KR0134342B1 - Coding apparatus and method of motion - Google Patents

Abstract

The present invention relates to a method and apparatus for encoding by motion estimation. In particular, a motion estimation unit for estimating motion between adjacent screens in a moving picture composed of continuous screens, and a classification for displaying directionality according to the estimated motion characteristics By quantizing the image data by the quantization matrix memory which can output the quantization step of the quantization matrix sequentially according to the classification and the motion components classified, not only the compression efficiency can be increased but also the overall image quality can be improved. Provide the effect.

Description

Encoding method by motion estimation and apparatus

1 is a block diagram showing an encoding apparatus by motion estimation according to the present invention.

2 is a detailed block diagram showing the same classification in the FIG. 1 apparatus.

* Explanation of symbols for main parts of the drawings

11 DCT unit 12 quantization unit

13 variable length encoding unit 14 buffer

15: inverse quantization unit 16: inverse DCT unit

17: frame memory 18: winter weight

19: Compensation Compensation Part 20: Compensation Division

21: separating unit 22: calculating unit

23: comparison unit 24: decoder

30: quantization matrix memory 31: quantization factor selector

32: multiplier

The present invention relates to an encoding system for compressing a data amount by encoding image data, and more particularly, by quantizing an image signal by adaptively weighting a quantization factor to a quantization matrix selected according to the motion estimation information data of the image signal. The present invention relates to a method and apparatus for encoding by motion estimation that can increase the compression efficiency of data.

Recently, in a system for transmitting and receiving video and sound, a video signal audio signal is encoded into a digital signal, transmitted or stored in a storage unit, and decoded and reproduced. In such an encoding and decoding system, a technique for further compressing the amount of transmission data is required to maximize data transmission efficiency. In general, a method used for encoding an image signal includes a transform coding method, a differential pulse code modulation (DPCM) method, a vector quantization method, and a variable length coding method. These coding schemes are used to compress the total amount of data by removing redundancy data included in the digital video signal. In order to perform such an encoding method, a screen is divided into blocks having a predetermined size, and predetermined conversion is performed on each block or a difference signal between blocks to convert image data into a conversion coefficient of a frequency domain. Examples of data transformation methods for each block include DCT (Discrete Cosine Transform), WHT (Walsh-Hadamard Transform), DFT (Discrete Fourier Transform), and DST (Discrete Sine Transform). The transmission data is compressed by appropriately encoding and transforming such transform coefficients according to data characteristics. In addition, since human vision is sensitive to high frequency and low frequency, the amount of transmission data can be further reduced by attenuating the signal of the high frequency part in processing the video signal.

In general, when a video signal is compressed using a variable length encoding and decoding device, a buffer is provided so that the amount of input / output data is kept constant. In order to prevent overflow or underflow of the buffer, the amount of code to be output is controlled by adjusting the quantization step according to the state of the buffer. In addition, since there are many similar parts between the screen and the screen, the motion vector is calculated by estimating the motion and the data is compensated using the motion vector, and the difference between the adjacent screens is very small. The data can be further compressed. In addition, how to determine the quantization step, which is one of the most important parameters in the encoding, is a very important problem. The conventional quantization step is determined by multiplying the quantization factor output from the quantization matrix memory by the quantization factor selected by the quantization factor selector according to the state of the buffer connected to the output terminal. If the quantization step is large, the output code amount is small, but the error is large when reconstructed. On the contrary, when the quantization step is small, an error is small when the output code amount is restored instead. Since the quantization step determination method adjusts the quantization factor according to the output buffer amount when the output bit amount is determined, the overall bit amount can be precisely adjusted, but there is a problem in that the nature of the screen is not considered at all.

Therefore, an object of the present invention is to determine the quantization step by adaptively matching the motion estimation information data and the buffer state of the image data in consideration of the visual characteristics of the human in encoding the image data into the transmission data compressed data amount The present invention provides a method for improving the visual quality of the entire screen.

Another object of the present invention is to provide an apparatus for implementing the encoding method according to the above-described motion estimation. As described above, an object of the present invention is to estimate an operation between an input image data and a previous frame signal in a method for encoding image data divided into blocks having a predetermined size using an optimal quantization step by motion estimation. Detecting a motion direction of the motion estimation value, sequentially outputting a quantization step of the quantization matrix corresponding to the detected motion direction, and weighting a predetermined quantization factor to the quantization step of the quantization matrix Obtaining an optimal quantization step and quantizing each sample data of the block data according to the optimal quantization step.

