, k} for all i and R _{i, sum, k} for all i. Is stored as encoded information in the k-th stage.

The fourth of the present invention, in the first invention, the in k stages of the block in the D + lambda _{i, j} of the _{j, k} code amount in encoding syntax to minimize R R _{i, j , k} , j is the sum of R _{i, j, k} from 1 to M _{, where} R _{i, sum, k} is λ _{1, j, k} <λ _{2, j, k} <λ _{3, j, k} < ...... <λ _{L, j, k} holds for all k, and when i = 1, 2, 3,..., L _, the maximum value of R _{i, sum, k} is greater than R _target and i = 1, 2, 3, ..., L, the Lagrange multiplier is determined so that the minimum value of R _{i, sum, k} is smaller than R _target , and R _{i, sum} held up to the k-th stage _{, k} and R _{target from} the values λ _{1, j, k} ≦ λ _{1, j, k + 1} and λ _{L, j, k + 1} ≦ λ _{L, j, k} and λ _{L, j, k + 1} − λ _{i, j, k + 1} (i = 1, 2, 3,..., L and j = 1 so that λ _{1, j, k + 1} <λ _{L, j, k} −λ _{1, j, k} , 2, 3, ..., M) .

The fifth aspect of the present invention is the same as the third aspect of the present invention, in which L = 3, and when R _{2, sum, k} <R _target , λ _{1,1, k + 1} = λ _{1,1, k} and λ _{2,1, k + 1} = (λ _{1,1, k} + λ _{2,1, k} ) / 2 and λ _{3,1, k + 1} = λ _{2,1, k} , R _{2, sum, k} > In the case of R _target , λ _{1,1, k + 1} = λ _{2,1, k} and λ _{2,1, k + 1} = (λ _{2,1, k} + λ _{3,1, k} ) / 2 and λ _{3 , 1, k + 1} = λ _{3, 1, k,} and when the absolute value of the difference between R _{2, sum, k} and R _target becomes smaller than a certain value, k is output as N and the encoded stream is output. Features.

In the sixth aspect of the present invention, in the first aspect of the present invention, when P is an integer of 1 or more and p is an integer from 1 to P, in the k-th stage, D + λ _{i, j in the j-th} block _{, k} R is selected for each of i = 1, 2, 3,..., L, and λ _{i, j, k} for i = 1, 2, 3 _,. And R _{i, j, k} (where R _{i, j, k} is the code amount in the coding syntax that minimizes D + λ _{i, j, k} R in the j-th block), P parameters C _{j, k, 1} , C _{j, k, 2} , C _{j, k, 3} ,..., C _{j, k, P} representing the λ dependence of R are calculated, and for all j and p C _{j, k, p} is stored as encoded information.

Further, the seventh aspect of the present invention is the same as the sixth aspect of the present invention, in which P = 1, and from the relationship between λ _{i, j, k} and R _{i, j, k} for i = 1, 2, 3,. as lambda dependence of R in the block of the j, C of _{R = C j, k, 1} / (λ 1/2) and then C _j when _{the, k, 1} calculates, j is from 1 to M _When the sum of _{j, k, 1} is C _{sum, k, 1} , in the (k + 1) th stage, λ _{1, j, k + 1} , λ _{2, j, k + 1} , λ _{3, j, k} Λ _{i, j, k + 1} is selected such that any _one of ₊₁ ,..., Λ _{L, j, k + 1} is equal to (C _{sum, k, 1} / R _target ) ² .

The eighth invention is characterized in that, in the first invention, L and M are variable with respect to k.

Further, the ninth aspect of the present invention is characterized in that any one of the method of the third aspect of the present invention or the method of the sixth aspect of the present invention is selected for each k-th stage.

FIG. 1 is a diagram illustrating a configuration example of a moving image encoding apparatus according to the present invention. The moving image encoding apparatus 1 includes an image input unit 10, a Lagrange multiplier λ candidate selection unit 11, and a syntax selection unit 12-i (i = 1, 2,..., L) for each Lagrange multiplier λ _{i, j, k} . , A Lagrange multiplier λ evaluation unit 13, and an encoded stream output unit 14.

