JP5306282B2 - Moving picture coding apparatus, moving picture coding method, and program - Google Patents

Description

  The present invention relates to a moving image encoding apparatus, a moving image encoding method, and a program.

  Currently, when encoding a moving image, predictive encoding and block division transform encoding are performed. In predictive coding, a prediction error signal is generated by performing intraframe predictive coding and motion compensation interframe predictive coding (see, for example, Non-Patent Document 1). The prediction error signal is a signal representing a difference between a pixel value in a macroblock and a predicted value of the pixel value. A macroblock is an area in which a plurality of adjacent pixels (for example, 8 × 8 pixels) are collected, and is an object to be encoded. In intra-frame predictive coding, a high compression rate is realized by predicting pixel values in a macroblock from pixel values of adjacent coded macroblocks. On the other hand, in motion compensation interframe predictive coding, in addition to interframe predictive coding using a difference signal between adjacent frames, information indicating how much an object in a frame has moved from a coded frame (motion vector) ) Is used to predict the pixel value of the macroblock. Thereby, a high compression rate is realized. In block division transform coding, a prediction error signal is coded by intraframe prediction coding and motion compensation interframe prediction coding. That is, a high compression rate is realized by performing discrete cosine transform (DCT) on the prediction error signal in units of macroblocks.

FIG. 4 is a block diagram showing a functional configuration of the moving picture coding apparatus 20 that performs coding combining prediction coding and block division transform coding.
The moving picture coding apparatus 20 includes a predictive coding unit 21, a block division transform coding unit 22, and a quantization unit 23. The predictive encoding unit 21 receives a video signal to be encoded (hereinafter referred to as an input video signal), and performs intraframe prediction encoding and motion compensation interframe prediction encoding on a macroblock in the input video signal. I do. Then, the predictive encoding unit 21 outputs the generated prediction error signal to the block division transform encoding unit 22. The block division transform encoding unit 22 performs discrete cosine transform on the input prediction error signal, thereby transforming the frequency component into a frequency component and using the transformed frequency component as an orthogonal transform coefficient. This removes spatial redundancy. The quantization unit 23 rounds the result obtained by dividing the orthogonal transform coefficient transformed by the block division transform coding unit 22 by a specified value (hereinafter referred to as a quantization step) to an integer value, and quantizes the result rounded to an integer value. It is a digitized value. That is, the quantization unit 23 associates the input orthogonal transform coefficient with one of the predetermined quantization representative values. As a result, the code amount is significantly reduced while allowing distortion due to quantization.

  Here, when performing encoding, the quantization unit 23 first determines a quantization parameter (an integer value from 0 to 51) for each macroblock, and performs a quantization step based on the determined quantization parameter. Is derived. The quantization parameter is a parameter that determines the quantization step. For example, H.M. In H.264 / AVC, the logarithm of the quantization parameter and the quantization step is proportional. Specifically, when the quantization parameter is increased by 6, the quantization step is doubled. Then, the quantization unit 23 calculates a quantization value by rounding the result obtained by dividing the orthogonal transform coefficient by the quantization step to an integer value.

  Non-Patent Document 2 describes SSIM (Structural SIMality), which is a method for objectively evaluating the similarity of images.

ITU-T Rec. H.264-ISO / IEC 14496-10, "Advanced video coding for generic audiovisual service", Mar. 2010. http://www.itu.int/rec/T-REC-H.264 2009 , 2009. Zhou Wang, Alan Conrad Bovik, Hamid Rahim Sheikh, Eero P. Simoncelli, "Image Quality Assessment: From Error Visibility to Structural Similarity," IEEE Transactions on Image Pro-cessing, vol. 13, no. 4, pp. 600-612, Apr. 2004.

  In the prior art, the amount of code is reduced by performing quantization in the quantizing unit 23. However, since distortion due to quantization is allowed at this time, distortion may occur. However, it is impossible to know in advance how much the generated distortion is and how much it is distorted. Therefore, it is difficult to optimize in advance the quantization parameter used for quantization, and as a result, there is a problem that distortion due to quantization may adversely affect image quality.

