KR20100129091A - Method of rate control in frame-level for h.264/avc - Google Patents

Abstract

PURPOSE: A method of rate control in a frame-level for h.264 is provided to control bit rate adaptively in an image. CONSTITUTION: A parameter, including an initial quantization parameter value of an image, is initialized. A frame complexity weight value for the prediction of a target bit of a current frame is calculated. A target bit amount of the current frame is calculated by using a bit amount considering the target buffer level and the virtual buffer occupation amount and the bit amount for encoding each frame using the frame complexity weighted value. The QP(Quantization Parameter) value about the current frame is determined.

Description

Method of rate control in frame for H.264 {Method of Rate Control in Frame-level for H.264 / AVC}

The present invention relates to a frame rate control method for H.264, and more particularly to an adaptive frame rate control method for real time H.264 based on the frame unit.

H.264 is a digital video codec standard that has a very high data compression rate, also called MPEG-4 Part 10 or Advanced Video Coding (AVC). This standard is an international compression standard developed by ITU-T's Video Coding Expert Group (VCEG) and ISO / IEC's Motion Picture Experts Group (MPEG) to provide an improved video coding standard. MPEG-4 Part 10 of .264 and ISO / IEC are technically the same standard. H.264 is a type of H.26x video standard proposed by ITU-T. It is usually called H.264 / AVC or AVC / H.264, H.264 / MPEG-4 AVC, MPEG-4 / H.264 AVC. Call.

Through a variety of predictive and entropy coding methods, H.264 / AVC video coding reduces bit rates by up to 50% compared to existing video coding standards such as H.263 + or MPEG-4 (part2). This can provide improved coding efficiency. In the case of transmitting image data in real time, the amount of bits generated according to the characteristics and types of the image may be changed in time because the characteristics of the transmitted channel are very limited and have a fixed bandwidth. Therefore, in order to provide the best encoding performance by adjusting the balance between the quality of the image and the capacity of the channel, it is essential to adjust the coding parameters.

To control the coding performance, H.264's rate control algorithm is based on the Rate Distortion Optimization (RDO) method to efficiently adjust the ratio of generated bits and quantization distortion for each frame before encoding. The appropriate quantization parameter QP (Quantization Parameter) is determined, and the coefficient of each coding mode is measured by this coefficient to determine the minimum cost mode as the final mode.

Therefore, the bit rate control method in H.264 / AVC has a problem that the optimal QP should be selected before the mode decision and the motion vector decision. In order to solve this problem, a method of estimating the distortion value of the macroblock from the macroblock of the previous frame by means of Mean Absolute Different (MAD) is used. M. Jiang improves this MAD technique by using the predicted MAD rate of the current frame relative to the average MAD value of the coded P frames in the GOP interval, and the P code coded with this MAD rate. The average PSNR error rate of the frames is used to predict the complexity of the current frame and shows improved performance (see "On Enhancing H.264 / AVC Video Rate Control by PSNR-Based Frame Complexity Estimation", IEEE Trans. Consumer. Electronics.vol. 51, no. 1, Feb. 2005.). However, these MAD techniques require a large amount of computation, which places a lot of restrictions on the application of video.

The bit rate control method used in H.264 / AVC is to find the optimal coding parameter by actual coding. The Lagrangian optimization technique is used to predict that the given bit rate condition is minimized and the distortion is minimized among the possible coding parameter combinations. I'm using. The bit rate control method used in H.264 / AVC is based on the Quadratic R-D model where bits and distortions are expressed as a function of the quantization stem. In spite of the characteristics of the video encoding and the direct adjustment method for quantization, this method may show an inefficient bit rate adjustment result due to various characteristics of video encoding. This method is ideal, but it is not suitable for real-time multimedia transmission due to the problem of large amount of computation because the actual bit rate and distortion values for all quantization parameters must be obtained. In real-time transmission of image information, it is desirable to reduce unnecessary computation and minimize image distortion in the process of estimating a bit amount suitable for a target application.

Therefore, in the present invention, in order to reduce the amount of computation in the process of predicting the amount of bits, a technique for calculating the amount of bits appropriate for the target application and minimizing the distortion of the image and encoding the image information in real time using a bit rate control method based on a statistical model We want to implement

When transmitting a bitstream encoded by an H.264 / AVC encoder over a fixed channel, a major problem of the conventional bit rate control method is that a large amount of computation is required because the amount of differential code information of the previous frame is used to encode the current frame. In order to solve these problems, the present invention implements a method of controlling a bit rate adaptively to image characteristics by using statistical information of various images and previously generated bit amounts as statistical information, not the difference code information amount of a previous frame. I would like to. That is, the present invention uses a method of reducing complexity than the conventional method based on a linear model of QP and the actual bit amount generated after encoding, thereby accurately predicting the amount of bits generated in a frame even with a small amount of computation. To reduce the overall computation of the bit rate control algorithm rather than Although H.264 / AVC supports both VBR (variable bit rate coding method) and CBR (fixed bit rate coding method) at the same time, the present invention intends to implement a CBR bit rate control method based on a frame unit.

