KR20010104058A - Adaptive quantizer according to DCT mode in MPEG2 encoder - Google Patents

Abstract

The present invention provides an adaptive quantizer according to an encoding mode in which an MPEG2 video encoder using an adaptive quantizer determines a quantization level by calculating a measure for determining the complexity of an image in consideration of an encoding mode for each macroblock. The present invention provides a frame distribution processing unit for obtaining minimum block variance of input macroblock data for a frame structure for a rate adjustment; A field variance processor for obtaining a minimum block variance of the input macroblock data with respect to a field structure; A DCT mode determination unit which determines a DCT mode by using a difference signal between an original image to be currently encoded and a motion compensation prediction image; A block variance selecting unit which selects one output from among the output of the frame distribution processing unit and the output of the field distribution processing unit according to the DCT mode determined by the DCT mode determination unit; An activity processing unit which obtains and normalizes an activity of an input macroblock using the block variance selected by the block variance selecting unit; A quantization level determiner which determines a quantization level according to a state of a video output buffer; The quantization level determiner determines the final quantization level of the current macroblock by multiplying the quantization level determined by the quantization level determiner and the activity normalized by the activity processor, thereby improving coding performance and improving image quality. .

Description

Adaptive quantizer according to coding mode of moving picture encoder {DCT mode in MPEG2 encoder}

The present invention relates to a Moving Picture Experts Group 2 (MPEG2) video encoder such as High Density TeleVision (HDTV). In particular, the present invention provides a method for determining the complexity of an image by considering an encoding mode in an MPEG2 video encoder using an adaptive quantizer. The present invention relates to an adaptive quantizer according to a coding mode of a video encoder that can improve image quality by calculating a measure to determine a quantization level.

A typical MPEG2 video encoder uses a method of adjusting a quantization level according to a state of an output buffer in order to transmit a certain amount of image data through a limited transmission channel. In this process, an adaptive quantization method that slightly varies the quantization level determined according to the characteristics of an image is used. The present invention relates to a method for improving its performance.

In recent years, interest in video encoding has been raised, and many standards such as H.261, MPEG1 and MPEG2 have been enacted. These standards are applied to various applications such as video telephony, digital broadcasting, and high-definition TV. Most of them adopt complex motion compensation / discrete cod coding as video coding. These standards are all standards for decoders, and the implementation of an encoder that satisfies them is variable in a range that can be decoded.

1 shows the structure of an MPEG2 video encoder using complex motion compensation / discrete cosine coding.

As shown, first, the motion estimator 2 performs motion estimation using an input image and a previous image stored in the frame memory 1 for prediction. Using the motion vector obtained here, the motion compensator 3 reads the motion compensation prediction value of the block to be currently processed from the previous image stored in the frame memory 1.

The difference between the predicted block thus obtained and the current original block is calculated by the subtractor (4), then discrete cosine coded by the discrete cosine coding unit (DCT unit) 5, and then quantized by the quantization unit 6. send. At the same time, the inverse quantization unit and the inverse discrete cosine coding are performed again through the inverse quantization unit 7 and the inverse DCT unit 8 to transmit the error signal, which is the same operation as the decoder, so that the coded signal and the motion compensation unit 3 The motion compensation prediction signal is summed in summer 9. This summed signal is then stored in frame memory 1 for motion estimation and compensation of the next image.

Image data compression in such a video compression system is performed through motion estimation and compensation, discrete cosine coding, and variable length coding. The MPEG2 video compression method supports various encoding modes for each picture. That is, I-Picture mode, which compresses and encodes using only the information in the current image, and P-Picture and B-Picture mode, which performs motion compensation from the previous or subsequent image in time and compresses and encodes the difference from the original image. Three are allowed. Therefore, the amount of data generated according to the encoding mode of each screen is greatly changed, and the change occurs according to the characteristics of the image.

However, in general, the constant bit rate (CBR) scheme, which is widely considered, requires that the amount of transmitted data be constant on average to allow transmission to a given channel. Therefore, like most video encoding methods, a method for adjusting a transmission bit rate is used in MPEG2. The main method used in MPEG2 application is to output the output data through the output buffer, and to measure the complexity of the current macroblock and the occupancy of the output buffer for each macroblock which is a unit of encoding. activity and adjust the quantization level used for quantization accordingly.

