KR100480698B1 - MPEG encoding method - Google Patents

Abstract

In particular, the present invention relates to a method of encoding an MPEG. The method includes: detecting an edge using a mask having a predetermined size for an input image, outputting an edge parameter as a detection result, and applying the edge parameter to determine a type of a macroblock to be currently encoded. And determining the final quantization step to be used for actual quantization by multiplying the spatial activity obtained by the edge parameter of the macroblock to be encoded currently with the reference quantization parameter, thereby accurately calculating the performance of the encoder by accurately estimating the amount of bits generated after encoding. Can be improved. In addition, edge detection can be implemented only with an adder, and in particular, by applying the same edge parameter to macro block type and bit rate control, hardware complexity can be reduced.

Description

MPEG encoding method

The present invention relates to an MPEG (MPEG) encoder, and more particularly, to an MPEG encoding method for controlling the macroblock type and bit rate of an MPEG encoder using a parameter by edge detection.

The use of digital video, including video conferencing / telephone, digital video cassette recorder (VCR), digital versatile disc (DVD), and digital camera surveillance and security systems, is expanding worldwide.

In particular, since a lot of resources are required for storing and transmitting a video, compression and encoding are performed to store and transmit the video.

A widely used standard for this purpose is the Moving Picture Experts Group (MPEG). The MPEG is largely based on a discrete cosine transform (DCT) quantization, a motion estimation and compensation technique, and variable length coding based on a code generation probability, and includes bit rate control for fixed bit rate transmission.

At this time, the algorithm mainly used in the MPEG encoder is the material of MPEG-2 under international standardization in IS / IEC JTC1 / SC29 / WG11, which is an organization of ISO (International Organization for Standardization) (document number AVC-491, TEST MODEL 5) This is the suggested method.

FIG. 1 is a block diagram illustrating a general MPEG encoder, and includes an encoder 100, a bit rate controller 200, and an MC controller 300 that controls motion compensation of the encoder 100.

That is, the subtractor 101 outputs the difference data with the motion-compensated data from the motion compensation unit 109 to the DCT unit 102 if the input data is an I picture or the P or B picture. The DCT unit 102 DCTs the input data and outputs the quantized unit to the quantization unit 103.

In this case, the DCT unit 102 removes the correlation of data through two-dimensional axis transformation. To this end, after dividing the picture into block units, the DCT unit 102 transforms each divided block according to the DCT equation. The axially transformed data tends to converge in one direction (lower frequency side). Only the collected data are quantized by the quantization unit 103 and then output to the variable length coding (VLC) unit 104. The VLC unit 104 outputs the frequently-valued values to the buffer 110 after reducing the total number of bits by displaying a small number of bits and a rare value to a large number of bits. The buffer 110 temporarily stores the VLC data and outputs the data for transmission or storage. The buffer 110 calculates the fullness of the buffer and outputs the buffer to the bit rate controller 200.

In addition, the DCT coefficients quantized by the quantization unit 103 are input to the IQ unit 105 again, dequantized, and then output to the IDCT unit 106. The IDCT unit 106 IDCTs the inverse quantized DCT coefficients and outputs them to the adder 107. The adder 107 adds the IDCT value and the motion compensated value to restore the complete image, which is the final pixel value, and then stores the result in the memory 108 for motion compensation. The motion compensator 109 performs motion compensation using the previous frame read from the memory 108 according to the MC controller 300 and then outputs the motion compensation to the subtractor 101 and the adder 107.

In this case, the MC control unit 300 determines whether the input data is a P-picture, MC / No-MC determination, intra / non-intra determination, or B-picture, forward MC, back. Variance and Mean Squared Error (MSE) are used in word MC, interpolated MC decision, intra / non-intra decision. Here, it is determined whether motion compensation is performed or not for the macro block determined as non-intra. At this time, if it is determined as No-MC, only the difference value between the current and previous macro blocks is output without outputting the motion vector.

First, the determination of the motion prediction method selects the smallest MSE of the current macroblock and the past or future macroblock obtained by the motion prediction or the average of the past and future macroblock.

In this case, in the case of the P-picture, MC / no MC is determined as shown in FIG. 2 by applying Equations 1 and 2 to the motion predicted result.

In Equation 1 and 2, blk (n) represents the nth coefficient of the macroblock currently encoded. blk _mc (n) is the n th coefficient of the reference macro block of the past image obtained by motion prediction, and blk _no-mc (n) is the n th coefficient of the macro block at the same position as the current macro block of the current image in the past image. Indicates.

In the case of determining MC / No-MC in FIG. 2, more emphasis is placed on No-MC because, in the case of P-picture, if No-MC does not transmit a motion vector, the actual coding efficiency increases.

When the motion prediction method is determined by the above-described method, intra / non-intra determination is performed by applying Equations 3 and 4 below.

Equation 3 represents the variance of the macroblock currently encoded, and Equation 4 represents the MSE of the current macroblock and the reference macroblock used to determine the motion prediction method.