  The image input unit 10 inputs an image signal to be encoded. The λ candidate selection unit 11 selects, for each encoding target block (j), for example, at least two or more Lagrange multipliers λ candidates that provide a code amount larger than a target code amount and a smaller code amount. Here, the Lagrange multiplier λ is selected so that the difference between the maximum value and the minimum value of the Lagrange multiplier λ becomes smaller as the stage (k) increases.

The syntax selection unit 12-i for each Lagrange multiplier λ _{i, j, k} has an encoding syntax for minimizing D + λ _{i, j, k} R for each of the L Lagrange multipliers λ _{i, j, k.} elect. The Lagrangian multiplier λ evaluation unit 13 evaluates each Lagrange multiplier λ _{1, j, k} , λ _{2, j, k} ,..., Λ _{L, j, k} until a predetermined number of steps is reached or a predetermined end. Until the condition is satisfied, the Lagrange multiplier λ candidate selection unit 11, the syntax selection unit 12-i for each Lagrange multiplier λ _{i, j, k,} and the Lagrange multiplier λ evaluation process are repeated.

The encoded stream output unit 14 has an encoding syntax for minimizing D + λ _{i, j, N} R from the final Lagrange multipliers λ _{i, j, N at} the N-th stage, M from j = 1 to M. The encoded target blocks are output as encoded streams.

  FIG. 2 is a diagram showing an example of an encoding process flow according to the first aspect of the present invention. In the moving image encoding process for calculating the code amount distortion D and the code amount R and determining the encoding syntax, L, M, and N are integers of 2 or more, i is an integer from 1 to L, and j is 1 An integer up to M and k is an integer from 1 to (N−1).

When the processing of the j-th block at the k-th stage is started, i = 1, 2, 3,..., L based on the encoded information held up to the (k−1) -th stage. On the other hand, a Lagrange multiplier λ _{i, j, k} is selected (steps S1 to S5). In the first stage (k = 1), L Lagrange multipliers λ _{i, j, 1} determined in advance are given as initial values.

In the processing of the j-th block at the k-th stage, for each of i = 1, 2, 3,..., L, an encoding thin that minimizes D + λ _{i, j, k} R in _{the j-th} block. A tax is selected (step S6).

  If there is an unprocessed block (step S7), j is incremented (step S8), and the process returns to step S4. If the processed block is the last block (if j = M), the kth The encoded information of the stage is stored and held in the memory (step S9).

If there is an unexecuted stage (step S10), k is incremented (step S11), and the process returns to step S3. If it is the last stage (when k = N), the jth block is From j = 1, the encoding syntax that minimizes D + λ _{i, j, N} R is output as an encoded stream for the j-th block from among the encoding syntaxes that are candidates for encoding. Repeat until M (step S12).

  According to the present invention, a moving picture coding apparatus that calculates the coding amount distortion D and the coding amount R and determines the coding syntax, and the coding is divided into a plurality of stages, and at any stage. In a moving picture coding apparatus that achieves high coding efficiency using coding information in the previous stage, it is possible to perform more efficient moving picture video coding with a small increase in the amount of calculation.

[First Embodiment]
FIG. 3 is an example of an encoding process flow according to the first embodiment of the present invention. The first embodiment corresponds to the case of employing the first to fifth present onset light simultaneously.

In the first embodiment, M and N are integers greater than or equal to 2 and i is an integer from 1 to 3 in a video encoding process for calculating the code amount distortion D and the code amount R and determining the encoding syntax. An integer, j is an integer from 1 to M, k is an integer from 1 to (N−1), and λ _{i, 1, k} = λ _{i, 2, k} = λ _{i, 3, k} =,. Lagrange multipliers are determined so that λ _{i, M, k} holds for all i and k.