FIG. 2 shows a configuration of a bit rate controller block of a typical MPEG2 video encoder. Referring to the operation thereof, a target bit rate of a current video to be encoded is set for each video according to an encoding mode of the video.

That is, the activity calculator 12 estimates the complexity using the average value of the quantization levels of all macroblocks in the previously encoded image having the same mode as the image to be currently encoded. Then, the target bit rate setting unit 20 sets the target bit rate of the current video by using the complexity and the number of pictures not yet encoded in the GOP (Group of pictures), which is a coding unit of MPEG2. Thus, the video encoder encodes the current video in the set target rate range.

The buffer fullness monitoring unit 21 monitors the fullness of the buffer 11 to provide the current fullness information to the quantization level determiner 22. Accordingly, the quantization level determiner 22 determines the primary quantization level for the macroblock using information such as the target data rate allocated to the current image and the fullness of the current buffer 11.

That is, the fullness of the current buffer 11 is obtained by using the transmission rate and the allocated transmission rate that are actually used to encode the current macroblock before the current macroblock in order to match the target transmission rate of the current image. The quantization level will be determined.

Then, the quantization level adjusting unit 23 slightly adjusts the quantization level obtained by the quantization level determining unit 22 according to the complexity of the current macroblock. This is because the complex image area is not easily seen by human vision even if the picture quality is slightly deteriorated, so that the complex part increases the quantization level slightly, thereby reducing the amount of data generated and the simple part lowers the quantization level to obtain better image quality, thereby improving the subjective image quality. For the purpose. This part is called adaptive quantization.

3 is a block diagram of a conventional adaptive quantizer, and its operation will be described.

First, when the macroblock data having the size of 16 * 16 pixels to be coded is input, the block variance calculator 30 calculates a block variance only for a luminance signal. In this case, the variance is calculated for an 8 * 8 pixel block. As shown in FIGS. 4A and 4B, in MPEG2, a block structure for DCT and quantization supports two types of frame structures and field structures in consideration of interlace scan characteristics. In the field structure, four 8 * 8 blocks are possible, and in the field structure, four 8 * 8 blocks are possible. Therefore, the variance of eight blocks in one macroblock is calculated as shown in [Equation 1].

Here, since the number of blocks is eight as shown in Figs. 4A and 4B, k = 1, 2, ..., 8, and the average value m _{b (k)} of pixels in the block is obtained as shown in [Equation 2].

Then, the activity calculator 31 calculates the activity in the same manner as in Equation 3 using a variance having a minimum value among the eight variances calculated by the block variance calculator 30.

In other words, this value is an activity value used as a measure for determining the complexity of the j-th macroblock.

The activity normalization unit 32 then normalizes the calculated activity. Therefore, the activity normalized by Equation 4 has a value between 0.5 and 2.0.

The average activity used here uses the average value of all macroblock activities of a previously encoded image. Thus, the quantization level determination unit 23 multiplies the obtained normalized activity by the quantization level obtained by the quantization level determination unit 22 of FIG. 2 to obtain the final quantization level of the current macroblock, as shown in [Equation 5]. Will be created.

mquant _j = Nact _j × Q _j

On the other hand, since the quantization level determination process is determined irrespective of the actual DCT and the encoding scheme used for the quantization for real time processing, the quantization level determination process may have an error in encoding efficiency.

That is, the actual encoding mode is determined using the original picture in the I-Picture, and is determined using the differential signal between the original picture and the motion compensation predictive picture in the P-Picture or B-Picture. In addition, since the criterion used for the mode decision is not determined using the variance of blocks but the difference signal of each line of the macroblock, the activity used in the adaptive quantizer is applied for the actual coding. There was a problem that a case different from the mode is generated. That is, since the activity is calculated using the least variance among the variances of the eight blocks, for example, in the adaptive quantizer, the activity of the frame structure is applied and the actual coding is made into the field structure. There is a problem in that encoding performance is degraded.

Therefore, the present invention has been proposed to solve the conventional problems as described above, the object of the present invention is

In the MPEG2 video encoder using the adaptive quantizer, the video encoder is designed to improve encoding performance and image quality by calculating a quantization level by calculating a measure for determining the complexity of an image in consideration of the encoding mode for each macroblock. An adaptive quantizer according to an encoding mode is provided.

