JP6495834B2 - Video encoding method, video encoding apparatus, and video encoding program - Google Patents

Description

  The present invention relates to a video encoding method, a video encoding device, and a video encoding program.

  Conventionally, as a typical video coding standard, the video coding international standard H.264 has been used. H.264 / AVC (Advanced Video Coding, see Non-Patent Document 1, for example) and H.264 / AVC. H.265 / HEVC (High Efficiency Video Coding, see Non-Patent Document 2, for example) is known. In order to perform video coding using these standards with high efficiency, there is “entropy coding” as the last stage process for outputting a compressed bit string code. H. H.264 / AVC and H.264. In H.265 / HEVC, a CABAC (Context Adaptive Binary Arithmetic Coding) method is used as the entropy coding.

  In high-efficiency video encoding, there are many options called encoding modes when encoding individual encoding units (blocks obtained by dividing a frame constituting a video into a predetermined size). An appropriate encoding mode can be selected. At this time, it is important which of the coding modes is optimal. That is, it is necessary to select an encoding mode that minimizes distortion at a constant code amount, in other words, an encoding mode that minimizes code amount under a certain amount of distortion. A method based on Lagrange's undetermined multiplier method is used to select an optimal encoding mode, which is called “RD optimization (Rate-Distortion optimization)”.

There are three variables used in RD optimization. The first “R” is the number of bits when the mode is selected and the encoding unit is encoded (hereinafter referred to as “code amount estimation value”). The second “D” is the sum of squared errors between the decoded signal and the original signal. The third “λ” is a Lagrange undetermined multiplier. λ is arbitrarily selected by the video encoding apparatus according to the target image quality and code amount. Then, for each selectable encoding mode, an encoding mode that minimizes the quantity C called “cost function” in equation (1) is selected.
C = D + λR (1)

  H. H.265 / HEVC and H.264. In actual coding software of H.264 / AVC, when RD optimization is performed, the generated code amount is often obtained by estimation without performing actual CABAC coding. This is because the amount of processing of CABAC encoding is relatively large, so that the amount of processing for obtaining the code amount estimation value R for each selectable encoding mode is avoided.

  As such a generated code amount estimation method, for example, the method of Non-Patent Document 3 obtains an estimated code amount by linear regression from several numerical values appearing in the preceding stage of encoding.

  As a more general code amount estimation method, there is a method using an occurrence probability of a signal to be encoded (see, for example, Non-Patent Documents 4, 5, and 6). By these methods, it is possible to reduce the processing amount by 1 to 25% while estimating the CABAC generated code amount with high accuracy.

For example, the method of Non-Patent Document 6 is described in H.H. This is a method used in the reference software of H.265 / HEVC, and the code amount estimation value R is obtained by the equation (2) using the already output bit amount Bw and the fixed-point number Bf.
R = Bw + (Bf >> 15) (2)
In Expression (2), “>>” is a right bit shift operation. That is, in the equation (2), the code amount estimated value R is obtained by adding Bw to the value obtained by shifting the fixed-point number Bf to the right by 15 bits.

ISO / IEC 14496-10: 2014 Information technology-Coding of audio-visual objects-Part 10: Advanced Video Coding ISO / IEC 23008-2: 2015 Information technology-High efficiency coding and media delivery in heterogeneous environments-Part 2: High efficiency video coding S. Guo, Z. Liu, D. Wang, Q. Han and Y. Song, "Linear rate estimation model for HEVC RDO using binary classification based regression" Data Compression Conference 2014, p. 406, Snowbird, March 2014 J. Hahm and C.-M. Kyung, "Efficient CABAC rate estimation for H.264 / AVC mode decision," IEEE Trans. CSVT, vol. 20, no. 2, pp. 310-316, February 2010. K. Won, J. Yang and B. Jeon, "Fast CABAC rate estimation for H.264 / AVC mode decision," Electronics Letters, vol. 48, no. 19, pp. 1201-1203, September 2012. F. Bossen, "CE1: Table-based bit estimation for CABAC," JCTVCG763, Geneva, November 2011

  By the way, in any of the methods described in Non-Patent Documents 3, 4, 5, and 6, although the estimated number of decimals was obtained in the process of calculating the estimated code amount, the estimated code is rounded or rounded off. The quantity was an integer. This was considered natural because the code generated by CABAC is a sequence of 0 or 1 bits.

However, the code amount of the signal occurring with the probability p is exactly log ₂ (p) bits and may not necessarily be an integer. Where log ₂ (p) is the logarithm of p with 2 as the base. For example, the code amount of a signal that occurs with a probability of 0.6 is about 0.737 bits. Thus, the code amount generally takes a decimal value of 0 or more.