In the coding syntax that minimizes D + λ _{i, 1, k} R in the j-th block in the k-th stage, R _target is the total target code amount from the first block to the M-th block. When the code amount is R _{i, j, k} and the sum of R _{i, j, k} from j = 1 to M is R _{i, sum, k} , λ _{1,1, k} <λ _{2,1, k} <λ _{3, 1, k} is satisfied for all k, and R _i when the i = _{1,2,3, sum,} maximum value of _k was set at R greater than _target, and i = 1, 2, 3 The Lagrange multiplier is determined so that the minimum value of R _{i, sum, k} is smaller than R _target .

When the k-th stage is started, i = 1, 1, depending on the magnitude relationship between the encoded information (R _{2, sum, k-1} ) and R _target held up to the (k-1) -th stage. Λ _{i, 1, k} is selected for 2 and 3 (steps S21 to S24). The processing in step S24 will be described in detail later by taking the (k + 1) th stage as an example. In the first stage, the encoded information does not exist held until before _{step, λ 1,1,1 <λ 2,1,1 <λ} 3,1,1 and R _{1, sum , 1> R target> R 3} , sum, 1 and so as to three of the Lagrangian multiplier determined as an initial value lambda _{1, 1, 1,} lambda _2,1,1, selects the lambda _3,1,1 .

An encoding syntax that minimizes D + λ _{i, 1, k} R in the j-th block is selected from the encoding syntaxes that are candidates for encoding the j-th block with respect to the j-th block. = 1, 2, and 3 are selected (step S26), and the (j + 1) th block is processed in the same manner until j becomes 1 to M (steps S27 and S28). R _{2, sum, k} is held as information (step S29).

At the stage of the (k + 1), at step _{S24, R 2, sum, k} <λ 1,1 in the case of _{R target, k + 1 = λ} 1,1, k cutlet λ _{2,1, k + 1} = (Λ _{1,1, k} + λ _{2,1, k} ) / 2 and λ _{3,1, k + 1} = λ _{2,1, k} , and if R _{2, sum, k} > R _target λ _{1 , 1, k + 1} = λ _{2,1, k} and λ _{2,1, k + 1} = (λ _{2,1, k} + λ _{3,1, k} ) / 2 and λ _{3,1, k + 1} = λ _{3,1, k} .

It is determined whether the absolute value of the difference between R2 _{, sum, k} and _Rtarget is smaller than a certain value (step S30), and the absolute value of the difference between R2 _{, sum, k} and _Rtarget is not smaller than a certain value. Increments k (step S31) and performs the next stage of processing. As k increases, R _{2, sum,} since the absolute value of the difference between _k and R _target becomes smaller, the k when it becomes smaller than R _{2, sum,} there is an absolute value difference values _k and R _target N is an encoding syntax that minimizes D + λ _{2, j, N} R among encoding syntaxes that are candidates for encoding the j-th block in the N-th stage, which is the last stage. Are output as an encoded stream for the j-th block from j = 1 to M (step S32).

In the above process, the values of λ _{1,1, k + 1} and λ _{3,1, k + 1} are equal to λ _{1,1, k} or λ _{2,1, k} or λ _{3,1, k} . Therefore, in the (k + 1) -th stage, the coding syntax that minimizes D + λ _{i, 1, k} R in the j-th block from among the coding syntaxes that are candidates for coding the j-th block. It is not necessary to select a tax for each of i = 1, 2, 3; it is sufficient to select only for i = 2.

In the first embodiment, L in the first aspect of the present invention is 3, and when R _{2, sum, k} <R _target , λ _{1,1, k + 1} = λ _{1,1, k} and λ _{2,1, k + 1} = (λ _{1,1, k} + λ _{2,1, k} ) / 2 and λ _{3,1, k + 1} = λ _{2,1, k} , R _{2, sum, k} > R _In the case of _target , λ _{1,1, k + 1} = λ _{2,1, k} and λ _{2,1, k + 1} = (λ _{2,1, k} + λ _{3,1, k} ) / 2 and λ _{3, Although} an example in which _{1, k + 1} = λ _{3,1, k} has been described, it is not necessarily limited to this expression, and L is not limited to 3, and λ _{1,1, k} ≦ λ _{1,1, k +1} and λ _{L, 1, k + 1} ≤λ _{L, 1, k} and λ _{L, 1, k + 1} −λ _{1,1, k + 1} <λ _{L, 1, k} −λ _{1,1, k} It goes without saying that another formula that holds can be used.