의 가중치 적용값과 직전 프레임의 QP값과 동일한 QP값을 갖는 직전 프레임 이전의 프레임들의 평균 비트량

의 가중치 적용값의 합과 직전 프레임의 QP값(

)을 이용하여 현재 프레임의 목표 비트량을 예측하기 위한 프레임 복잡도 가중치

를 예측 계산하는 복잡도가중치 갱신단계; 현재 프레임의 목표 비트량을 예측하기 위한 프레임 복잡도 가중치

를 이용한 프레임당 부호화할 수 있는 비트량

과 가상버퍼 점유량

과 목표 버퍼레벨

을 고려한 프레임당 부호화할 수 있는 비트량

을 이용하여 현재 프레임에 대한 목표 비트량

을 계산하는 목표비트량 산출단계; 및 QP값 인덱스에 대한 예측된 발생 비트량

구간에서 현재 프레임에 대한 목표 비트량

을 반복 루프를 통하여 찾아 현재 프레임에 대한 QP값(

)을 결정하는 프레임 QP값 결정단계; 를 포함하여 이루어지고,In order to achieve the above object, according to one aspect of the present invention, in the frame-by-frame rate control method for H.264, the parameters including the initial information and the initial quantization parameter (QP) value of the buffer to initialize the frame information and the buffer A parameter initialization step of initializing; Coded Bit Amount for Previous Frame

Average bit quantity of frames before the previous frame having a weighted value of and the same QP value as the previous frame's QP value

The sum of the weighted values of and the QP value of the previous frame (

Frame complexity weights for predicting the target bit rate of the current frame using

A complexity weighted updating step of predicting and calculating the complexity; Frame Complexity Weights for Predicting Target Bitrate of Current Frame

The amount of bits that can be encoded per frame using

And virtual buffer occupancy

And target buffer levels

Bit rate that can be encoded per frame

Target bit rate for the current frame

Calculating a target bit amount; Amount of occurrence bits for the index and QP value index

Target bit rate for the current frame in the interval

Through the loop to find the QP value for the current frame (

Determining a frame QP value; , ≪ / RTI >

Here, the method for controlling the frame rate of a frame for H.264 characterized in that the encoding is performed with the QP value of the frame to be currently encoded determined by performing the complexity weighting step, the target bit amount calculation step, and the frame QP value determination step. Suggest.

는 다음의 식에 따라 결정되는 것을 특징으로 한다.Preferably, in the complexity weighting step, the frame complexity weighting

Is characterized by the following equation.

Here, lambda is a weight value for reducing the error between the target bit amount and the actually generated bit amount. Preferably λ is 0.67.

은 다음의 식에 따라 결정되며,Preferably, in the target bit amount calculation step, the target bit amount for the current frame

Is determined by the equation

과 목표 버퍼레벨

에 대한 의존도를 나타내는 가중치 값이고, 바람직하게 β는 0.5이고,Β is the virtual buffer occupancy

And target buffer levels

Is a weight value representing dependence on, preferably β is 0.5,

를 이용한 프레임당 부호화할 수 있는 비트량

및 가상버퍼 점유량

과 목표 버퍼레벨

을 고려한 프레임당 부호화할 수 있는 비트량

는 다음의 식에 따라 결정된다.Frame complexity weights

The amount of bits that can be encoded per frame using

And virtual buffer occupancy

And target buffer levels

Bit rate that can be encoded per frame

Is determined by the following equation.

는 가중치 상수이고, 바람직하게는 0.8이다.Where R is the amount of bits currently remaining, N _p is the number of P frames in one picture group (GOP), u is the bit rate for constantly transmitting the video sequence compressed at the channel bandwidth, and F _r is The frame rate encoded per second,

Is a weight constant, preferably 0.8.

과 목표 버퍼레벨

을 산출하는 단계; 산출된 가상버퍼 점유량

과 목표 버퍼레벨

을 이용하여 예측 버퍼량을 결정하는 단계; 및 예측 버퍼량과 프레임당 부호화할 수 있는 비트량

및

를 고려하여 현재 프레임에 대한 목표 비트량

을 결정하는 단계;를 포함하여 이루어진다.Preferably, the step of calculating the target bit amount is: virtual buffer occupancy amount

And target buffer levels

Calculating; Calculated Virtual Buffer Occupancy

And target buffer levels

Determining a predicted buffer amount using; And the amount of prediction buffer and the amount of bits that can be encoded per frame

And

Target bit rate for the current frame

Determining the;