In order to achieve the above object, the adaptive quantizer according to the encoding mode of the video encoder according to the present invention,

A frame dispersion processing unit for obtaining a minimum block variance of the input macroblock data for the frame structure for the rate adjustment;

A field variance processor for obtaining a minimum block variance of the input macroblock data with respect to a field structure;

A DCT mode determination unit which determines a DCT mode by using a difference signal between an original image to be currently encoded and a motion compensation prediction image;

A block variance selecting unit which selects one output from among the output of the frame distribution processing unit and the output of the field distribution processing unit according to the DCT mode determined by the DCT mode determination unit;

An activity processing unit which obtains and normalizes an activity of an input macroblock using the block variance selected by the block variance selecting unit;

A quantization level determiner which determines a quantization level according to a state of a video output buffer;

Technical features of the present invention include a quantization level determination unit that determines the final quantization level of the current macroblock by multiplying the quantization level determined by the quantization level determination unit and the activity normalized by the activity processor.

1 is a block diagram of a typical MPEG2 video encoder block;

2 is a block diagram of a block rate control unit of a typical MPEG2 video encoder;

3 is a block diagram of a conventional adaptive quantizer;

4A is a view showing a frame structure block of a typical macro block;

Figure 4b shows a field structure block of a typical macroblock.

5 is a block diagram of an adaptive quantizer block according to an encoding mode of a video encoder according to the present invention;

6 is a block diagram illustrating an internal block structure of the frame dispersion processing unit and the field dispersion processing unit of FIG. 5;

7 is a block diagram of an internal block converting unit and a block distribution calculating unit of the corresponding structure shown in FIG.

<Explanation of symbols for the main parts of the drawings>

100: adaptive quantization control unit 200: frame distribution processing unit

300: field dispersion processing unit 400: delay unit

450: subtractor 500: DCT mode determination unit

600: block distribution selector

Hereinafter, the adaptive quantizer according to the encoding mode of the video encoder according to the present invention will be described in detail with reference to the accompanying drawings.

5 shows an adaptive quantizer block configuration according to an encoding mode of a video encoder according to the present invention.

As shown, the frame dispersion processing unit 200 for obtaining the minimum block variance of the input macroblock data for the frame structure for the rate adjustment; A field distribution processor (300) for obtaining a minimum block variance of the input macroblock data with respect to a field structure; A delay unit (400) for delaying the input macro block data by a time necessary for motion compensation and estimation; A subtractor 450 for calculating a difference between an output block of the delay unit 400 and a motion predicted block; A DCT mode determination unit 500 which determines a DCT mode using an output value of the subtractor 450; A block variance selecting unit (600) for selecting one side output from the output of the frame distribution processing unit (200) and the output of the field distribution processing unit (300) according to the DCT mode determined by the DCT mode determination unit (500); An activity processor 700 for obtaining and normalizing an activity of an input macroblock using the block variance selected by the block variance selector 600; A quantization level determiner 900 which determines a quantization level according to the state of the video output buffer 11; The quantization level determiner 900 is configured to multiply the quantization level determined by the activity processor 700 and the activity normalized by the quantization level determiner 800 to determine the final quantization level of the current macroblock.

As shown in FIG. 6, the frame distribution processor 200 includes a frame structure converter 210 for dividing a format of an input macroblock into four blocks of a frame structure; The block structure calculator 220 is configured to calculate the variance of the frame block transformed by the frame structure converter 210.

As shown in FIG. 7, the frame structure converting unit 210 includes a frame storage unit address adjusting unit 211 for controlling the storage and retrieval of the frame structure type of the input macroblock data; The macroblock storage unit 212 stores and retrieves a macroblock in a frame structure according to the control of the frame storage unit address adjusting unit 211.

As illustrated in FIG. 7, the block dispersion calculator 220 includes a block average value calculator 221 for calculating an average value of a frame structure block output from the frame structure converter 210 in units of macroblocks, and the block average. A frame structure in macroblock units using a block data delay unit 222 for delaying the frame structure block data by a calculation time and a block average value output from the block average value calculator 221 and the delayed frame structure block data. A variance comparison for extracting the minimum frame structure block variance by comparing the variance of data calculated by the data operator 223 and the block variance calculated by the data operator 223 with each other and delaying the DCT mode of the current macroblock. Section 224.

As shown in FIG. 6, the field distribution processor 300 includes: a field structure converter 310 for dividing a format of an input macroblock into four blocks of a field structure; The field structure converter 310 includes a block variance calculator 320 that calculates a variance of the field blocks converted by the field structure converter 310.