In the first embodiment, λ _{1,1, k} <λ _{2,1, k} <λ _{3,1, k} <,..., <Λ _{L, 1, k} holds for all k, and i = The maximum value of R _{i, sum, k} when 1, 2, 3,..., L is larger than R _target and R _{i, sum, k} when i = 1, 2, 3 _,. Although the example using the fourth aspect of the present invention in which the Lagrange multiplier is determined so that the minimum value of R _i is smaller than R _target has been described _, the magnitude relationship of λ _{i, 1, k} does not necessarily follow this equation, and R _i, Others such that the absolute value of the difference between _{sum, k} and R _target is smaller than a certain value, and λ _{i, j, k} is within a certain range when j = 1, 2, 3,. It goes without saying that the method may be used.

[Second Embodiment]
The second embodiment corresponds to the case where the first invention, the second invention, the sixth invention, and the seventh invention are simultaneously employed. FIG. 4 shows an encoding processing flow in the second embodiment.

In a moving picture coding apparatus that calculates a coding amount distortion D and a coding amount R to determine a coding syntax, L, M, and N are integers of 2 or more, i is an integer from 1 to L, and j is 1 An integer up to M, k is an integer from 1 to (N−1), an overall target code amount from the first block to the Mth block is R _target, and in the kth stage, in the jth block D + λ _{i, j, k} Let R _{i, j, k be} the code amount in the coding syntax that minimizes R _, and R _{i, sum, k} be the _sum of R _{i, j, k} from j = 1 to M. .

In the k-th stage, L Lagrangian multipliers λ _{1, j, k} , λ _2, for the j-th block based on the coding information held up to the (k−1) -th stage _{. j, k} , λ _{3, j, k} ,..., λ _{L, j, k} are selected (steps S41 to S45). The processing in step S45 will be described in detail later by taking the (k + 1) th stage as an example. In step S45, if k = 1, the initial value λ _{i, j, 1} is given.

Among the encoding syntaxes that are candidates for encoding the j-th block, the encoding syntax that minimizes D + λ _{i, j, k} R in _{the j-th} block is i = 1, 2, 3, .., L are selected (step S46).

Next, from the relationship between λ _{i, j, k} and R _{i, j, k} for i = 1, 2, 3,..., L, R = C _{j, k} as the λ dependency of R in the j-th block. _{, 1} / (λ ^1/2 ), C _{j, k, 1} is calculated from a method such as the least squares method (step S47).

Similarly, the processing of the (j + 1) -th block is repeated until j becomes 1 to M (step S49), and the sum of C _{j, k, 1} from j = 1 to M is calculated as C _{sum, k, 1} And _{Csum, k, 1} is held as encoded information (step S50).

It is determined whether the k-th stage is the last stage (step S51). If the k-th stage is not the last stage, the process of the (k + 1) -th stage is performed (step S52). The determination as to whether or not it is the last stage is λ _{i, j} when the absolute value of the difference between R _{i, sum, k} and R _target is smaller than a certain value and j = 1, 2, 3 _{,. ,} depending on whether _k is within a certain range.

In the (k + 1) th stage, in step S45, λ _{1, j, k + 1} , λ _{2, j, k + 1} , λ _{3, j, k + 1} ,..., Λ _{L, j, k + 1} By selecting λ _{i, j, k + 1} so that either of them becomes equal to (C _{sum, k, 1} / R _target ) ² , the convergence of the Lagrange multiplier giving the code amount close to R _target is accelerated.