  Therefore, in the conventional method, the estimated code amount is rounded or rounded down to round the numerical value to an integer, so that the code amount estimated value R includes a rounding error of less than 1 bit. Since it is originally an estimated value, there is naturally an error with the actual CABAC code amount. Conventionally, however, RD optimization is performed using a value that includes the above-described rounding error in addition to the error with the actual code amount. It was broken. That is, since the cost function C based on the code amount estimation value R that contains a relatively large amount of error has been minimized, there is a possibility that an optimal encoding mode may not be obtained, and further, the encoding efficiency There is a problem that decreases.

The rounding error of less than 1 bit is a negligible amount when the code amount estimation value R is as large as several thousand bits. For example, when the probability is 0.6, if the code amount is rounded to 1 bit, The ratio of the error added to the original code amount is 35.7% as shown in the equation (3), which occupies a large ratio that cannot be ignored.
(1-log ₂ (p)) / log ₂ (p) × 100 = 35.7 [%] (3)

  The present invention has been made in view of such circumstances, and an object thereof is to provide a video encoding method, a video encoding device, and a video encoding program capable of improving the accuracy of the code amount estimation value R. To do.

  One aspect of the present invention is a video encoding method performed by a video encoding device that encodes a signal of an input video, and floating calculation of a code amount estimation value used for cost function calculation when an encoding mode is selected. This is a video encoding method including a code amount estimated value calculation step performed using decimal point arithmetic.

  One aspect of the present invention is the video encoding method, wherein the code amount estimation value calculation step calculates an integer part Bw of a code amount of a coding unit and a decimal part Bf of the code amount, The estimated code amount is calculated by “Bw + Bf ÷ predetermined value”.

One aspect of the present invention is the video encoding method, wherein the predetermined value is 2 ⁿ when the number of bit shifts when the code amount estimation value is calculated by a bit shift operation is ⁿ . .

  One aspect of the present invention is the video encoding method, wherein the predetermined value is 32768.

  One aspect of the present invention is a video encoding apparatus that encodes an input video signal, and performs calculation of a code amount estimation value used for cost function calculation when a coding mode is selected by using a floating-point operation. It is a video coding apparatus provided with a code amount estimated value calculation part.

  One aspect of the present invention is a video encoding program for causing a computer to execute the video encoding method.

  According to the present invention, since the code amount estimated value R is obtained using floating point arithmetic without using a fixed decimal point bit shift, the accuracy of the code amount estimated value R can be improved. By improving the accuracy of the code amount estimation value R, it is possible to refine RD optimization in video encoding, so that a more appropriate encoding mode is selected, and a higher quality is achieved with a smaller code amount. An effect of obtaining a decoded video is obtained.

It is a block diagram which shows the structure of a general video coding apparatus. It is a flowchart which shows operation | movement of the video coding apparatus shown in FIG. It is a flowchart which shows operation | movement of the encoding control part 115 shown in FIG. It is a flowchart which shows operation | movement of step S16 which calculates the code amount estimated value R shown in FIG.

  A video encoding apparatus according to an embodiment of the present invention will be described below with reference to the drawings. Note that in this specification, an image means a still image or an image for one frame constituting a moving image. A video has the same meaning as a moving image, and is a set of a series of images.

  First, H.C. H.265 / HEVC or H.264 A configuration of a general video encoding device including H.264 / AVC will be described. FIG. 1 is a block diagram showing a configuration of a general video encoding device 10. The video encoding device 10 shown in FIG. 1 receives the video signal 100 to be encoded, divides it into blocks, and encodes each block, thereby generating and outputting smaller encoded data 118.

  The video encoding device 10 includes a subtraction unit 102, a transform / quantization unit 104, an inverse quantization / inverse transform unit 106, an addition unit 107, a distortion removal filter 108, a frame memory 109, an in-screen prediction unit 111, and a motion estimation unit 112. , An inter-screen prediction unit 114, an encoding control unit 115, and an entropy encoding unit 117.

  When the configuration of the video encoding device 10 is described with reference to FIG. 1, the well-known functions and configurations that the video encoding device 10 normally has are described unless they are directly related to the description of the present invention. And illustration of a structure is abbreviate | omitted.

  Next, the operation of the video encoding device 10 shown in FIG. 1 will be described with reference to FIG. FIG. 2 is a flowchart showing the operation of the video encoding device 10 shown in FIG. In this case, the video encoding device 10 is H.264. In the following description, it is assumed that the video encoding apparatus conforms to H.265 / HEVC.

  First, a video signal input from the outside is divided into blocks for each processing unit called a prediction unit (Prediction Unit). The subtraction unit 102 subtracts the prediction signal 101 from the video signal 100 for each prediction unit, and generates and outputs a prediction residual signal 103 from the result (step S1). The prediction signal 101 is a signal generated based on one of the prediction results as a result of performing intra prediction by the intra prediction unit 111 or inter prediction by the inter prediction unit 114.