  The present invention has been made in consideration of such circumstances, and its purpose is to set an appropriate quantization parameter used for encoding, and to suppress subjective image quality by suppressing distortion due to quantization. It is to improve and improve the encoding efficiency.

前記画質劣化評価指標をｆ（ｘ _ｉ ）とし、前記発生符号量を正規化した値をｘ _ｉ として式（１）におけるａ _ｉ を次の式（２）により算出し、

発生符号量に対する画質劣化評価指標を表す目標符号化効率をＥ _ｐｉｃ として、発生符号量の目標値である目標発生符号量ＴＭｂＢｉｔ _ｉ を次の式（３）により算出し、

当該目標発生符号量に基づいて前記量子化パラメータを算出することを特徴とする動画像符号化装置である。 (1) The present invention has been made to solve the above-described problems, and one aspect of the present invention generates a bitstream by encoding an input video signal using a predetermined quantization parameter. A first encoding unit; a decoding unit that decodes the bitstream generated by the first encoding unit to generate a decoded video signal; and a difference between the decoded video signal decoded by the decoding unit and the input video signal. Based on an evaluation unit for calculating an image quality degradation evaluation index indicating the degree of distortion in the decoded video signal, and on the basis of an image quality degradation evaluation index calculated by the evaluation unit and a generated code amount in the first encoding unit comprising a quantization parameter calculation unit that calculates a parameter, and a second encoding unit that performs encoding on the input video signal by using a quantization parameter, wherein the quantization parameter calculation unit has calculated, Serial evaluation unit calculates the image quality degradation evaluation index using SSIM (Structural SIMILARITY), the quantization parameter calculation unit relates defined rate-distortion curve f _{(x i)} by the following equation (1) ,

The image quality degradation evaluation index is set to f (x _i ), the value obtained by normalizing the generated code amount is set to x _i , and a _i in the formula (1) is calculated by the following formula (2):

The target encoding efficiency TMbBit _i , which is the target value of the generated code quantity, is calculated by the following equation (3) , where E _pic is the target coding efficiency representing the image quality degradation evaluation index for the generated code quantity ,

The moving picture coding apparatus is characterized in that the quantization parameter is calculated based on the target generated code amount .

前記画質劣化評価指標をｆ（ｘ _ｉ ）とし、前記発生符号量を正規化した値をｘ _ｉ として式（４）におけるａ _ｉ を次の式（５）により算出し、

発生符号量に対する画質劣化評価指標を表す目標符号化効率をＥ _ｐｉｃ として、発生符号量の目標値である目標発生符号量ＴＭｂＢｉｔ _ｉ を次の式（６）により算出し、

当該目標発生符号量に基づいて前記量子化パラメータを算出することを特徴とする動画像符号化方法である。 ( 2 ) Further, according to one aspect of the present invention, the first encoding unit encodes an input video signal using a predetermined quantization parameter to generate a bitstream, and the decoding unit Decoding the bitstream generated by the first encoding unit to generate a decoded video signal; and the evaluation unit performing the decoding based on a difference between the decoded video signal decoded by the decoding unit and the input video signal. A step of calculating an image quality degradation evaluation index indicating a degree of distortion in the video signal; and a quantization parameter calculation unit that performs quantization based on the image quality degradation evaluation index calculated by the evaluation unit and the generated code amount in the first encoding unit. A step of calculating a quantization parameter, and a step in which the second encoding unit encodes the input video signal using the quantization parameter calculated by the quantization parameter calculation unit. If, have a, the evaluation unit calculates the image quality degradation evaluation index using SSIM, the quantization parameter calculation unit, the rate-distortion curve f (x _i defined by the following equation (4) )

The image quality degradation evaluation index is set to f (x _i ), the value obtained by normalizing the generated code amount is set to x _i , and a _i in the formula (4) is calculated by the following formula (5):

A target encoding efficiency TMbBit _i , which is a target value of the generated code quantity, is calculated by the following equation (6) , where E _pic is the target coding efficiency representing the image quality degradation evaluation index for the generated code quantity ,

The moving picture coding method is characterized in that the quantization parameter is calculated based on the target generated code amount .