The absolute value of the difference between R _{i, sum, k} and R _target is smaller than a certain value, and λ _{i, j, k} is within a certain range when j = 1, 2, 3,. Where i and k are i ₀ and N, respectively, and in the Nth stage, which is the last stage, D + λ _{i0, j} from among the encoding syntaxes that are candidates for encoding the jth block. _{, N} R is repeatedly output from j = 1 to M to output the encoded syntax as an encoded stream for the j-th block (step S53).

In the second embodiment, an example of calculating C _{j, k, 1} when R = C _{j, k, 1} / (λ ^1/2 ) is assumed as the λ dependency of R in the j-th block. As described above, it is not necessarily limited to this formula, and it goes without saying that another formula may be used as the λ dependence of R. The formula to be used can be determined based on information obtained through experiments.

  In the second embodiment, an example in which the number of parameters P is 1 is described as the λ dependency of R in the j-th block. However, the number of parameters P is not necessarily limited to 1, and an R including a plurality of parameters is used. It goes without saying that the λ-dependency formula may be used.

[Third Embodiment]
The third embodiment of the present invention corresponds to the case where the first invention, the eighth invention, and the ninth invention are employed at the same time. From the first stage to the N'th stage, λ _{i, j, k} is calculated using the method of the sixth aspect of the present invention. However, even if k is increased, such as an image to which the model does not apply _, R _{i, sum, If} the absolute value of the difference between _k and R _target does not become smaller than a certain value, after the (N ′ + 1) th stage, the method of the third aspect of the present invention is used for R _{i, sum, k} and R _target . Make sure that the absolute value of the difference is small.

As the first to third embodiments, the representative method combining some of the present invention has been described, but from the above description, other combinations of the first to ninth present invention can be easily implemented. Is clear.

  The above moving image encoding processing can be realized not only by hardware and firmware but also by a computer and a software program, and the program is recorded on a computer-readable recording medium and provided. Or can be provided through a network.

It is a figure which shows the structural example of a moving image encoder. It is a figure which shows the example of an encoding process flow. It is a figure which shows the example of an encoding process flow. It is a figure which shows the example of an encoding process flow.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Moving image encoder 10 Image input part 11 λ candidate selection part 12-i Syntax selection part (i = 1, 2,..., L) for λ _{i, j, k}
13 Evaluation unit for λ 14 Encoded stream output unit

Claims (13)