은

로부터 산출되며. 여기에서 A(n-1)는 (n-1)번째 프레임에 대한 부호화된 비트량이고,More preferably, the virtual buffer occupancy amount

silver

Is calculated from. Where A (n-1) is the coded bit amount for the (n-1) th frame,

은 The target buffer level

silver

로부터 산출되며, 여기에서 codedPframe 은 영상그룹(GOP) 구간에서의 부호화된 P 프레임 개수이고, N_p 는 하나의 GOP 내에 있는 P 프레임의 개수이고,

Where codedPframe is the number of encoded P frames in a GOP section, N _p is the number of P frames in one GOP,

The determining of the prediction buffer amount may include:

으로부터 프레임 당 예측 버퍼량

을 산출하고,

Amount of prediction buffer per frame from

Yields,

If the amount of buffer per frame is estimated,

로부터 버퍼의 상한(U(n))과 하한(L(n)) 범위까지 고려하여 최종 예측 버퍼량(

) 값을 결정한다.

From the upper bound U (n) and the lower bound L (n) of the buffer to

) To determine the value.

가 결정될 때 프레임들간의 화질 열화를 줄이기 위하여 다음의 식에 따라 현재 프레임의 최종

가 결정되는 것을 특징으로 한다.Preferably, in the frame QP value determination step, for the current frame

To reduce image quality deterioration between frames when is determined, the final frame of the current frame is

Characterized in that is determined.

는 프레임간의 QP값을 제한하는 값이다. 바람직하게는, 상기

는 ±2로 설정된다.From here

Is a value for limiting the QP value between frames. Preferably, the

Is set to ± 2.

결정은 다음의 식에 따라 이루어진다.More preferably, for the current frame

The decision is made according to the following equation.

는 QP 인덱스에 의한 예측 비트량

구간에서 해당 목표 비트량 T 를 찾는 함수로써, 반복 루프를 통하여 구해진다.From here

Is the predicted bit amount by the QP index

This function is to find the corresponding target bit amount T in the interval and is obtained through an iterative loop.

According to an aspect of the present invention, in the embodiment, when transmitting a bitstream encoded in an H.264 / AVC coder through a fixed channel, the previous frame in encoding the current frame, which is a big problem of the existing bitrate control method, may be obtained. In order to solve the problem that requires a large amount of calculation because the difference code information amount is used, it is adaptive to the image characteristics by using the statistical information of various images and the previously generated bit amount as statistical information, not the difference code information amount of the previous frame. Now you can control the bit rate.

As can be seen from the experimental results of one embodiment of the present invention, the PSNR performance of the conventional bit rate control method is almost similar to that of the conventional bit rate control method. In particular, in terms of computational savings, it is shown that the proposed method is suitable for real-time processing because the computational amount is reduced by more than 99%. This can be used in many applications related to video.

It is apparent that various effects not directly mentioned in accordance with various embodiments of the present invention may be derived by those skilled in the art from various configurations according to embodiments of the present invention.

Embodiments of the present invention for achieving the above object will be described with reference to the accompanying drawings. In describing the embodiments, the same reference numerals refer to the same configuration, and additional descriptions that may overlap or limit the meaning of the invention may be omitted in describing the embodiments of the present invention.

FIG. 1 shows the overall hierarchical structure in the bit rate control method according to an embodiment of the present invention, and is composed of four main processing steps.

1) Initialization of main parameters for encoding

2) Calculation of target bit rate for frame

3) Determination of the QP value for the target bit amount

4) Update complexity for frame

When the initialization is completed in step 1, the process of determining the QP value for the P frame is repeated in steps 4 through 2 and 3. That is, the process structure to be performed is as follows.

Step 1-> ...-> (Step 4-> Step 2-> Step 3)-> ...-> (Step 4-> Step 2-> Step 3) ...

The detailed description of the contents of each step is as follows.

1) Initialization of main parameters for encoding (step 1)

In step 1, parameters such as frame information for encoding, a buffer initialization, and an initial QP value of an image are initialized.

As shown in FIG. 1, in the parameter initialization step, the instantaneous allowable source transmission rate estimation is a step of estimating a bit rate to be transmitted using the speed of a network that can be transmitted and the number of frames that need to be transmitted per second. And / or information such as the size of the image, how many frames per second should be sent and / or frame information for incubation. In addition, buffer initialization initializes the buffer, which sets the initial virtual buffer occupancy of the buffer and the initial values of the upper and lower bounds of the buffer to prevent overflow and underflow of the buffer.

The bit amount allocated to the I frame has a great influence on the image quality of the entire image. Therefore, determining the QP of an I frame affects the image quality of the frames following the I frame, so it is very important to determine the QP of the I frame.