  Next, the transform / quantization unit 104 transforms and quantizes the prediction residual signal 103 and outputs it (step S2). The conversion referred to here is to convert the information to be compressed into another information representation form in which the semantic content is not lost in order to further increase the compression rate. For this conversion, DCT (Discrete Cosine Transform), DST (Discrete Sine Transform), or the like is used.

  Next, the entropy encoding unit 117 entropy-encodes the quantization coefficient 105 that is the result of the transform / quantization unit 104 and outputs the result as encoded data 118 (step S3).

  On the other hand, the value of the quantization coefficient 105 is subjected to inverse quantization / inverse transform in the inverse quantization / inverse transform unit 106 (step S4). The inverse transformation here is to perform inverse DCT (inverse discrete cosine transformation) or inverse DST (inverse discrete sine transformation).

  Next, the addition unit 107 adds the prediction signal 101 output from either the intra-screen prediction unit 111 or the inter-screen prediction unit 114 and the output signal of the inverse quantization / inverse conversion unit 106. Subsequently, the distortion removal filter 108 performs distortion removal on the signal output from the addition unit 107 (step S5).

  The output of the distortion removal filter 108 is a reproduction of the converted decoded image, and the signal is output as the decoded video signal 110 (step S6). The decoded video signal 110 is stored in the frame memory 109 (step S7) and is output to the encoding control unit 115.

  The motion estimation unit 112 uses the decoded video signal 110 and the video signal 100 stored in the frame memory 109 to estimate how much the target block has moved over time, and generates motion information 113 as a result of the estimation. Output. Using the motion information 113 and the decoded video signal 110 stored in the frame memory 109, the inter-screen prediction unit 114 generates the prediction signal 101 of the target block. Alternatively, the intra-screen prediction unit 111 generates and outputs the prediction signal 101 of the target block using the decoded video signal 110 stored in the frame memory 109 (step S8).

  The entropy encoding unit 117 encodes the control data 116 output from the encoding control unit 115, the quantization coefficient 105 output from the transform / quantization unit 104, and the motion information 113 output from the motion estimation unit 112. Is output as converted data 118.

  In the series of processes described above, some processing units operate based on the control data 116 output from the encoding control unit 115. The encoding control unit 115 refers to the video signal 100 and the decoded video signal 110 and outputs control data. For example, in the transform / quantization unit 104 and the inverse quantization / inverse transform unit 106, the control data 116 designates a quantization width and a block size to be transformed.

  In addition, the encoding control unit 115 specifies whether the target block is predicted by intra prediction or inter prediction in the intra prediction unit 111 and the inter prediction unit 114. Then, the encoding control unit 115 performs prediction from two frames for what prediction method is used for intra-screen prediction, what block size is used for prediction, and for inter-screen prediction. Specify whether to start from one frame. In addition, the encoding control unit 115 outputs a size of a unit for performing inter-screen prediction and a past encoded frame to be referred to in the motion estimation unit 112.

  The “quantization width” and “block size to be converted” information specified by the control data 116 output from the encoding control unit 115 described here are called “modes”. By selecting a mode with the smallest cost function from among a plurality of mode options, the highest image quality can be obtained at a given bit rate.

  Next, the operation of the encoding control unit 115 shown in FIG. 1 will be described with reference to FIG. FIG. 3 is a flowchart showing the operation of the encoding control unit 115 shown in FIG. First, the encoding control unit 115 inputs a value of an undetermined multiplier λ given separately (step S11). Subsequently, the encoding control unit 115 inputs the video signal 100 of the target block (step S12).

  Next, the encoding control unit 115 sequentially lists modes that can be tried in the processing unit in the target video encoding device 10 and sets them as the trial mode (step S13). The encoding control unit 115 attempts encoding using the set trial mode (step S14), and sets the sum of the squares of the difference between the obtained decoded image signal and the original image signal as the distortion amount D (step S15). .

  Also, the encoding control unit 115 estimates the corresponding code amount estimation value R by the technique described in Non-Patent Document 6 or the like (step S16). Then, the encoding control unit 115 calculates the cost function C based on the equation (1) (step S17). The encoding control unit 115 determines whether or not the obtained cost function C is the smallest among the trials (step S18), and if it is the smallest, stores the mode at this time (step S19).

  Next, it is determined whether all trialable modes have been completed in the target processing unit (step S20), and if not completed, the process returns to step S13 to repeat the process. If completed, the encoding control unit 115 outputs the stored mode in which the cost function C is minimum (step S21), and the process ends.

  Next, the details of the operation of step S16 for calculating the code amount estimation value R shown in FIG. 3 will be described with reference to FIG. FIG. 4 is a flowchart showing the operation of step S16 in which the encoding control unit 115 calculates the code amount estimated value R shown in FIG.