( 3 ) One aspect of the present invention is a program for causing a computer to execute any one of the above-described moving image encoding methods.

  According to the present invention, the quantization parameter used when the video input signal is encoded again is calculated using the image quality degradation evaluation index for the result of encoding the video input signal once. For this reason, an appropriate quantization parameter can be set, subjective image quality can be improved, and coding efficiency can be improved.

It is a block diagram which shows the function structure of the moving image encoder by this embodiment. It is a flowchart which shows the procedure of the encoding process by this embodiment. It is a graph which shows the result of having experimented with the moving image encoder in this embodiment. It is a block diagram which shows the function structure of the moving image encoder in a prior art.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a block diagram showing a functional configuration of a moving picture encoding apparatus 10 according to the present embodiment.
The moving image encoding apparatus 10 includes a first encoding unit 11, a decoding unit 12, an evaluation unit 13, a quantization parameter calculation unit 14, and a second encoding unit 15. The first encoding unit 11 receives a video signal to be encoded (hereinafter referred to as an input video signal), encodes the input video signal, and sets the encoded result as a bit stream. Then, the first encoding unit 11 outputs the bit stream to the decoding unit 12. Specifically, the first encoding unit 11 first performs predictive encoding and block division transform encoding on each macroblock in the input video signal. Next, the first encoding unit 11 divides an orthogonal transform coefficient, which is a result of performing block division transform coding, by a quantization step. Then, the first encoding unit 11 rounds the division result to an integer value and calculates a quantized value. At this time, the first encoding unit 11 calculates a quantization step from the quantization parameter set in advance for each macroblock. In the present embodiment, the logarithm of the quantization parameter and the quantization step is proportional. For this reason, the first encoding unit 11 calculates the quantization step QS according to the following equation (7). Here, QP is a quantization parameter. Hereinafter, the encoding performed by the first encoding unit 11 is referred to as first encoding.

  The decoding unit 12 decodes the input bit stream and uses the decoded result as a decoded video signal. Then, the decoding unit 12 outputs the decoded video signal to the evaluation unit 13, and outputs the encoded information of the bitstream acquired at the time of decoding to the quantization parameter calculation unit 14 and the second encoding unit 15. The encoding information is information when the input video signal is first encoded, and includes a macroblock type and a macroblock generation code amount. The macro block type is a result of predictive coding and block division transform coding of a macro block. The macroblock generated code amount (generated code amount) is a code amount generated when a macroblock is encoded.

The evaluation unit 13 receives the input video signal and the decoded video signal, and based on the difference between the input video signal and the decoded video signal in each macroblock (hereinafter referred to as a differential signal), the image quality degradation evaluation index by the first encoding Is calculated. The image quality degradation evaluation index is an index indicating the degree of distortion. The image quality degradation evaluation index in the present embodiment is an SSIM obtained from the luminance value difference, variance, and covariance indices in each macroblock. SSIM indicates that the larger the value, the smaller the image distortion, and the smaller the value, the larger the distortion. The image quality degradation evaluation index may be a sum of absolute differences or a sum of squares of the difference signals. Then, the evaluation unit 13 outputs the image quality degradation evaluation index to the quantization parameter calculation unit 14.
Here, the evaluation unit 13 calculates SSIM by the following equation (8). x represents an image in the input video signal, and y represents an image in the decoded video signal. Α1, β1, and γ1 are constants.

  The elements constituting the expression (8) are defined by the following expressions (9), (10), and (11), respectively.