The video sequence is composed of a video group GOP (Group of Picture), the first frame of the GOP is encoded in the Intra mode, and the subsequent frames are generally encoded in the Inter mode. An I-frame refers to an intra coded frame, and a P-frame refers to a frame predicted (forward predicted) with reference to a previous frame. Since the P-frame is encoded using motion prediction correction with reference to the previous frame, the I-frame, which is an intra coded frame, may be referred to as the most important frame as a reference frame for the P-frame. The first connected picture or random access point is typically an Intra code. In this method, when the average QP value of the P-frames in the previous GOP is high, the QP value of the I-frame of the current GOP is also high, and when the QP value of the I-frame of the current GOP is high, the high QP value is low PSNR (Peek). Signal To Noise Rate), the P-frames in the current GOP referring to the I-frame will have a low PSNR in the chain. Therefore, the image quality of the entire image may vary depending on the QP determination method of the I-frame.

The parameter initialization procedure for the I frame in the present invention uses the reference model JM12.1 scheme of H.264 / AVC. In this method, the QP value is determined in consideration of the number of bits per pixel (bpp) that can be allocated to each pixel.

(14)

(14)

In Equation (14), u is a bit rate for constantly transmitting a compressed video sequence with a channel bandwidth, width and height represent a width and height of an image, and F _r is an encoded frame rate per second.

In order to control the bit rate per frame, first, the virtual buffer occupancy for the first frame must be determined. If the first virtual buffer occupancy is B _c (1), then the initial virtual buffer occupancy is Similarly, set it to 1/8 of the buffer size (B _s ).

(2)

(2)

In order to prevent the overflow and underflow of the buffer, initial values of the upper limit U (1) and the lower limit L (1) of the buffer are set as in Equation (15). In this case, u is a bit rate for constantly transmitting a compressed video sequence with a channel bandwidth, and F _r is an encoded frame rate per second.

(15)

(15)

2) Calculate the target bit rate for the frame (step 2)

In step 2, the target bit amount of the frame to be encoded is predicted by using the information of the complexity update of the frame after encoding, taking into account the complexity weight, the virtual buffer occupancy, and the target buffer level of the frame to be encoded. To predict the target bit amount.

The virtual buffer occupancy amount B _c (n) and the target buffer level TBL (n) in the present invention are obtained using equations (16) and (17), respectively.

(16)

(16)

(17)

(17)

을 산출한다.In equation (16), A (n-1) is the coded bit amount for the (n-1) th frame, codedPframe in equation (17) is the number of coded P frames in the GOP interval, and N _p is one The number of P frames in the GOP. As shown in equation (17), the target buffer level decreases the target buffer level in the buffer occupancy in the case of the first P frame after the I frame is encoded, and the value in the previous target buffer level after the first P frame is encoded. . In encoding the current frame, in order to predict a target bit amount that can be allocated to a frame, a prediction of the buffer amount is performed by using the buffer occupancy amount and the target buffer level calculated as in Equation (18).

To calculate.

(18)

(18)

(8)

(8)

) 값을 결정한다.If the buffer amount per frame is predicted, the final predicted buffer amount (

) To determine the value.

The final target bit amount T (n) for the n-th frame is finally determined as in Equation (20) in consideration of the predicted buffer amount and the code bit amount predicted in Equation (19).

(19)

(19)

(20)

20

In Equation (19), W _p is a value predicting the complexity weight of the current P frame. In the present invention, W _p is calculated as the generated bit amount weight and the QP value of previously encoded frames. See Equation (28) below. R is the amount of bits remaining. T _r1 and T _r2 are bits that can be encoded per frame using complexity weights for P frames, and bits that can be encoded per frame considering buffer occupancy and target buffer levels. In Eq. (20), β means the dependence on the buffer occupancy and the target buffer level and was set to 0.5 by experiment.

3) Determination of the QP value for the target bit amount (step 3)

In this step, the QP value of a frame to be encoded is determined.

In the MAD prediction and the quantization coefficient prediction based on the conventional H.264 / AVC method, the MAD (Mean Absolute Difference) indicating the difference between the current frame and the reference frame is first used for the bit rate control. To predict _c , calculate as in Eq. (11).

(11)

(11)

In Equation (11), MAD _p represents the MAD of the previous frame and C ₂ is a compensation value for the prediction error of the predicted values. The initial values for C ₁ and C ₂ are 1 and 0, respectively. The quantization coefficient for the frame unit for controlling the bit rate of the conventional H.264 / AVC uses a quadratic equation rate-distortion model as shown in Equation (12).

(12)

(12)

In Equation (12), T (n) is the target bit quantity in the nth frame, MAD _n is the MAD in the _nth frame, QP is the quantization coefficient, and x ₁ and x ₂ are two-dimensional bit rate-distortions obtained in each frame. Variable representing the value.