It is another object of the present invention to encode an image data divided into blocks having a predetermined size using DPCM, and to provide an encoding apparatus having a buffer for transmitting encoded data at a predetermined bit rate, the image data being applied through an input terminal. Extracts the data having the most similar pattern from the previous screen data and calculates the motion vector representing the motion between the two data, and checks the components of the motion vector applied from the motion tracking part to check the binary signal A quantization matrix memory for storing a plurality of quantization matrices according to a moving direction, selecting and outputting an optimal quantization matrix in response to a binary signal applied from the same sorting unit, and controlling the buffer state A quantization factor selector that receives and outputs a corresponding quantization factor and the optimal quantity A multiplier that obtains a quantization step by weighting a quantization factor to a magnetization matrix, and a quantization unit that sequentially quantizes each sample data of the block data according to the quantization step.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram showing an apparatus for encoding video data by motion estimation according to the present invention. The apparatus of FIG. 1 has a first adder A1 for calculating the error data between the block data and the predetermined feedback data applied through the input terminal. The output terminal of the first adder A1 is connected to a DCT section 11 for converting error data into data in the frequency domain. The output terminal of the DCT unit 11 is connected to a quantization unit 12 for converting the transformation coefficients of representative values of a predetermined level. A variable length coding unit 13 for variable length coding the quantized data and compressing the data is connected to the output terminal of the quantization unit 12. The output of the variable length coding unit 13 is connected to a buffer 14 for keeping the bit rate of the transmission data V _CD constant. An output terminal of the quantization unit 12 is also connected to an inverse quantization unit 15 and an inverse DCT unit 16 which inversely quantize a quantization coefficient and then inversely transform it into image data of a spatial domain. A second adder A2 and a frame memory 17 for reconstructing a screen by adding inverse transform data and a predetermined feedback data are connected to an output terminal of the inverse DCT unit 16. Between the frame memory 17 and the input end, a tracking unit 18 for finding block data having a pattern most similar to the input image data in the frame memory 17 and calculating the motion vector MV representing the movement between blocks is provided. Also connected to the output end of the dynamic estimation unit 18 is a dynamic compensation unit 19 for compensating data using the dynamic vector (MV). An output terminal of the copper weighing unit 18 is also connected to a copper classifier 20 that generates a binary signal according to the horizontal and vertical components of the copper vector MV. A quantization matrix memory 30 for outputting a corresponding quantization matrix in response to a binary signal output from the same classification unit 20 is connected to an output terminal of the coclassification unit 20. The output terminal of the buffer 14 is connected to a quantization factor selector 31 for selecting and outputting the quantization factor SF according to the buffer state detection value. The quantization matrix memory 30 and the quantization factor selector 31 are constituted by a multiplier 32 that multiplies the quantization matrix by the quantization factor to obtain a quantization step QS.

In FIG. 1, image data in block units (8 pixels x 8 pixels) input through an input terminal calculates error data in a predetermined feedback data and the first adder A1. The error data is converted into data in the frequency domain by the DCT unit 11, and the energy of the converted conversion coefficient is mainly collected toward the low frequency side. The quantization unit 12 converts the transform coefficients into representative values of a predetermined level through a predetermined quantization process. In other words, the quantization step of the quantization unit 12 is obtained by multiplying the quantization matrix and the quantization factor SF of the quantization matrix memory 30 prepared in consideration of the spatial visual characteristics of the human being according to the frequency by the multiplier 32. If the quantization step is large, the amount of code output is small and the error is large. On the contrary, when the quantization step is small, the amount of code output is large and the error is small. For the quantized DCT coefficients, the variable-length encoding unit 13 transmits a short length code to a symbol having a low frequency for a pawn having a high frequency after Zigzag Scan and 0 Run Length. In this case, the transmitted data (V _CD ) is compressed by giving a long code. Then, the buffer 14 outputs the buffer state detection value so that the amount of compressed data supplied from the variable length coding unit 13 and the amount of data output from the buffer are the same. The buffer state detection value is applied to the quantization factor selector 31 to select one of a plurality of quantization factors stored in the quantization factor selector 31. The selected quantization factor SF is multiplied by the multiplier 32 and the quantization matrix value output from the quantization matrix memory 20 to output a quantization step. This quantization step determination process will be described later in detail. On the other hand, since there are many similar parts between the screen and the screen, in the case of a screen with a slight movement, the motion vector (MV) is calculated by estimating the motion, and the difference signal between the adjacent screens is compensated by using the movement vector Is very small, which can further compress the transmission data. In order to perform such dynamic compensation, the inverse quantization unit 15 and the inverse DCT unit 16 inversely quantize the quantization coefficients output from the quantization unit 12 and then inversely transform the image data in the spatial domain. The image data output from the inverse DCT unit 16 is added to the predetermined feedback data by the second adder A2 and stored in the frame memory 17 to reconstruct the screen.