Here, l (x, y) is a luminance value comparison function, c (x, y) is a contrast value comparison function, and s (x, y) is a structure comparison function. Further, μ _x used here is defined by the following equation (12). Here, lx _i is the luminance value of the i-th macroblock in the input video signal. N is a positive integer and is the number of macroblocks.

Μ _y is defined by the following equation (13). Here, ly _i is the luminance value of the i-th macroblock in the decoded video signal.

Σ _x is defined by the following equation (14).

Σ _y is defined by the following equation (15).

Σ _xy is defined by the following equation (16).

  The quantization parameter calculation unit 14 is a quantization used when the second encoding unit 15 performs encoding based on the encoding information output from the decoding unit 12 and the image quality degradation evaluation index output from the evaluation unit 13. Parameters are calculated for each macroblock. Here, the quantization parameter calculation unit 14 uses a macroblock generation code amount as the encoding information. Details of the quantization parameter determination method will be described later. Then, the quantization parameter calculation unit 14 outputs a quantization parameter set that is a combination of quantization parameters in each macroblock to the second encoding unit 15.

  The second encoding unit 15 performs encoding using the input video signal, the quantization parameter set output by the quantization parameter calculation unit 14, and the encoding information output by the decoding unit 12. That is, the second encoding unit 15 performs encoding while applying the value specified by the quantization parameter set to the macroblock type included in the encoding information, and outputs the encoded result as an encoded stream. To do. Specifically, the second encoding unit 15 first calculates a quantization step from the quantization parameter specified in the quantization parameter set. Then, the second encoding unit 15 divides the macroblock type by the quantization step, rounds the result of the division to an integer value, and calculates a quantization value. Hereinafter, the encoding in the second encoding unit 15 is referred to as second encoding.

Next, the encoding process by the moving image encoding device 10 will be described with reference to FIG. FIG. 2 is a flowchart showing the procedure of the encoding process according to this embodiment.
First, in step S101, the first encoding unit 11 performs first encoding on the input video signal. Next, in step S102, the decoding unit 12 decodes the bitstream that is the result of the first encoding. In step S103, the evaluation unit 13 calculates an image degradation evaluation index SSIM from the difference between the input video signal and the decoded video signal.

Next, in step S104, the quantization parameter calculation unit 14 calculates the target encoding efficiency (E _pic ). Here, the target coding efficiency E _pic = (∂D / ∂R) is a value representing distortion with respect to the generated code amount. D represents distortion and is a value of an image quality degradation evaluation index obtained by the evaluation unit 13. That is, the value of SSIM is entered in D. Also, R is a macroblock generation code amount included in the encoding information. For example, when using the average value of the macroblock generation code amount (average macroblock generation code amount) and the SSIM average value (average SSIM) in the first encoding result, E _pic = (∂D / ∂R) = Average SSIM / average macroblock generation code amount.

Next, in step S105, the quantization parameter calculation unit 14 obtains a rate distortion curve f (x _i ) in each macroblock. Here, i, which is a subscript of x, is a positive integer and means the i-th macroblock. The rate distortion curve f (x _i ) is defined by the following equation (17).

Here, x _i is a value obtained by normalizing the macroblock generation code amount at the time of the first encoding obtained from the decoding unit 12. Further, f (x _i ) is an image quality degradation evaluation index at the time of the first encoding obtained by the evaluation unit 13. a _i is a variable that uniquely defines f (x _i ). That is, the quantization parameter calculation unit 14 calculates a _i by the following equation (18) in order to obtain the equation (17).

Here, the macroblock generation code amount is defined as 0 to 3200 bits, but in Expression (18), the macroblock generation code amount is divided by 3200 and the value range from 0 to 1 falls into x _i .