In the present invention, instead of using the QP conversion algorithm as shown in Equation (12) of the existing H.264 / AVC method, it is determined by using statistical information by experiments of various images as shown in FIG. FIG. 2 shows the generated bit rate characteristics of a P frame according to the QP value according to the characteristics of each image. Table 1 shows experimental data calculated to predict the bit rate of each frame. It is shown.

The images used in the experiment to use various types of image statistics are fast and detailed images of "Mobile" and "Stefan" and slow and flat images "News" and "Container". The image "Foreman" with even detail is used.

<Table 1> Average bit rate of P frame according to each QP

QP container foreman mobile news stefan QR ... 17 12,849 20,692 47,388 7,136 42,849 24,559 18 10,621 17,164 42,426 6,103 38,352 21,819 19 9,091 14,943 38,998 5,426 35,062 19,301 20 7,212 12,277 34,350 4,645 31,002 16,823 21 5,285 10,522 30,895 4,108 27,926 14,893 22 5,007 9,018 27,715 3,636 24,809 13,095 23 4,006 7,575 24,176 3,153 21,849 11,313 24 3,224 6,358 21,084 2,749 18,947 9,905 25 2,659 5,538 18,972 2,458 17,057 8,593 26 2,093 4,572 16,009 2,100 14,467 7,317 27 1,704 3,921 13,845 1,856 12,594 6,303 ...

It can be seen from the experiment (Table 1) that the average bit amount of P frames has a close relationship with the QP value, and as shown in FIG. 2, the bit amount allocated according to the QP value changes linearly.

The relationship between the QP value and the generated bit amount can be derived as follows.

(21)

(21)

는 QP값 인덱스 n에 대한 예측된 발생 비트량을 의미한다. α와 β는 도 2과 표 1의 다양한 영상의 통계 특성에 의해 결정된 상수 값이다. <표 1>에서의

은 현재 부호화하고자 하는 프레임에 할당 가능한 비트량 구간을 나타낸 것으로서, 양자화 인덱스 함수로서의 비트량이 가우시안 분포를 가질 수 있기 때문에 식 (21)의 QP 인덱스에 대한 예측 가능한 발생 비트량

는 식 (22)와 같이 재계산될 수 있다. N in formula (21) is the QP value index,

Denotes the estimated amount of generated bits for the QP value index n. α and β are constant values determined by statistical characteristics of various images of FIGS. 2 and 1. In <Table 1>

Denotes a bit amount interval that can be allocated to the frame to be currently encoded, and since the bit amount as a quantization index function can have a Gaussian distribution, the predictable generated bit amount for the QP index of Equation (21)

Can be recalculated as Equation (22).

(22)

(22)

은 초기화 단계에서 한 번만 계산되며 1차원 테이블에 값을 저장하여 사용한다. Thus the estimated occurrence bit rate for the QP index

Is calculated only once in the initialization phase and stored in a one-dimensional table for use.

After that, the QP _c value for the current frame is determined using the target bit amount calculated in Equation (20) as shown in Equation (23).

(23)

(23)

는 QP 인덱스에 의한 예측 비트량

구간에서 해당 목표 비트량 T 를 찾는 함수로써, 반복 루프를 통하여 구해진다. 그리고 P 프레임의 QP_c값이 결정될 때는 프레임들 간의 화질 열화를 줄이기 위하여 식 (24)와 같이 ΔQP 값으로 제한한다.In equation (23)

Is the predicted bit amount by the QP index

This function is to find the corresponding target bit amount T in the interval and is obtained through an iterative loop. When the QP _c value of the P frame is determined, the QP _c value is limited to the ΔQP value as shown in Equation (24) to reduce image quality deterioration between frames.

(24)

(24)

In Eq. (24), QP _p represents the QP value in the previous frame, and the ΔQP value that limits the QP value between frames is set to ± 2. If the QP value difference between the frames exceeds ± 2, visual heterogeneity may be visually shown.

4) Update Complexity for Frame (Step 4)

In step 4, the complexity of the current frame is calculated using the actual bit rate generated after the previous frame is encoded. The process of preventing the occupancy of the virtual buffer and the overflow and underflow of the buffer, and the A complexity weight of a P frame to be used in later encoding is calculated using the generated bit amount and the generated bit amount of the current frame.

The upper limit U (n) and lower limit L (n) values of the buffer for preventing the occupancy amount B _c of the virtual buffer and the overflow and underflow of the buffer are obtained through the following equations (25) to (27).

(25)

(25)

(26)

(26)

(27)

(27)

Then the complexity of the P-frame to predict the target bit quantity of the P frame to be encoded in a weight W _p is obtained as shown in equation (28).

