KR20110099554A - Qp deciding method of h.264/avc encoder using complexity of frame - Google Patents

Abstract

A method for determining a quantization parameter of an H.264 / AVC encoder using the complexity of the video, the method comprising: inputting a video image, calculating a screen complexity for the video image, actual generation bits and GOP allocation bits of the video image Calculating the target bit using the ratio of the ratio, calculating the initial quantization parameter using the complexity of the screen and the target bit, determining whether the scene change has been performed, and calculating the target bit when the scene change has been determined. And determining the scene change quantization parameter using the target bit and the complexity in the image. As a result, by determining the initial quantization parameter and the scene change quantization parameter separately, it is possible to prevent the bit rate control efficiency from being degraded during the initial video input and the sudden image quality conversion such as the scene change, thereby improving the image quality of the entire image.

Description

Quantization Parameter Determination Method of H.264 / AVC Encoder Using Image Complexity {JP Deciding Method of H.264 / AVC Encoder using Complexity of Frame}

The present invention relates to a method for determining a quantization parameter of an H.264 / AVC coder using image complexity. The present invention provides a method for adaptively selecting an initial quantization parameter and a quantization parameter during scene change, thereby improving efficiency of bit rate control. The present invention relates to a method for determining a quantization parameter of an H.264 / AVC encoder using image complexity.

Recently, as hardware is advanced and consumers want high-definition service, ITU-T and MPEG have jointly defined a video coding standard with high-performance compression efficiency called H.264 / AVC. The new technology of the H.264 / AVC video standard has a similar compression structure as previously defined H.263, MPEG-4, etc., but maximizes the compression efficiency. H.264 / AVC obtains the quantization coefficient by DCT transforming and quantizing the difference between the prediction pixel determined as the optimal mode and the current image through intra prediction, and the obtained quantization coefficient is entropy. It is transmitted to the decoder after the encoding process.

By using this compression technology, the real-time video transmission efficiency is increased, and the higher coding efficiency than the conventional one enables higher quality video transmission. Of course, if the channel band is large enough to transmit the original uncompressed image information, such a technique may not be very efficient, but at present, it is impossible to transmit sufficiently large information at once due to the limitation of the transmission channel. Therefore, by compressing and transmitting information using a compression technique having high coding efficiency, it is possible to receive better quality information in real time.

In this case, although the capacity of a normal channel is capable of transmitting information of a certain size, the image information has a different amount of information for each screen. Accordingly, it is necessary to transmit each screen information in accordance with a constant channel capacity. This is why a bit rate control algorithm is needed. Bit rate control basically aims to have an optimal encoding performance in consideration of the amount of bits allowed by the channel capacity and the amount of bits and image quality of the video. The most efficient way to control the image quality and compression rate is to use quantization parameters. The overall flow of bit rate control is to adjust this to improve the image quality while making the bit amount of the image close to the target bit.

Bit rate control is largely performed through the following process. First, the target bit is obtained, then the quantization parameter is determined, and a mean of absolute differential (MAD) value is calculated and encoded through an RDO process. In this case, the target bit is determined in consideration of the buffer state and the channel capacity, and the quantization parameter is calculated using the predicted MAD value and the target bit.

The bit rate control algorithm of H.264 / AVC has some problems. First, there is no bit rate control algorithm for the I-frame and the first P-frame. This causes a reduction in overall bit rate control efficiency if the initial quantization parameter is not properly selected. Therefore, the adaptive quantization parameter selection algorithm for I-frame improves the rate control efficiency.

The following is the problem due to the linear model used for MAD prediction. In general, the MAD predicted by the linear model has a relatively similar value to the actual MAD, and there is a very large error when a sudden change in image quality such as a scene change occurs. Since the error of the predictive MAD also affects the quantization parameter determination for the bit rate control, it is necessary to introduce a new algorithm for scene transition.

An object of the present invention relates to a method of determining a quantization parameter of an H.264 / AVC coder using the complexity of an image to improve bit rate control efficiency by determining a first frame and a separate quantization parameter when changing a scene. .

The objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

The object is the step of inputting a video image; Calculating the complexity of the screen; Calculating a target bit using a ratio of the actual generation bit to the GOP allocation bit; And calculating an initial quantization parameter using the complexity of the screen and the target bit. The method of claim 1 is achieved by the method of determining a quantization parameter of an H.264 / AVC encoder.

According to the method of determining the quantization parameter of the H.264 / AVC encoder using the complexity of the image, the initial quantization parameter and the scene change quantization parameter are determined separately, thereby improving the bit rate control efficiency during the initial video input and the sudden image quality conversion such as the scene change. The fall can be prevented. In addition, the first quantization parameter is encoded as an image that is close to the average image quality of the image. At this time, it is confirmed that the average image quality is most efficient. In addition, when encoding with the scene change quantization parameter, the image is encoded to be close to the average quality of the images after the scene change. Accordingly, the problem of early target bit depletion, which is a problem of the existing bit rate control scheme, is solved, thereby improving the quality of the entire image.