The specific method of Non-Patent Document 6 is described in H.C. Detailed description is omitted because it is described in the reference software of H.265 / HEVC. First, the encoding control unit 115 calculates the integer code amount Bw from the internal state of the CABAC when the encoding is attempted in the designated mode (step S41). Subsequently, the encoding control unit 115 calculates a decimal code amount Bf (step S42). Finally, the encoding control unit 115 calculates and outputs a code amount estimated value R based on the equation (4) (step S43). Equation (4) is calculated using an arithmetic operation using a floating point.
R = Bw + Bf ÷ 32768
= Bw + Bf ÷ 2 ¹⁵ (4)

  Thus, by calculating the code amount estimated value R using a floating point, it is possible to improve the calculation accuracy rather than calculating by bit shift. With this configuration, a roundoff error is not included when the code amount estimated value R is calculated, and the code amount estimated value R can be calculated accurately, so that the possibility of obtaining an optimal encoding mode is increased. Thus, the effect that the encoding efficiency can be expected to be improved is obtained.

  Note that the value “32768” does not have to be a fixed value, and even if it fluctuates by about ± 1% (32440 to 33095), the decoded image can be prevented from being affected, and the right bit shift can be performed. The accuracy can be improved compared with the calculation using.

In the above description, an example in which division is performed by 32768 (= 2 ¹⁵ ) has been shown. However, when the expression (2) is set to R = Bw + (Bf >> n), that is, when n bits are shifted to the right, The equation (4) may be the equation (4) ′.
R = Bw + Bf ÷ 2 ⁿ (4) ′
In this case as well, division may be performed by floating point arithmetic. However, ²ⁿ is calculated in advance, and the calculation speed is improved by substituting the value.

  Note that the floating-point arithmetic using the equations (4) and (4) ′ can be applied to any encoding software having a process for converting the real number R into an integer at the final stage of the process of outputting the code amount estimation value. It is. For example, the present invention can be applied to the calculation processes described in Non-Patent Documents 4 and 5.

  As described above, the code amount estimation procedure is configured to perform estimation with less error. According to this configuration, it is possible to reduce the possibility of erroneous selection of a coding mode that increases efficiency in coding optimization, which is a key to improving image quality in video coding. Therefore, higher encoding efficiency is realized in video encoding, and higher quality decoded video can be obtained with a smaller code amount.

  You may make it implement | achieve all or one part of the video coding apparatus in embodiment mentioned above with a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory inside a computer system serving as a server or a client in that case may be included and a program held for a certain period of time. Further, the program may be a program for realizing a part of the above-described functions, and may be a program capable of realizing the functions described above in combination with a program already recorded in a computer system. It may be realized using hardware such as PLD (Programmable Logic Device) or FPGA (Field Programmable Gate Array).

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Therefore, additions, omissions, substitutions, and other modifications of the components may be made without departing from the technical idea and scope of the present invention.

  In lossy encoding of images / videos, the present invention can be applied to applications where it is indispensable to encode / decode images for the purpose of improving video quality and reducing the encoding bit rate.

  DESCRIPTION OF SYMBOLS 100 ... Video signal 101 ... Prediction signal 102 ... Subtraction part 103 ... Prediction residual signal 104 ... Transformation / quantization part 105 ... Quantization coefficient 106 ... Inverse quantization / inverse transformation part 107 ... Addition part , 108 ... distortion removal filter, 109 ... frame memory, 110 ... decoded video signal, 111 ... intra prediction unit, 112 ... motion estimation unit, 113 ... motion information, 114 ... inter prediction unit, 115 ... encoding control unit, 116: Control data, 117: Entropy encoding unit, 118: Encoded data

Claims (4)

A video encoding method performed by a video encoding device that encodes an input video signal,
Have a code amount estimation value calculation step performed using floating point arithmetic calculations code amount estimation value used for the cost function calculation in the case of selecting the coding mode,
In the code amount estimated value calculation step, an integer part Bw of a code amount of a coding unit and a decimal part Bf of the code amount are calculated, and the code amount estimated value is calculated by “Bw + Bf ÷ predetermined value”.
The predetermined value is either a predetermined fixed value or a value that varies ± 1% from the fixed value.
Video encoding method. The fixed value, the video encoding method according to claim 1 which is 32768. A video encoding device that encodes an input video signal,
A code amount estimated value calculation unit that performs calculation of a code amount estimated value used for cost function calculation in the case of selecting an encoding mode using a floating-point operation ,
The code amount estimated value calculation unit calculates an integer part Bw of a code amount of a coding unit and a decimal part Bf of the code amount, calculates the code amount estimated value by “Bw + Bf ÷ predetermined value”,
The predetermined value is either a predetermined fixed value or a value that varies ± 1% from the fixed value.
Video encoding device. A video encoding program for causing a computer to execute the video encoding method according to claim 1 .

Images