(28)

(28)

In one embodiment of the present invention, the maximum number of encoded P frames for predicting the complexity weight for the P frame is limited to 20. In Equation (28), S _bits is an average bit amount of P frames having the same QP value as QP _p in the generation bit amounts A (n-2) to A (n-20) of the previous P frame. This is because it is necessary to know the frame bit amount generated according to each QP value in order to determine the QP value through the bit amount that can be allocated to the frame. In addition, the P frame to be encoded has a close correlation with adjacent frames. Therefore, in one preferred embodiment, the weight lambda value of A (n-1) and S _bits is set to 0.67 to reduce the error between the target bit amount and the actually generated bit amount. This lambda value is a result obtained by simulating each fixed QP value and a reference frame for motion estimation to 20.

[Experiment result]

In the present invention, an efficient bit rate control method based on an H.264 / AVC coder is proposed. H.264 / AVC reference software 12.1 encoder was used to verify the performance of the embodiment of the present invention. The images used in the experiments were QCIF-level CONTAINER, NEWS, FOREMAN, COASTGUARD, PARIS, SIGN IRENE, and TABLE images, which are widely used as performance evaluation tests in the encoder. The GOP structure of the image consists of the initial I frame and all subsequent frames consisting of P frames and no B frames. All the images were tested by applying the QP value frame by frame without skipping frames. JM12.1, an existing bit rate control algorithm, and a reference paper (M. Jiang and N.Ling, "On Enhancing H.264 / AVC Video Rate Control by PSNR-Based Frame Complexity Estimation", IEEE Trans. Consumer.Electronics.vol. 51, no. 1, Feb. 2005.) And the bit rate control method according to an embodiment of the present invention used the Peak Signal to Noise Ratio (PSNR) for the objective evaluation of the image quality, 8bits of M × N size The noise ratio of the decoded image to the original image is defined as follows.

(29)

(29)

는 유클리드 놈(Euclidean norm)을 나타내고,

는 원 영상,

은 복호화 된 영상을 나타낸다. 그리고 각 알고리즘의 계산량 비교를 위하여 고밀도 측정 타이머를 사용하였으며, 계산량 측정 범위는 각 알고리즘의 데 이터 및 버퍼의 초기화 부분을 제외한 비트율을 제어하는 부분에서만 측정하였다. <표 2>, <표 3> 및 <표 4>는 채널 전송대역폭이 각각 32, 64, 128Kbps이고 프레임율이 30fps일 때 기존 비트율 제어 방법과 본 발명의 하나의 실시예에 따른 비트율 제어 방법에 대한 부호화된 비트량, PSNR 및 계산량 비교이다.In equation (29)

Represents Euclidean norm,

Circle image,

Represents the decoded image. High-density measurement timers were used to compare the computational quantities of each algorithm, and the computational measurement range was measured only in the part that controls the bit rate except for the data and buffer initialization of each algorithm. <Table 2>, <Table 3> and <Table 4> show the existing bit rate control method and the bit rate control method according to an embodiment of the present invention when the channel transmission bandwidth is 32, 64, 128 Kbps and the frame rate is 30fps, respectively. The coded bit rate, the PSNR and the computed amount are compared.

<Table 2> Coded Bit Rate, PSNR and Computation Rate of Each Image in 32 kbps Channel Transmission Environment

Sequence
(video) JM 12.1 Reference paper Example Bit
rate PSNR Calculation
(㎲) Mad% Bit
rate PSNR Calculation
(㎲) Mad% Bit
rate PSNR Calculation
(㎲) container 31.99 34.74 1,148.783 14% 32.02 34.74 1,318.194 12% 31.67 34.54 1.277 news 32.09 31.94 1,137.806 14% 32.04 31.80 1,313.095 12% 32.01 31.68  1.289 foreman 32.08 28.39 1,138.822 13% 32.00 28.58 1,310.744 12% 31.95 28.52  1.306 coastguard 32.05 27.46 1,135.405 13% 32.06 27.44 1,309.780 12% 32.03 27.48  1.294 paris 32.18 26.57 1,157.744 14% 32.10 26.53 1,328.955 12% 32.01 26.56  1.283 signirene 32.01 31.23 1,120.029 13% 32.04 31.18 1,294.267 12% 32.00 31.15  1.314 table 32.03 29.27 1,145.141 14% 32.04 29.51 1,319.145 12% 31.94 29.21  1.311

<Table 3> Coded bit rate, PSNR, and computation amount of each image in 64 kbps channel transmission environment