1 is a block diagram of an H.264 / AVC encoder according to the present invention,
2 is a schematic diagram showing a principle in which complexity is obtained in a macroblock;
3 is a flowchart illustrating a process of determining a quantization parameter in an H.264 / AVC encoder according to the present invention.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention.

1 is a block diagram of an H.264 / AVC encoder according to the present invention.

The H.264 / AVC coder includes a bit rate controller 10, a scene change detector 20, a quantizer 30, a DCT converter 40, an entropy coder 50, an inverse quantizer 60, and an IDCT converter 40. ), An image memory 80, and a motion predictor 90.

In such an encoder, a frame image input by a frame or a parallel scan that is input through a video camera or a camcorder is input. Field video supports perforation scanning, which is widely used as the video format of TV signals. A parallel scan is a scheme in which pixel lines are divided into even lines and odd lines, that is, divided into two field signals.

One frame is composed of the "parent field" (the spatially located one of the two fields constituting one frame) and the "subfield" (the spatially located one of the two fields constituting one frame). It consists of two fields.

The bit rate controller 10 controls the bit rate of the transmitted frame, and allocates a large number of bits to a part where a defect of a reconstruction screen caused by an encoding operation is sensitive to the human eye according to the type and characteristics of the screen, and the human eye In the part that is not easily recognized, it allocates relatively few bits to adjust the overall image quality.

The bit rate controller 10 according to the present invention adaptively determines an initial quantization parameter required at the beginning of video input and a scene change quantization parameter required at the scene changeover, and provides the quantizer 30 to the quantizer 30.

The process of determining the initial quantization parameter in the bit rate controller 10 will now be described in detail.

First, in order to determine the initial quantization parameter, the bit rate controller 10 calculates a complexity degree of the screen with respect to the input image, which is calculated by using a complexity calculation method in the screen shown in Equation 1 below. .

Here, I _{x , y, z} are pixel values of the (x, y) position in the k-th frame, G _f (k) is the frame complexity of the k-th frame, M, N is the horizontal and vertical size of the frame.

Equation 1 shows, as shown in FIG. 2, a relationship between pixels in a plurality of macroblocks included in one frame in complexity.

The bit rate controller 10 then sets a target bit for initial quantization parameter determination. The target bit is determined using the ratio of the actual generation bit generated by encoding the selective quantization parameter to the GOP allocation bit, as shown in Equation 2.

는 비트율, N_GOP는 GOP전체의 프레임수, N_totalP는 GOP전체에서 P프레임의 총 수를 의미하며, Fr은 프레임율을 의미한다.here,

Is the bit rate, N _GOP is the number of frames of the entire GOP, N _totalP is the total number of P frames in the GOP, and Fr is the frame rate.

When the target bit is determined in this way, an initial quantization parameter may be calculated using Equation 3 below, and the calculated initial quantization parameter is transferred to the quantizer 30.

Here, QP (k) means the quantization parameter value in the k-th frame. The quantization parameter has a value between '1' and '51'.

Meanwhile, the process of calculating the scene change quantization parameter in the bit rate controller 10 is as follows.

First, in order to calculate the scene change quantization parameter in the bit rate controller 10, it is necessary to determine whether the scene change has been made, which is performed by the scene change detector 20. The scene change detector 20 obtains a complexity ratio and a difference value in the image to determine whether a scene change has been performed. In this case, the complexity ratio uses a ratio of the previous image to the current image and may be represented by Equation 4.

Here, G _f (k) represents the complexity in the k th frame, and G _f (k-1) represents the complexity in the k-1 th frame.

The difference value is obtained by differentially calculating the thermal complexity of the previous image and the thermal complexity of the current image, which can be represented by Equation 5 below.

Here, GMB, R (j, k) is the sum of columns in the macroblock complexity G _MB (i, j, k), which represents the column complexity, and M and N are the horizontal and vertical size of the frame. do.

The scene change detector 20 has a threshold set for each of the complexity ratio and the difference value, and determines that the scene change has been made when the complexity ratio and the difference value satisfy the following Equation 6, respectively.

Here, G _{F , ratio} (k) denotes a complexity ratio between the k th frame and the k-1 th frame, and G _{F , diff} (k) denotes a difference in complexity between the k th frame and the k-1 th frame.

When the scene change detector 20 determines that the scene change has been made, the bit rate controller 10 sets a target bit for determining the scene change quantization parameter. The target bit for determining the scene change quantization parameter may be calculated by using the average of the generation bits during the scene change and the average of the generation bits of the interframe in consideration of the available bits.