As shown in FIG. 7, the field structure converting unit 310 includes: a field storing unit address adjusting unit 311 which controls storing and retrieving field structure type of input macroblock data; The macro block storage unit 312 stores and retrieves macro blocks in a field structure under the control of the field storage unit address adjusting unit 311.

As illustrated in FIG. 7, the block variance calculator 320 includes a block average value calculator 321 for calculating an average value of the field structure block output from the field structure converter 310 in macroblock units, and the block average. By using the block data delay unit 322 for delaying the field structure block data by the time required for calculation, the block average value output from the block average value calculator 321 and the delayed field structure block data, the field block in macroblock units. A data comparator 323 for calculating the variance of the variance, and a variance comparator for comparing the block variances calculated by the data operator 323 with each other to extract the minimum field structure block variance and delaying the DCT mode of the current macroblock 324.

The operation of the adaptive quantizer according to the encoding mode of the video encoder according to the present invention as described above is as follows.

The input image is input for motion estimation and compensation, and the encoder includes a delay unit 400 using a frame memory to adjust the delay time required for motion estimation and compensation, which takes the most delay time. The delay time required for this is inevitably larger than the time for calculating the variance in the adaptive quantizer according to the present invention. Therefore, the present invention almost eliminates the problem caused by the delay time.

If the input image is simultaneously input to the frame dispersion processor and the field dispersion processor, the minimum dispersion should be calculated and stored in each of them. This is because the delay time of the delay unit for the input image and the delay time of the DCT mode determination unit have to wait for the compensation of the delay of the motion estimation and compensation unit. After the decision is made by the DCT mode decision unit, one of the minimum variance of the frame or field structure is selected according to the result, and the activity processor calculates the normalized activity.

The macroblock data to be encoded is input to the frame dispersion processor 200, the field dispersion processor 300, the delay unit 400, and the motion estimation unit 2 in the video encoder.

Then, the frame dispersion processing unit 200 and the field dispersion processing unit 300 reconstruct the frame structure and the field structure as shown in FIGS. 4A and 4B. Since the macroblocks are divided into four blocks for each structure, the frame dispersion processor 200 and the field dispersion processor 300 each calculate four block variances and output the four block variances to the block variance selector 600.

Meanwhile, using the macroblock data through the subtractor 450, the DCT mode determiner 500 determines the DCT mode, which is an encoding mode, and provides the block dispersion selector 600.

Therefore, the block variance selector 600 receives the DCT mode determined by the DCT mode determiner 500, selects the minimum variance of the corresponding structure according to the mode, and outputs it to the activity processor 700. 700) calculates the normalized activity using this minimum variance.

The quantization level determiner 900 determines the quantization level according to the state of the bit string output buffer 11 and provides the quantization level determiner 800 to the quantization level determiner 800. Accordingly, the quantization level determiner 800 multiplies the normalized activity and the quantization level to adjust the final quantization level for the current macroblock and output the result to the quantization unit 6.

In other words, while the input macroblock data is processed for compression encoding, the minimum variance is calculated for each of the frame and field structures. When the DCT mode is determined, one activity is selected to calculate activity according to the result. The final quantization level will be determined. Accordingly, adaptive quantization can be performed in consideration of an encoding mode.

Herein, operations of the frame dispersion processing unit 200, the field dispersion processing unit 300, and the block distribution selecting unit 600 will be described in detail.

The macroblock data inputted through the frame storage unit address control unit 211 and the field storage unit address control unit 311 is divided into the macroblock storage unit 212 in order to be divided into appropriate block formats according to respective encoding methods. ) 312 is stored in an appropriate location and read in block scan format.

At this time, the frame storage unit address adjusting unit 211 reads block data in a frame structure format, and the field address control unit 311 reads block data in a field structure format. The frame structure block data thus read is input to the block average value calculator 221 for calculating the frame structure block variance, and the average value is obtained for each block. The field structure block data read by the field storage unit address adjusting unit 311 is input to the block average value calculating unit 321 for field structure block variance calculation, and the average value is obtained for each block. In this case, block data delay units 222 and 322 are required to compensate for the delayed time in each average value calculation process.

The average value of each structure and each block data obtained through this process are input to the corresponding data calculation units 223 and 323, and four variances are calculated for each macroblock through the subtraction process and the multiplication process.