Sequence
(video) JM 12.1 Reference paper Example Bit
rate PSNR Calculation
(㎲) Mad% Bit
rate PSNR Calculation
(㎲) Mad% Bit
rate PSNR Calculation
(㎲) container 64.03 37.42 1,175.506 14% 64.04 37.42 1,346.126 12% 63.94 37.27 1.317 news 64.04 35.82 1,168.595 14% 63.97 35.87 1,342.050 12% 63.89 35.51 1.304 foreman 64.03 32.15 1,183.141 14% 63.99 32.17 1.353.496 12% 63.79 32.10 1.302 coastguard 64.02 29.76 1,173.391 14% 64.00 29.76 1,346.029 12% 64.01 29.75 1.312 paris 64.34 29.05 1,184.514 14% 64.12 29.05 1,356.702 12% 64.06 29.03 1.277 signirene 64.04 33.85 1,148.641 14% 63.97 33.93 1,323.441 12% 64.03 33.74 1.311 table 64.03 33.11 1,185.203 14% 64.02 33.16 1,357.471 12% 63.94 33.33 1.314

<Table 4> Coded bit rate, PSNR and amount of each video in 128 kbps channel transmission environment

Sequence
(video) JM 12.1 Reference paper Example Bit
rate PSNR Calculation
(㎲) Mad% Bit
rate PSNR Calculation
(㎲) Mad% Bit
rate PSNR Calculation
(㎲) container 128.04 40.24 1,201.869 14% 127.95 40.16 1,372.095 13% 127.20 40.19 1.305 news 128.09 40.15 1,193.926 14% 128.07 40.17 1,368.743 12% 127.66 40.08 1.314 foreman 128.01 35.75 1,225.928 14% 128.02 35.78 1,399.074 12% 127.83 35.64 1.313 coastguard 128.14 32.17 1,212.770 14% 127.96 32.21 1,387.609 12% 128.06 31.97 1.314 paris 128.21 33.65 1,204.631 14% 128.01 33.69 1,374.881 12% 128.13 34.00 1.296 signirene 128.15 37.65 1,184.677 14% 127.91 37.64 1,358.640 12% 128.06 37.55 1.272 table 128.00 36.56 1,215.251 14% 128.04 36.63 1,386.727 12% 127.71 36.65 1.307

As can be seen from the experimental results table, the reference paper saves 0.07Kbps less bit on average than JM 12.1, but shows better performance by 0.02 ~ 0.03dB higher on the PSNR side. However, the method of reference paper controls bit rate based on PSNR. Therefore, it is shown that about 13% more calculation amount is needed than the method of JM 12.1 due to the increase of computation. And the method of JM 12.1 and the reference paper is over 0.1Kbps bit on average, which is unsuitable for channel transmission environment.

On the other hand, the method according to the embodiment of the present invention is about 0.02 dB lower than the JM 12.1 in terms of PSNR, and about 0.04 dB lower than the reference paper, but the amount of bits generated in the channel transmission environment is about 0.5 Kbps than the JM 12.1 method. It saved about 0.12Kbps than the reference paper method.

In addition, it can be seen that the method according to the embodiment of the present invention adjusts the bit amount appropriately for the overall channel transmission environment. In terms of the calculation amount, both the JM 12.1 method and the reference paper method show that the amount of calculation for complex MAD calculation and the bit amount adjustment accordingly is much higher than the method according to the embodiment of the present invention.

In particular, the experimental results show that the MAD calculation (MAD%) used to predict the complexity of the current frame occupies about 12% to 14% of the total bit rate control calculation.

As can be seen from the experimental results, the method according to the embodiment of the present invention was almost similar to the conventional JM 12.1 method and the reference paper method in terms of PSNR, but the average bit rate was controlled with a very small amount of 0.1%. It was confirmed that it can be done.

In the method according to the embodiment of the present invention, a large amount of computational savings does not follow complicated MAD calculations and calculation processes, unlike conventional methods, and utilizes statistical information of existing images and regenerates the generated bit amount in real time. This is because the bit rate control is adaptively applied to the characteristics of the image by utilizing the.

3 and 4 illustrate a comparison between the bit rate and PSNR coded for each frame in the method and the existing methods according to an embodiment of the present invention when a TABLE image is transmitted through a 128 Kbps channel. JM 12.1 is based on the existing JM 12.1 model, REF 14 is a method according to the reference papers of M. Jiang and N. Ling, and Proposed is according to one embodiment of the present invention. The method according to the embodiment of the present invention shows a higher PSNR performance while generating a more stable amount of generated bits than the existing methods.

5, 7, and 8 are PSNR performance comparisons between the method according to the embodiment of the present invention and the existing methods when the CONTAINER, NEWS, and TABLE images are transmitted through various channels, respectively. JM 12.1 is based on the existing JM 12.1 model, REF 14 is a method according to the reference papers of M. Jiang and N. Ling, and Proposed is according to one embodiment of the present invention. It can be seen that for most transport channels, the PSNR performance is similar to that of the conventional method.