That is, this process is continued for one frame. After all processing of one frame is completed, the same screen as the screen that has already been compressed and transmitted is stored in the frame memory 17. It means that the picture is the same as the picture reproduced at the receiving side or at the time of reproduction. Next, when the screen is input, the tracking unit 18 finds the block data of the pattern most similar to the input image data from the frame data stored in the frame memory 17 and calculates a moving vector MV representing the movement between the two blocks. . This dynamic estimation is a block match, in which a screen is divided into a predetermined block size, and a block of the current screen is obtained by accumulating and accumulating a difference for each pixel of the block size corresponding to the position of the previous screen. The cumulative sum refers to the difference between the block in the current picture and the block in the previous picture. And this process continues by moving one pixel. The maximum momentum and calculation between adjacent screens generally do not exceed the range of -8 to +7 to -16 to +15 at the reference point. The position with the lowest cumulative sum within such a range can be regarded as moving from the reference point of the previous frame to that position.

Therefore, the vector V is equal to the movement amount. This moving vector (MV) is transmitted to the variable length coding unit 13 for use in the decoding system, and also to the moving compensation unit 19 and the same classification unit 20. The compensating unit 19 reads out the corresponding block data from the moving vector MV from the frame data of the frame memory 17 and supplies it to the first adder A1. Then, the first adder A1 calculates error data between the input image data and the block data output from the compensator 19, and the error data is encoded again and transmitted to the receiving side. The case of encoding a series of input screens as described above is called an intra frame (I), and the case of encoding by estimating motion from a previous screen is called a predicted frame (P). In addition, in order to increase the compression efficiency, the method of predicting from the future screen is called a backward predicted frame, and the method of configuring the screen by predicting and interpolating from the past and future screens is called an interpolated frame (B). do. The order of the screens can also be configured in various ways, such as [IBBPBBPBBPI], depending on the purpose. First degree is [IPIP ... ]. The co-classifier 20 supplies a binary signal corresponding to the co-vector to the quantization matrix memory 30.

2 shows a detailed block diagram of the same classifier 20. The same sorting part of FIG. 2 is provided with the separating part 21 which isolate | separates the copper vector MV supplied from the copper weighting part 18 into horizontal and vertical components. An arithmetic unit 22 having an arithmetic unit corresponding to the output terminal of the separating unit 21 is provided to the arithmetic unit, and the arithmetic unit 22 which obtains the absolute value of each component value input to the arithmetic unit is connected to the separating unit 21. There is a comparator 23 having a plurality of comparators COMP1 and COMP2 and generating a binary signal by receiving an absolute value from each corresponding calculator of the calculator 22 and comparing it with a preset value. A decoder 24 for decoding a binary signal is connected to an output terminal of the comparator 23.

First, the copper vector MV output from the sinusoidal unit 18 is applied to the separating unit 21 of the same classifier 20. The separating unit 21 receives the copper vector MV by the horizontal component X component (Y component) separating unit 21 (MV _bX, MV _Y ) to each calculator to obtain an absolute value of the component value. The comparison unit 23 receives an absolute value of each motion component from a corresponding calculator and compares it with a preset reference value. At this time, if the value of the inputted motion component is greater than the threshold, the motion of the component is insignificant. If the value of the inputted motion component is less than the reference value, the binary signal is generated. In this case, a binary signal in a low state is generated.

The decoder 24 decodes binary signals applied from the comparators COMP1 and COMP2 of the comparator 23 to select one of a plurality of quantization matrices stored in the quantization matrix memory 30. The relationship between the motion components and the output of the decoder 24 is shown in Table 1 below.

TABLE 1

The quantization matrix memory 30 selects and outputs a quantization matrix corresponding to a signal applied from the decoder 24 of the same classifier 20. As shown in Table 2 below, the quantization matrix has a larger quantization step in the high frequency region (lower right side of the table) than in the low frequency region (upper left side of the table). This is because the visual effect of the error between the quantization coefficients is less sensitive in the high frequency region than in the low frequency region when quantizing the image data.