In a video encoding method by an encoding device that calculates a code amount distortion D and a code amount R and determines an encoding syntax that minimizes D + λR based on a Lagrange multiplier λ,
L, M, N are integers of 2 or more, i is an integer from 1 to L, j is an integer from 1 to M, k is an integer from 1 to N , and the first block to the Mth block Let R _{target be the} overall target code amount, and let R _{i, j, k} , j be from 1 in the encoding syntax that minimizes D + λ _{i, j, k} R in _{the j-th} block at the k-th stage. Let R _{i, sum, k be the sum} of R _{i, j, k} up to M ,
In the k-th stage,
The block of the j, the (k-1) code quantity is retained at the stage of R _{i, sum,} using the coded information of _k-1, the target code amount R _target and the code amount R _From the magnitude relationship with _{i, sum, k-1} , the difference between the maximum and minimum values of L Lagrange multipliers λ _{i, j, k} is L Lagrange multipliers λ in the (k−1) _-th stage. _The L Lagrange multipliers λ _{i, j, k} (provided as initial values when k = 1 so that the difference between the maximum and minimum values of _{i, j, k−1} is smaller) Selecting the given L Lagrange multipliers λ _{i, j, 1} ),
Among the encoding syntaxes that are candidates for encoding the j-th block, the encoding syntax that minimizes D + λ _{i, j, k} R in _{the j-th} block is i = 1, 2, 3 ,..., L are repeatedly selected until j becomes 1 to M.
The process of holding the encoding information of the code amount R _{i, sum, k in} the selected encoding syntax in _{the k-th} stage, adding 1 to k, and moving to the next stage is changed from k to 1 Repeat until
In the Nth stage,
D + λ _{1, j, N} R, D + λ _{2, j, N} R, D + λ _{3, j, N} R,..., D + λ _L, out of the encoding syntaxes that are candidates for encoding the j th block _. moving picture encoding method for minimizing one of _{j and N} R as an encoded stream for the j-th block is repeated until j becomes 1 to M. . The moving image encoding method according to claim 1,
Λ _{i, j,} when the absolute value of the difference between the code amount R _{i, sum, k} and the target code amount R _target is smaller than a certain value and j = 1, 2, 3,. _{When k} is within a certain range, i is i ₀ and k is N.
In the Nth stage,
A step of outputting an encoding syntax that minimizes D + λ _{i0, j, N} R as an encoded stream for the j-th block from among encoding syntaxes that are candidates for encoding the j-th block Is repeated until j becomes 1 to M. In the moving image encoding method according to claim 1 or 2,
The Lagrangian multiplier is determined so that λ _{i, 1, k} = λ _{i, 2, k} = λ _{i, 3, k} =, ..., = λ _{i, M, k} holds for all i and k. moving picture coding method, characterized in that Monodea Ru. In the moving image encoding method according to claim 2 or 3,
In the step of selecting the previous Symbol L number of Lagrange multiplier λ _{i, j, k} Te stage smell of the k,
λ _{1, j, k} <λ _{2, j, k} <λ _{3, j, k} ≦ <λ _{L, j, k} holds for all j and k, and i = 1, 2, 3,. , L _, the maximum value of R _{i, sum, k} is greater than R _target and i = 1,2,3, ..., L _, the minimum value of R _{i, sum, k} is determined from R _target The Lagrange multiplier λ _{i, j, k} is determined to be small,
From the value of the code amount R _{i, sum, k−1} and the target code amount R _target held until the (k−1) -th stage in the k-th stage after the second stage, λ _{1, 1, k-1} ≤λ _{1,1, k} and λ _{L, 1, k} ≤λ _{L, 1, k-1} and λ _{L, 1, k} −λ _{1,1, k} <λ _{L, 1, k-} Λ _{i, 1, k} (i = 1, 2, 3,..., L) is determined such that ₁₋ λ _{1,1, k−1} holds. In the moving image encoding method according to claim 3,
L is 3, and in the k-th stage, λ _{1, j, k} <λ _{2, j, k} <λ _{3, j, k} and R _{1, sum, k} > R _target > R _{3, sum,} three Lagrange multipliers so that the _{k λ 1, j, k,} λ 2, j, k, λ 3, j, elected _k,
In the k-th stage after the second stage, if R _{2, sum, k-1} <R _target , λ _{1,1, k} = λ _{1,1, k-1} and λ _{2,1, k} = (Λ _{1,1, k-1} + λ _{2,1, k-1} ) / 2 and λ _{3,1, k} = λ _{2,1, k-1} and R _{2, sum, k-1} > R _target In this case, λ _{1,1, k} = λ _{2,1, k-1} and λ _{2,1, k} = (λ _{2,1, k-1} + λ _{3,1, k-1} ) / 2 and λ _{3, 1, k} = λ _{3,1, k-1}