That is, when it is determined whether motion compensation is used, the result is compared with the variance of the reconstructed block and the variance of the encoding macroblock to determine whether to encode the intra macroblock or the non-intra macroblock. In this case, if it is determined to be intra, the input macro block is encoded through DCT, quantization and variable length coding. If it is determined as non-intra, the difference between the input macro block and the reference macro block of the previous frame is determined through DCT, quantization, and VLC. Is encoded.

On the other hand, the MPEG standard uses variable length coding (VLC) as described above. For this reason, the bits generated by the encoder are also variable. Therefore, the bit rate control unit 200 controls bit rate for use in applications requiring fixed bit rate transmission or the size of an encoded image.

The bit rate controller 200 adjusts the generation amount of data by varying the step size of the quantization unit 103 mainly according to the buffer fullness of the buffer 110. That is, if the number of bits generated is greater than or equal to the reference value, the amount of data to be filled in the buffer 110 is increased. Therefore, the quantization step size is increased to reduce the number of bits to be generated later. The number of bits is increased to adjust the state of the buffer 110 as a whole. The generation bit is controlled by the quantization step. The image quality is greatly influenced by the bit rate control.

At this time, referring to TM5, the bit rate controller 200 performs the following three steps.

First, a first step is bit allocation, in which complexity is predicted and target bits are allocated. That is, a constant bit rate is allocated in units of GOP (Group Of Picture) according to the transmission bit rate, and bits to be allocated to each picture in the GOP are allocated according to the complexity of each picture (I, P, B frame). The complexity of a picture is obtained by multiplying the generated bit amount of one picture by the average value of the steps.

The second step is to adjust the rate (i.e., bit rate), and the bit rate control is performed in units of macro blocks. That is, the reference quantization parameter calculator 201 calculates a reference quantization parameter for each macro block according to the fullness of the virtual buffer 110. The bit rate is controlled so that each picture can be encoded according to the bit allocated in the first step. In this case, the reference quantization parameter for the current macroblock is determined by the virtual buffer fullness by the generation bits up to the previous macroblock.

Here, it is assumed that each picture has an arbitrary virtual buffer, and a method of adjusting the quantization parameter according to the state of the buffer is used.

The third step is adaptive quantization, and the spatial activity calculating unit 202 of the bit rate control unit 200 calculates the spatial activity of the macroblock to be encoded as shown in Equation 5 below.

sub-blocks (n = 1 ... 4): the frame-organized sub-bloks

sub-blocks (n = 5 ... 6): the frame-organized sub-bloks

In Equation 5, vblk _n is a subblock value obtained by variance. After normalizing the spatial activity thus obtained, the quantization parameter generator 203 multiplies the normalized spatial activity by the reference quantization parameter obtained in the second step to determine the quantization parameter to be actually used for quantization. That is, the quantization unit 103 determines the final quantization step by multiplying a quantization parameter output from the quantization parameter generator 203 by a constant, for example, 2, and quantizes the DCT coefficients using the determined quantization step size. That is, adaptive quantization is a method of increasing the subjective picture quality, and is a method of changing the reference quantization parameter according to the complexity of the current macroblock.

As such, macroblock type determination and bit rate control are important factors that determine the performance of the encoder, and the algorithm proposed in TM5 is mainly used.

That is, the type of the macroblock is determined by the MSE with the reference macroblock obtained by variance and motion prediction of the current macroblock. In addition, in case of bit rate control, variance is used to obtain spatial activities in the adaptive quantization step.

This is to estimate the amount of bits generated after performing VLC using variance and MSE, quantizing the result, and then VLC.

In this case, when the same value is distributed or MSE, when the error value is widely spread over the block and when the partial value is biased, the high frequency component occurs after DCT because the partial value is biased. Because the value is large, even quantization generates more bits. At this time, an error in the amount of bits generated is large even at the same dispersion value. Therefore, dispersion or MSE does not reflect high frequency components well. In other words, a parameter that can more accurately predict the amount of bits generated in encoding is required.

In addition, since a multiplier and an adder are required to obtain variance and MSE, hardware is complicated.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to provide an MPEG encoding method capable of better reflecting a high frequency component to improve an image quality of a corresponding bit.

Another object of the present invention is to provide an MPEG encoding method which can reduce the complexity of hardware in implementation.

In order to achieve the above object, the MPEG encoding method according to the present invention applies an edge detection algorithm that reflects a high frequency component to perform macro block type determination and bit rate control, while simultaneously performing macro block type determination and bit rate control. It is characterized by using a parameter obtained by the edge detection algorithm.

The MPEG encoding method according to the present invention includes detecting an edge using a mask having a predetermined size with respect to an input image, outputting an edge parameter as a detection result, and applying the edge parameter to determine a type of a macroblock to be currently encoded. And determining a final quantization step to use for actual quantization by multiplying a reference quantization parameter by obtaining a spatial activity from an edge parameter of a macroblock to be currently encoded.

The determining of the type of the macro block may include determining the type of the macro block by the edge of the current macro block and the edge of the reference macro block obtained by the motion prediction.

The determining of the type of the macroblock may include comparing an edge parameter with a set threshold to determine whether to input the intrablock or non-intracoded macroblock, and determine the edge parameter of the current macroblock and the edge of the previous macroblock. It is characterized by determining whether to compensate for motion by comparing the parameters.