In the above, the present invention has been described with reference to the preferred embodiments with reference to the accompanying drawings. The accompanying drawings and the foregoing embodiments are described by way of example to help those skilled in the art to understand the present invention. Therefore, various embodiments of the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention, the foregoing embodiments are to be considered as illustrative and not restrictive. Therefore, the scope of the present invention should be interpreted according to the invention described in the appended claims rather than the above-described embodiments, and various modifications, alternatives, and equivalents described by those skilled in the art are described above. It is obvious that it is included in the scope of the.

1 is a diagram illustrating a hierarchical structure of a bit rate control method according to an embodiment of the present invention;

2 is a graph illustrating a state in which the amount of generated bits varies according to characteristics of each image in one aspect of the present invention.

3 is a graph illustrating a comparison of encoded bit amounts (128 Kbps) per frame of one embodiment of the present invention and existing methods for a table image;

4 is a graph illustrating PSNR comparison (128 Kbps) per frame of one embodiment of the present invention and existing methods for a table image.

5 is a graph showing an average PSNR Y (dB) (@ 30 Hz) of one embodiment of the present invention and existing methods for a container image.

FIG. 6 is a graph showing an average PSNR Y (dB) (@ 30 Hz) of one embodiment of the present invention and existing methods for a News image.

7 is a graph showing an average PSNR Y (dB) (@ 30 Hz) of one embodiment of the present invention and a conventional method for a table image.

Claims (7)

In the H.264 bit rate control method, A parameter initialization step of initializing the frame information and the buffer and initializing parameters including an initial quantization parameter (QP) value of the image; Coded Bit Amount for Previous Frame

Average bit quantity of frames before the previous frame having a weighted value of and the same QP value as the previous frame's QP value

The sum of the weighted values of and the QP value of the previous frame (

Frame complexity weights for predicting the target bit rate of the current frame using

A complexity weighted updating step of predicting and calculating the complexity; Frame Complexity Weights for Predicting Target Bitrate of Current Frame

The amount of bits that can be encoded per frame using

And virtual buffer occupancy

And target buffer levels

Bit rate that can be encoded per frame

Target bit rate for the current frame

Calculating a target bit amount; And Estimated amount of generated bits for the QP value index

Target bit rate for the current frame in the interval

Through the loop to find the QP value for the current frame (

Determining a frame QP value; , &Lt; / RTI &gt; Here, the method for controlling the frame rate of a frame for H.264 characterized in that the encoding is performed with the QP value of the frame to be currently encoded determined by performing the complexity weighting step, the target bit amount calculation step, and the frame QP value determination step. . The method of claim 1, wherein in the complexity weighting step, The frame complexity weight

The frame rate control method for H.264 characterized in that determined by the following equation.

Here, λ is a weight value for reducing the error between the target bit amount and the actually generated bit amount. The method according to claim 1, wherein in the target bit amount calculation step, Target bit rate for current frame

Is determined by the equation

Β is the virtual buffer occupancy

And target buffer levels

Is a weight value that indicates the dependence on Frame complexity weights

The amount of bits that can be encoded per frame using

And virtual buffer occupancy

And target buffer levels

Bit rate that can be encoded per frame

The frame rate control method for H.264 characterized in that determined by the following equation.

Where R is the amount of bits currently remaining, N _p is the number of P frames in one picture group (GOP), u is the bit rate for constantly transmitting the video sequence compressed at the channel bandwidth, and F _r is The frame rate encoded per second,

Is a weight constant. The method according to claim 1, wherein the target bit amount calculating step: Virtual buffer occupancy

And target buffer levels

Calculating; Calculated Virtual Buffer Occupancy

And target buffer levels

Determining a predicted buffer amount using; And The amount of prediction buffer and the amount of bits that can be encoded per frame

And

Target bit rate for the current frame

Determining the; frame-rate bit rate control method for H.264 characterized in that it comprises a. The method according to claim 4, Virtual buffer occupancy

silver

Is calculated from. Where A (n-1) is the coded bit amount for the (n-1) th frame, Target buffer level

silver

Where codedPframe is the number of encoded P frames in a GOP section, N _p is the number of P frames in one GOP, The determining of the prediction buffer amount may include:

Amount of prediction buffer per frame from

Yields, If the amount of buffer per frame is estimated,

From the upper bound U (n) and the lower bound L (n) of the buffer to

The frame rate control method for H.264 characterized in that it determines the value. The method according to claim 1, wherein in the frame QP value determination step, For the current frame

To reduce image quality deterioration between frames when is determined, the final frame of the current frame is

The frame rate control method for H.264, characterized in that is determined.

From here

Is a value that limits the QP value between frames. The method according to claim 6, For the current frame

The frame rate control method for frame rate for H.264 characterized in that the determination is made according to the following equation.

From here

Is the predicted bit amount by the QP index

This function finds the corresponding target bit quantity T in the interval, and is obtained through an iterative loop.

Images