The variances obtained here are each determined by comparing the minimum value in the dispersion comparison unit 224 and waiting for the DCT mode in the encoding portion to be determined. Then, when the DCT mode is determined, the block variance selector 600 determines one of the minimum frame variance and the minimum field variance according to the DCT mode, and sends it to the activity processor 700. That is, when the DCT mode is the frame DCT, the minimum frame variance is output to obtain the activity of the macroblock, and when the DCT mode is the field DCT, the minimum field variance is output.

When the variance is calculated in this way, the activity of the frame or field structure used in the actual encoding is calculated to determine the quantization level, thereby improving the subjective picture quality more than the conventional method.

As described above, the adaptive quantizer according to the encoding mode of the video encoder of the present invention calculates a measure for determining the complexity of the image in consideration of the encoding mode for each macroblock when compressing the input image having the interlaced scan format. By determining the quantization level, it is possible to improve coding performance and to improve image quality.

Claims (7)

In the adaptive quantizer of the video encoder, A frame dispersion processing unit for obtaining a minimum block variance of the input macroblock data for the frame structure for the rate adjustment; A field variance processor for obtaining a minimum block variance of the input macroblock data with respect to a field structure; A DCT mode determination unit which determines a DCT mode by using a difference signal between an original image to be currently encoded and a motion compensation prediction image; A block variance selecting unit which selects one output from among the output of the frame distribution processing unit and the output of the field distribution processing unit according to the DCT mode determined by the DCT mode determination unit; An activity processing unit which obtains and normalizes an activity of an input macroblock using the block variance selected by the block variance selecting unit; A quantization level determiner which determines a quantization level according to a state of a video output buffer; Adaptive quantization according to the encoding mode of the video encoder, characterized in that it consists of a quantization level determiner to determine the final quantization level of the current macroblock by multiplying the quantization level determined by the quantization level determiner and the activity normalized by the activity processor. group. The apparatus of claim 1, wherein the frame distribution processor A frame structure conversion unit dividing a format of an input macro block into four blocks of a frame structure; Adaptive quantizer according to the encoding mode of the video encoder, characterized in that the block structure calculation unit for calculating the variance of the frame block transformed by the frame structure converter. The method of claim 2, wherein the frame structure conversion unit, A frame storage unit address control unit which controls storage and retrieval of a frame structure type of input macroblock data; An adaptive quantizer according to an encoding mode of a video encoder, characterized by comprising a macroblock storage unit for storing and retrieving a macroblock in a frame structure under the control of the frame storage unit address adjusting unit. The method of claim 2, wherein the block dispersion calculator, A block average value calculating unit for calculating an average value of the frame structure blocks output from the frame structure converting unit in macroblock units, a block data delay unit delaying the frame structure block data by the block average calculation time, and the block average value calculating unit The minimum frame structure block variance is compared by comparing the data variance unit that calculates the variance of the frame structure block in units of macroblocks with the block average value output from the delayed frame structure block data and the block variance calculated by the data operation unit. Adaptive quantizer according to the encoding mode of the video encoder, characterized in that for extracting and delaying while the DCT mode of the current macroblock is determined. The method of claim 1, wherein the field dispersion processing unit, A field structure converting unit for dividing the format of the input macro block into four blocks of the field structure; Adaptive quantizer according to the encoding mode of the video encoder, characterized in that the block structure calculation unit for calculating the variance of the field block transformed by the field structure converter. The method of claim 5, wherein the field structure conversion unit, A field storage address control unit for controlling field structure format storage and retrieval of input macroblock data; An adaptive quantizer according to an encoding mode of a video encoder, characterized by comprising a macroblock storage unit for storing and retrieving a macroblock in a field structure under the control of the field storage unit address adjusting unit. The method of claim 5, wherein the block dispersion calculator, A block average value calculating unit for calculating an average value of the field structure blocks output from the field structure conversion unit in macroblock units, a block data delay unit for delaying the field structure block data by the block average calculation time, and the block average value calculating unit By using the block average value output from the block data and the delayed field structure block data, the data calculation unit for calculating the variance of the field block in macroblock units and the block variance calculated by the data operation unit are compared with each other to obtain the minimum field structure block variance. Adaptive quantizer according to the encoding mode of the video encoder, characterized in that for extracting and delaying while the DCT mode of the current macroblock is determined.