Other objects, features and advantages of the present invention will become apparent from the following detailed description of embodiments taken in conjunction with the accompanying drawings.

Hereinafter, with reference to the accompanying drawings illustrating the configuration and operation of the embodiment of the present invention, the configuration and operation of the present invention shown in the drawings and described by it will be described as at least one embodiment, By the technical spirit of the present invention described above and its core configuration and operation is not limited.

In general, the amount of bits generated after VLC is greatly influenced by data in the frequency domain after DCT. In other words, as data is biased toward the lower frequency, the generation bit is smaller.

That is, most of the bits generated after encoding are for block coefficients.

Therefore, in the present invention, in order to more accurately estimate the amount of bits generated after encoding, the difference between adjacent pixels that reflect high frequency components well, instead of variance and MSE, that is, edge detection algorithm is applied to macroblock type determination and bit rate control. It is.

4 is a diagram illustrating a method of detecting characteristic values of a macroblock by a 3x3 edge detection algorithm.

Then, the characteristic parameter obtained by edge detection is called edge parameter.

The edge parameter E is obtained by applying Equation 6 below.

In Equation 6, Hr and Hc are operators used for edge detection, and W [i, j] represents a current macroblock or a reference macroblock obtained by motion prediction with the current macroblock.

Here, the larger the value of the edge parameter E, the more high frequency components are included in the macro block.

The edge parameters thus obtained can be equally applied to the spatial activity of the type determination of the macroblock and the adaptive quantization of the bit rate control. That is, the edge parameter may be applied instead of the variance and MSE used for determining the macroblock type, and may be applied when obtaining the spatial activity of the adaptive quantization.

For example, the edge parameter is compared with a preset threshold to determine whether to input the intra or non-intra coded macro block, and whether the motion compensation is performed by comparing the edge parameter of the current macro block with the edge parameter of the previous macro block. Determine. That is, the type of the macroblock is determined by the edge of the current macroblock and the reference macroblock obtained by the motion prediction.

In addition, the edge parameter may be applied instead of variance when calculating the spatial activity to determine the quantization step size. That is, the spatial activity is obtained by applying the edge parameter of the macroblock to be encoded instead of variance, and then the final quantization step to be used for actual quantization is determined by multiplying the reference quantization parameter obtained in the second step according to the fullness of the virtual buffer 110. .

5A to 5C show experimental results of the amount of bits generated by variance and edge detection. Experimental results show that the edge parameter enables more accurate estimation of the amount of bits generated.

6A to 6C are PSNR comparison graphs for the proposed method and TM5, and the present invention shows an improvement of about 1.2 dB.

As described above, according to the MPEG encoding method according to the present invention, the amount of generated bits is accurately estimated by estimating the amount of generated bits after encoding by performing macroblock type determination and bit rate control by applying edge parameters obtained by edge detection. This improves the performance of the encoder. In addition, edge detection can be implemented only with an adder, and in particular, by applying the same edge parameter to macro block type and bit rate control, there is an effect of reducing hardware complexity.

Those skilled in the art will appreciate that various changes and modifications can be made without departing from the spirit of the present invention.

Therefore, the technical scope of the present invention should not be limited to the contents described in the embodiments, but should be defined by the claims.

1 is a block diagram of a typical MPEG encoder

2 is a typical MC / No-MC determination graph

3 is a typical intra / non intra determination graph

4 illustrates an example of edge detection in an MPEG encoder according to the present invention.

5A to 5C show experimental results of the estimated amount of generated bits by applying the edge parameter and the variance of the present invention.

6A-6C are PSNR comparison graphs for the present invention and TM5

Explanation of symbols for main parts of the drawings

100: encoding unit 101: subtractor

102: DCT unit 103: quantization unit

104: VLC unit 105: inverse quantization unit

106: IDCT unit 107: adder

108: frame memory 109: motion compensation unit

110: virtual buffer 200: bit rate control unit

300: MC control unit

Claims (4)

In the MPEG encoding method for performing discrete cosine transform (DCT), quantization, variable length coding, motion compensation according to the type of macro block, Detecting an edge using a mask of a predetermined size on the input image and outputting an edge parameter as a detection result; Determining the type of macroblock to currently encode by applying the edge parameter; And And determining a final quantization step to use for actual quantization by multiplying a reference quantization parameter by obtaining a spatial activity by applying an edge parameter of a macroblock to be currently encoded. The method of claim 1, In the edge detection step, the edge parameter (E) is obtained by applying the following equation. Here, Hr and Hc are operators used for edge detection, and W [i, j] represents a current macroblock or a reference macroblock obtained by motion prediction with the current macroblock. The method of claim 1, wherein the determining of the type of macro block comprises: And the type of the macroblock is determined by the edge of the current macroblock and the reference macroblock obtained by motion prediction. The method of claim 3, wherein the determining of the type of macro block comprises: Compares the edge parameter with a set threshold to determine whether to input the intra or non-intra coded macro block, and compares the edge parameter of the current macro block with the edge parameter of the previous macro block to determine whether to compensate for motion. MPEG encoding method, characterized in that.