Here, Fr denotes a frame rate, N _codeP denotes the number of encoded P frames, N _GOP denotes the number of frames of the entire GOP, and b (n _i ) denotes the actual occurrence bit of the i-th frame.

The target bit calculated as described above and the complexity in the image are applied to Equation 3 to determine the scene change quantization parameter. The determined scene change quantization parameter is provided to the quantizer 30.

The DCT (Discrete Cosine Transform) transformer converts the video information of the spatial domain into frequency domain data and performs Discrete Cosine Transform (DCT) to generate DCT coefficient blocks in macroblock units.

The quantizer 30 quantizes the DCT coefficient block output from the DCT converter 40. In this case, the quantizer 30 may apply quantization parameters such as an initial quantization parameter and a scene change quantization parameter provided from the bit rate controller 10. The quantized DCT coefficient block is provided for each frame.

The entropy encoder 50 allocates a bit shorter than the reference value per code for a code with a high frequency of occurrence, and assigns a bit longer than the reference value with a code with a frequency of occurrence of a code with a low frequency of occurrence. It acts to be close to the entropy of information volume. The entropy coding scheme may use an encoding scheme such as an exponential Golomb code, Variable Length Coding (VLC), Context Adaptive Variable Length Coding (CAVLC), or Context Adaptive Binary Arithmetic Coding (CABAC).

Inverse quantizer 60 performs inverse quantization on quantized spectral coefficients when the current frame is needed for subsequent motion estimation / compensation.

The DCT inverse converter 70 reversely operates the DCT converter 40 to generate an inverse difference macroblock through inverse DCT transformation from the output of the inverse quantizer 60.

The inverse quantizer 60 and the DCT inverse transformer 70 remove overhead information (various header information, start codes, etc.) from the video bitstream, decode pure data information, and then perform inverse quantization and inverse DCT processes. Restore the original pixel values.

If the current frame is an inter frame (e.g., a P frame), the motion predictor 90 determines that the macroblock in the current frame is relative to the reference frame that is the reconstructed frame of the previous frame buffered in the image memory 80. Estimate and compensate for motion The difference between the macroblock (prediction macroblock) of the current frame and the macroblock of the original current frame compensated in the motion predictor 90 is provided to the DCT converter 40.

The image memory 80 temporarily stores and buffers output data in order to uniformly transmit the transmission channel having a constant transmission rate.

The process of determining the quantization parameters for the initial image input and the scene change in the encoder having such a configuration will be described below with reference to FIG. 3.

First, when a video image is input (S300), the bit rate controller 10 calculates a target bit for calculating an initial quantization parameter. The bit rate controller 10 calculates a complexity of the screen with respect to the input video image (S305), and calculates a target bit using the ratio of the actual generation bit and the GOP allocation bit of the input video image (S310). Then, the initial quantization parameter is calculated using the calculated screen complexity and the target bit (S315), and the calculated initial quantization parameter is transferred to the quantizer 30 (S320).

Meanwhile, while the video image is input, the bit rate controller 10 continuously calculates a target bit in real time and continuously controls the bit rate of the transmitted frame. In this case, the bit rate controller 10 calculates and provides a target bit so that the scene quantization parameter can be determined by the quantizer 30 during the scene change. First, the bit rate controller 10 is provided with information on whether a scene change has been made by the scene change detector 20. The scene change detector 20 calculates a complexity ratio and a difference value (S325), and determines whether the two values satisfy a condition for a preset threshold (S330). If both values satisfy the condition, the scene change detector 20 determines that the scene has been changed and provides it to the bit rate controller 10. The bit rate controller 10 calculates the target bit using the average of the generation bits in the scene change and the average of the generation bits of the interframe (340). Then, the bit rate controller 10 calculates the scene change quantization parameter using the complexity and the target bit in the image (S350), and transfers the calculated scene change quantization parameter to the quantizer 30 (S360).

If the complexity ratio and the difference value do not satisfy the condition and the scene change detector 20 determines that the scene change has not been performed, the bit rate controller 10 uses the available bits and the buffer allocation bits as in the prior art. The bit is calculated (S335).

To this end, first, the bit rate controller 10 obtains an allocation bit of a group of picture (GOP) using Equation (8), and then obtains an available bit using Equation (9).

는 비트율을 의미하고, F_r는 프레임율을 나타낸다. N_GOP와 N_{p ,r}은 각각 전체의 프레임수와 부호화되지 않은 P-프레임수를 의미하며, b(n_{i -1})은 i-1번째 프레임의 실제 발생 비트를 의미하며, N_{p ,r}은 부호화되지 않은 P프레임수를 의미한다.here

Denotes a bit rate, and F _r denotes a frame rate. N _GOP and N _{p , r} represent the total number of frames and the number of unencoded P-frames, respectively, and b (n _{i -1} ) represents the actual occurrence bits of the i-1th frame, N _{p , r} Denotes the number of unencoded P frames.