Next, in step S106, the quantization parameter calculation unit 14 obtains a target generated code amount TMbBit _i that satisfies the target encoding efficiency E _pic on the rate distortion curve f (x _i ). The target generated code amount TMbBit _i is a target value of the macroblock generated code amount in the second encoding. It is clear from the definition that the target encoding efficiency E _pic = (∂D / ∂R) means a slope on f (x _i ). That is, it can be said that f _{(x i)} was differentiated for _{x i} f _{'(x i)} and as _{E pic} equals _{x i} is the target generated code amount TMbBit _i. Therefore, the quantization parameter calculation unit 14 obtains TMbBit _i using the following equation (19).

Here, since x _i = TMbBit _i , the quantization parameter calculation unit 14 calculates the target generated code amount TMbBit _i by the following equation (20).

Next, in step S107, the quantization parameter calculation unit 14 determines the quantization parameter to the macroblock generation code so that the macroblock generation code amount after the second encoding becomes the TMbBit _i obtained previously. A curve (QP-GenBit curve: g (y _i )) representing the quantity relationship is calculated. The quantization parameter and the macroblock generation code amount have a monotonically decreasing relationship in which the macroblock generation code amount decreases as the quantization parameter increases. In addition, the monotonic decrease is characterized by a gradual decrease, and this is considered to apply to the fourth quadrant of the quadratic function. Therefore, the QP-GenBit curve (g (y _i )) is defined by a quadratic equation as in the following equation (21).

Here, y _i is the quantization parameter at the time of the second coding, alpha and β is a constant determined in advance, gamma _i is g (y _i) uniquely define constant, g (y _i) macroblock generated code Amount. I represents the i-th macroblock. In obtaining the quantization parameter at the time of the second encoding, the quantization parameter calculation unit 14 firstly sets the quantization parameter at the time of the first encoding obtained from the first encoding unit to the following equation (22): Substituting the macro block generation code amount to obtain γ _i , and uniquely determines the QP-GenBit curve (g (y _i )) of each macro block.

The QP-GenBit curve draws various curves depending on the macroblock, but each curve has a characteristic that it hardly intersects. On the other hand, the macroblock generation code amount at the same quantization parameter varies greatly depending on the characteristics of the macroblock. Therefore, the values of α and β that determine the shape of the curve and the position in the X-axis direction are determined in advance. In determining α and β, a plurality of macroblocks are previously coded for quantization parameters 0 to 51 by experiment to obtain macroblock generation code amount and SSIM data. From the encoding result, an average macroblock generation code amount and average SSIM of all macroblocks are taken for each quantization parameter, and an approximate curve is obtained by using the least square method for those points (52 points). The coefficients of y ² _i and y _i of the quadratic curve obtained here are defined as α and β, respectively. In the present embodiment, α = 0.91846975 and β = −95.079875 are determined.

Next, in step S108, the quantization parameter calculating unit 14 uniquely determines the variable γ _i that determines the QP-GenBit curve (g (y _i )) obtained by the equation (22) and the target obtained by the equation (20). Substitute the generated code amount TMbBit _i into the following equation (23) to obtain the quantization parameter y _i at the time of the second encoding.

Here, in order to uniquely find a solution for y _i , a constraint of 0 ≦ y _i ≦ 51 is provided.

In step S109, the quantization parameter calculation unit 14 determines whether the quantization parameter y _i has been calculated for all macroblocks. If the quantization parameter y _i has been calculated for all macroblocks, the process proceeds to step S110. On the other hand, if the quantization parameter y _i has not been calculated for all macroblocks, the process returns to step S105. Then, after obtaining y _i for all the macroblocks, in step S110, the second encoding unit 15 performs second encoding using the quantization parameter y _i .