The encoded stream is output in the Nth stage, where k is N when the absolute value of the difference between the code amount R2 _{, sum, k} and the target code amount _Rtarget is smaller than a certain value. A video encoding method. In a video encoding method by an encoding device that calculates a code amount distortion D and a code amount R and determines an encoding syntax that minimizes D + λR based on a Lagrange multiplier λ,
L, M, N are integers of 2 or more, i is an integer from 1 to L, j is an integer from 1 to M, k is an integer from 1 to N, P is an integer of 1 or more, and p is from 1 and an integer of up to P, and the target code amount of the whole from the first block to the block in the M and R _target, to the D in the j blocks + lambda _{i, j} of stages of the _k, the _k R to the minimum The encoding amount in the encoding syntax is R _{i, j, k} , where j is 1 to M _, the sum of R _{i, j, k} is R _{i, sum, k} , and the λ dependency of the encoding amount R is predetermined. other expressions of P parameters C _{j, k, p,} C _j from j 1 to _{M, k,} the sum of _p C _{sum, k,} as _p,
In the k-th stage,
For the j-th block, using the encoding information of the sum C _{sum, k−1, p} of the parameters held in the (k−1) -th stage, the sum of the target code amount R _target and the parameters From _{Csum, k-1, p} , the difference between the maximum and minimum values of the L Lagrange multipliers λ _{i, j, k} is the L Lagrange multipliers λ _{i, The} L Lagrange multipliers λ _{i, j, k} (provided as initial values when k = 1 so that the difference between the maximum value and the minimum value of _{j, k-1} is smaller ) Selecting L Lagrange multipliers λ _{i, j, 1} );
Among the encoding syntaxes that are candidates for encoding the j-th block, the encoding syntax that minimizes D + λ _{i, j, k} R in _{the j-th} block is i = 1, 2, 3 , ..., L to select for each
Based on the relationship between the L Lagrange multipliers λ _{i, j, k} and the code amount R _{i, j, k} in the selected coding syntax , a predetermined value representing the λ dependency of the code amount R in the j-th block is determined. Repeating the step of calculating C _{j, k, p} for the P parameters of the given expression until j becomes 1 to M,
A process of holding the encoding information of the sum C _{sum, k, p} of the calculated P parameters C _{j, k, p} in the k-th stage, adding 1 to k, and moving to the next stage is performed. Repeat until 1 goes to N,
In the Nth stage,
D + λ _{1, j, N} R, D + λ _{2, j, N} R, D + λ _{3, j, N} R,..., D + λ _L, out of the encoding syntaxes that are candidates for encoding the j th block _. A video encoding method characterized by repeating the step of outputting an encoding syntax that minimizes one of _{j and N} R as an encoded stream for the j-th block until j becomes 1 to M. . In the moving image encoding method according to claim 6,
P is 1, and from the relationship between λ _{i, j, k} and R _{i, j, k} for i = 1, 2, 3,..., L, the λ dependency of the code amount R in the j-th block , R = C _{j, k, 1} / (λ ^1/2 ), C _{j, k, 1} is calculated,
When the sum of C _{j, k, 1} where j is 1 to M is C _{sum, k, 1} , at the kth stage, λ _{1, j, k} , λ _{2, j, k} , λ _{3, j , k} ,..., λ _{L , j, k} is selected so that λ _{i, j, k} is selected to be equal to (C _{sum, k−1,1} / R _target ) ² Encoding method. The moving image encoding method according to claim 1,
The moving picture coding method, wherein L and M are variable with respect to k. A moving picture coding method, wherein either the moving picture coding method according to claim 3 or the moving picture coding method according to claim 6 is selected for each k-th stage. In a video encoding device that calculates a code amount distortion D and a code amount R and determines an encoding syntax that minimizes D + λR based on a Lagrange multiplier λ,
An image input means for inputting an image to be encoded;
L, M, N are integers of 2 or more, i is an integer from 1 to L, j is an integer from 1 to M, k is an integer from 1 to N , and the first block to the Mth block Let R _{target be the} overall target code amount, and let R _{i, j, k} , j be from 1 in the encoding syntax that minimizes D + λ _{i, j, k} R in _{the j-th} block at the k-th stage. Let R _{i, sum, k be the sum} of R _{i, j, k} up to M ,