Then, the bit rate controller 10 calculates a buffer allocation bit in consideration of the buffer state by using Equation (10). The target bit is calculated using Equation 11 through the weighted sum of the available bits and the buffer allocation bits.

는 목표 버퍼 레벨, B_c(n_i)는 실제 버퍼 점유율을 의미한다.here,

Is the target buffer level, B _c (n _i ) means the actual buffer occupancy.

β means a weight value and usually has a value of 0.5.

Then, the bit rate controller 10 calculates a predicted mean of absolute differential (MAD) (S345). Since the MAD is obtained through a ROD (Rate Distortion Optimization) process after the quantization parameter is determined, the MAD of the current image must be predicted using the MAD of the previous image. A linear algorithm is used as an MAD prediction algorithm for this purpose, and Equation 12 is an equation regarding a linear model for calculating a predicted MAD.

Here, a ₁ and a ₂ are linear model parameters, and are updated as each macroblock is encoded. The initial values are 1 and 0, respectively.

The quantization parameter for bit rate control is determined using the calculated target bit and the predicted MAD (S355). Equation 13 below is an equation calculated through a mathematical algorithm relating to the degree of distortion of bits and images.

Here, R denotes a target bit required for encoding, and X ₁ and X ₂ denote primary and secondary coefficient values of the two-dimensional bit rate-distortion model, respectively.

Then, the calculated quantization parameter is provided to the quantizer 30 (S360), and a value of Mean of Absolute Differential (MAD) is calculated and encoded through a rate distortion optimization (ROD) process.

Although the preferred embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (8)

Inputting a video image;
Calculating a complexity of a screen for the video image;
Calculating a target bit using a ratio of an actual generation bit and a GOP allocation bit of the video image; And,
Calculating an initial quantization parameter using the complexity of the screen and a target bit; and a method of determining a quantization parameter of an H.264 / AVC encoder. The method of claim 1,
The complexity of the screen is calculated using the following Equation 1 quantization parameter determination method of the H.264 / AVC encoder.
(Equation 1)

Where I _{x, y, z} is the pixel value at position (x, y) in the kth frame, G _f (k) is the frame complexity of the kth frame, and M, N are the horizontal and vertical dimensions of the frame. The method of claim 1,
The target bit is determined by using a ratio between actual generation bits generated by encoding a selective quantization parameter and GOP allocation bits, and is calculated using Equation 2 below. Quantization Parameter Determination Method.
(2)

here,

Is the bit rate, N _GOP is the total number of frames in the GOP, N _totalP is the total number of P frames in the GOP, and Fr is the frame rate. The method of claim 1,
The initial quantization parameter is a method of determining a quantization parameter of the H.264 / AVC coder, characterized in that using the following equation (3).
[Equation 3]

Here, QP (k) means the quantization parameter value in the k-th frame. The method of claim 4, wherein
Determining whether a scene change has been made;
If it is determined that the scene change has been made, calculating a target bit; And,
And determining the scene change quantization parameter by applying the target bit and the complexity in the image to Equation 3. The method of claim 1, further comprising: determining a scene transition quantization parameter. The method of claim 5, wherein
The scene change is determined using a complexity ratio in the image and a difference value, and the complexity ratio is calculated using a ratio of a previous image and a current image, as shown in Equation 4 below, and the difference value is A method of determining a quantization parameter of an H.264 / AVC encoder, characterized by calculating the difference between the thermal complexity of the previous image and the thermal complexity of the current image, as shown in Equation 5 below.
(4)

(5)

Here, G _f (k) represents the complexity in the k-th frame, G _f (k-1) represents the complexity in the k-1th frame, and GMB, R (j, k) represents the complexity of the macroblock stage G _MB The sum of the columns in (i, j, k) represents the column complexity, and M and N represent the horizontal and vertical size of the frame. The method according to claim 6,
If the complexity ratio and the difference value satisfies the condition of Equation 6 below, it is determined that the scene change is made.
(6)

Here, G _{F , ratio} (k) denotes the complexity ratio of the k th frame and the k-1 th frame, and G _{F , diff} (k) denotes the complexity difference between the k th frame and the k-1 th frame. The method of claim 5, wherein
If it is determined that the scene change has been made, the target bit is calculated using the average of the generation bits during the scene change and the average of the generation bits of the interframe in consideration of the available bits. A method for determining a quantization parameter of an H.264 / AVC encoder, characterized by being displayed.
(Equation 7)

Where Fr is the frame rate, N _codeP is the number of encoded P frames, N _GOP is the number of frames of the entire GOP, and b (ni) is the actual occurrence bit of the ni-th frame.

Images