  FIG. 3 is a graph showing a result of an experiment performed by the moving image encoding apparatus 10 according to the present embodiment. In this figure, the horizontal axis represents the macroblock generated code amount (unit: kbit / s), and the vertical axis represents the image quality degradation evaluation index SSIM. A solid line 101 indicates the result of the second encoding, and a solid line 102 indicates the result of the first encoding. The experiment in this figure was conducted on the standard definition television (SDTV) standard size (720 × 480), interlace, and the number of frames of 450 frames, which are one of standard images in the ITE (Video Information Media Society). SSIM indicates that the larger the value, the smaller the image distortion, and the smaller the value, the larger the distortion. For this reason, as shown in the figure, it can be confirmed that the result of the second encoding is improved in encoding efficiency as compared with the result of the first encoding.

  As described above, according to the present embodiment, the moving image encoding apparatus 10 calculates the quantization parameter from the image degradation evaluation index, and performs the second encoding using the calculated quantization parameter. For this reason, it becomes possible to set an optimal quantization parameter, subjective image quality can be improved, and coding efficiency can be improved.

2 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed, thereby executing an encoding process. You may go. Here, the “computer system” may include an OS and hardware such as peripheral devices.
Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.
The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time.
The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.
The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

  As described above, the embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to the above, and various design changes and the like can be made without departing from the scope of the present invention. It is possible to

  DESCRIPTION OF SYMBOLS 10 ... Moving image encoder 11 ... 1st encoding part 12 ... Decoding part 13 ... Evaluation part 14 ... Quantization parameter calculation part 15 ... 2nd encoding part

Claims (3)

A first encoding unit that encodes an input video signal using a predetermined quantization parameter to generate a bitstream;
A decoding unit that generates a decoded video signal by decoding the bitstream generated by the first encoding unit;
An evaluation unit that calculates an image quality degradation evaluation index indicating a degree of distortion in the decoded video signal based on a difference between the decoded video signal decoded by the decoding unit and the input video signal;
A quantization parameter calculation unit that calculates a quantization parameter based on an image quality degradation evaluation index calculated by the evaluation unit and a generated code amount in the first encoding unit;
A second encoding unit that encodes the input video signal using the quantization parameter calculated by the quantization parameter calculation unit;
Equipped with a,
The evaluation unit calculates the image quality degradation evaluation index using SSIM (Structural SIMality),
The quantization parameter calculation unit relates to the rate distortion curve f (x _i ) defined by the following equation (1) :

The image quality degradation evaluation index is set to f (x _i ), the value obtained by normalizing the generated code amount is set to x _i , and a _i in the formula (1) is calculated by the following formula (2):

The target encoding efficiency TMbBit _i , which is the target value of the generated code quantity, is calculated by the following equation (3) , where E _pic is the target coding efficiency representing the image quality degradation evaluation index for the generated code quantity ,

A moving picture coding apparatus that calculates the quantization parameter based on the target generated code amount . A first encoding unit that encodes an input video signal using a predetermined quantization parameter to generate a bitstream;
A decoding unit decoding the bitstream generated by the first encoding unit to generate a decoded video signal;
An evaluation unit calculating an image quality degradation evaluation index indicating a degree of distortion in the decoded video signal based on a difference between the decoded video signal decoded by the decoding unit and the input video signal;
A step of calculating a quantization parameter based on an image quality degradation evaluation index calculated by the evaluation unit and a generated code amount in the first encoding unit;
A second encoding unit encoding the input video signal using the quantization parameter calculated by the quantization parameter calculation unit;
I have a,
The evaluation unit calculates the image quality degradation evaluation index using SSIM,
The quantization parameter calculation unit relates to the rate distortion curve f (x _i ) defined by the following equation (4) :

The image quality degradation evaluation index is set to f (x _i ), the value obtained by normalizing the generated code amount is set to x _i , and a _i in the formula (4) is calculated by the following formula (5):

A target encoding efficiency TMbBit _i , which is a target value of the generated code quantity, is calculated by the following equation (6) , where E _pic is the target coding efficiency representing the image quality degradation evaluation index for the generated code quantity ,

A moving picture coding method, wherein the quantization parameter is calculated based on the target generated code amount . A program for causing a computer to execute the moving image encoding method according to claim 2 .

Images