k is the phase of each of the k from 1 to (N-1), the block of the j, the (k-1) retained by the code amount R _i at the stage _{of, sum, k-1} of Using the encoding information _, the maximum and minimum values of L Lagrange multipliers λ _{i, j, k} are calculated from the magnitude relationship between the target code amount R _target and the code amount R _{i, sum, k−1} . difference, so that the difference value smaller than the first L number of Lagrange multipliers lambda _{i, j,} the maximum value of the _k-1 and the minimum value in the stage of (k-1), the L Lagrangian multiplier lambda _{i , j, k} (where K = 1, L Lagrange multipliers λ _{i, j, 1} given as initial values) are selected as Lagrange multiplier candidate selection means;
For each selected Lagrange multiplier λ _{i, j, k} (i = 1, 2, 3,..., L), an encoding syntax that minimizes D + λ _{i, j, k} R in the jth block. A syntax selection means for selecting
Encoding information of the code amount R _{i, sum, k in} the selected encoding syntax in the k-th stage by the syntax selection means is held, and syntax selection for the Lagrange multiplier candidate selection means and the Lagrange multiplier Evaluation means for performing control by repeating the processing by the means until k becomes 1 to N;
D + λ _{1, j, N} R, D + λ _{2, j, N} R, D + λ _{3, j, N} R among the encoding syntaxes that are candidates for encoding the j th block in the Nth stage. ,..., D + λ _{L, j, N} is provided with encoded stream output means for outputting an encoded syntax that minimizes any one of R as an encoded stream for the j-th block. Device. In a video encoding device that calculates a code amount distortion D and a code amount R and determines an encoding syntax that minimizes D + λR based on a Lagrange multiplier λ,
An image input means for inputting an image to be encoded;
L, M, and N are integers of 2 or more, i is an integer from 1 to L, j is an integer from 1 to M, k is an integer from 1 to N, P is an integer of 1 or more, and p is from 1 As an integer up to P, the total target code amount from the first block to the M-th block is R _target D + λ in the j-th block at the k-th stage _{i, j, k} Let R be the amount of code in the coding syntax that minimizes R _{i, j, k} , J is R from 1 to M _{i, j, k} The sum of R _{i, sum, k} , P parameters of a predetermined formula representing the λ dependency of the code amount R are represented by C _{j, k, p} , J from 1 to M _{j, k, p} The sum of C _{sum, k, p} As
The sum C of the parameters held in the (k−1) -th stage for the j-th block in each k-th stage from k = 1 to (N−1). _{sum, k-1, p} Is used to encode the target code amount R _target And the sum C of the parameters _{sum, k-1, p} And L Lagrange multipliers λ _{i, j, k} Is the L Lagrange multipliers λ in the (k−1) th stage. _{i, j, k-1} L Lagrange multipliers λ so that the difference is smaller than the difference between the maximum value and the minimum value of _{i, j, k} (However, when k = 1, L Lagrange multipliers λ given as initial values are used. _{i, j, 1} ) Candidate selection means for selecting Lagrange multipliers,
Among the encoding syntaxes that are candidates for encoding the j-th block, D + λ in the j-th block _{i, j, k} An encoding syntax that minimizes R is selected for each of i = 1, 2, 3,..., L, and the L Lagrange multipliers λ are selected. _{i, j, k} And the code amount R in the selected coding syntax _{i, j, k} From the relationship with the above, P parameters of a predetermined formula representing the λ dependency of the code amount R in the j-th block are represented by C _{j, k, p} A syntax selection means for calculating
P parameters C at the k-th stage calculated by the syntax selection means _{j, k, p} Sum C _{sum, k, p} Evaluation means for holding the encoding information of the Lagrangian multiplier and processing by the syntax selection means for the Lagrangian multiplier until k becomes 1 to N.
Among the encoding syntaxes that are candidates for encoding the j-th block in the N-th stage, D + λ _{1, j, N} R, D + λ _{2, j, N} R, D + λ _{3, j, N} R, ..., D + λ _{L, j, N} Coding stream output means for outputting a coding syntax that minimizes any of R as a coded stream for the j-th block.
A moving picture coding apparatus characterized by the above. A moving picture coding program for causing a computer to execute the moving picture coding method according to any one of claims 1 to 9 . A computer-readable recording medium on which a moving image encoding program for causing a computer to execute the moving image encoding method according to any one of claims 1 to 9 is recorded.

Images