TABLE 2

Here, the quantization matrix shown in Table 2 is a matrix for 8x8 block data, and has a quantization step of 8x8. As shown in Table 3, the quantization matrix memory 30 multiplies the factor matrix according to the movement direction with the quantization matrix (shown in Table 2) to store and store optimal quantization matrices according to the movement direction.

TABLE 3

The quantization matrix memory 30 selects and outputs one of a plurality of quantization matrices stored in response to the decoding signal applied from the decoder 24 of the co-classifier 20. That is, the quantization matrix memory 30 selects and outputs the quantization matrix made by Table 3 (a) when the binary signal output from the decoder 24 of the same classifier 20 is 0, and if it is 1, Table 3 ( b), if 10, Table 3 (c) and if 11, Table 3 (d) selects and outputs the quantization matrix. When motion occurs, the quantization matrix is larger than its original value, which means that it is then insensitive in time. In addition, the quantization step becomes larger in the moving direction depending on the directional component, thereby increasing the compression efficiency while preventing visual errors from being unobtrusive. In addition, the buffer state detection value fed back from the buffer 14 is applied to the quantization factor selector 31 to selectively output the quantization factor. The multiplier 32 multiplies the quantization matrix, which is selected and output from the quantization matrix memory 30 by the operation estimation value, and the quantization factor SF, which is selected and output from the quantization factor selector 31 by the buffer state detection value, to quantize the quantization matrix. Step QS is determined. The quantization unit 12 receives the N × N block transform coefficients from the DCT 11 and sequentially receives the quantization step QS through the multiplier 32, and receives each transform coefficient of the N × N block data N × N. The quantization coefficient is correspondingly quantized according to each quantization step of the quantization matrix, and the quantization coefficient is transmitted to the variable length encoding unit 13.

As described above, the present invention relates to the encoding method and the apparatus by motion estimation, and the quantization factor is variably selected by the buffer state detection value returned from the buffer in accordance with the amount of data encoded. The quantization matrix is variably selected according to the dynamic components of and the block data is quantized according to the quantization step of the new quantization matrix obtained by multiplying the selected optimal quantization matrix by the quantization factor. Therefore, since the quantization step is determined according to the degree of motion on the screen, the visual quality of the entire screen can be improved.

Claims (5)

An encoding apparatus having a buffer for encoding image data divided into blows having a predetermined size using DPCM and transmitting the encoded data at a constant bit rate, wherein the data has a pattern most similar to the image data applied through an input terminal. Extracts the previous vector from the previous screen data, and calculates a motion vector representing the motion between the two data, the dynamic classification unit that checks the components of the motion vector applied from the motion tracking part and outputs a binary signal corresponding thereto, and the motion direction. A quantization matrix memory for storing a plurality of quantization matrices and selecting and outputting an optimal quantization matrix in response to a binary signal applied from the same classifier, and selecting and outputting the quantization factor under control of the buffer state. A quantization factor selection unit weights quantization factors to the optimal quantization matrix Coding device according to the multiplier and each sample data of the block data to obtain the quantization step in the motion estimation that includes a quantizer for sequentially quantized according to the quantization step. 2. The apparatus of claim 1, wherein the same classifier divides and outputs a horizontal vector and a vertical component of the dynamic vector applied from the sinusoidal unit, and calculates an absolute value of the horizontal component and vertical component values applied from the separating unit. And a comparator for generating a binary signal by comparing the absolute values of the horizontal component and the vertical component applied from the calculator with a preset reference value, and decoding the binary signal applied from the comparator for optimal quantization according to the movement direction. And a decoder for outputting a signal for selecting a matrix. 2. The encoding apparatus of claim 1, wherein the quantization matrix memory stores a matrix value obtained by multiplying a preset quantization matrix by a factor matrix having a large value in the movement direction. A method for encoding image data divided into blocks having a predetermined size using an optimal quantization step by motion estimation, the method comprising: estimating motion between input image data and a previous frame signal, and a motion direction of the motion estimation value Detecting a quantization step of the quantization matrix corresponding to the detected motion direction, weighting a predetermined quantization factor to the quantization step of the quantization matrix to obtain an optimal quantization step, and the block data. And quantizing each sample data according to the optimal quantization step. The method of claim 4, wherein the detecting of the direction of motion comprises: separating the motion estimation values into horizontal components and vertical components, respectively, and outputting the absolute values of the horizontal and vertical components; Generating a binary signal by comparing the respective reference values with each other, and outputting a signal for selecting an optimal quantization matrix according to a motion direction by decoding the binary signal. Encoding method.