KR100351568B1 - Apparatus and method for high compression to consider edge direction of motion compensated prediction - Google Patents

Abstract

The present invention is to provide a high compression device and a method thereof in consideration of the boundary direction of the motion compensation prediction block, the present invention is a video encoder comprising a motion compensation prediction block, a DCT block and a quantization block, the motion compensation A boundary line checker which finds a characteristic of boundary direction in the signal output from the prediction block and obtains a difference value based on the found boundary line; A compensation prediction unit obtaining a new motion compensation prediction value by compensating with the difference value obtained by the boundary checker; H.263, which is a standard image compression method, by including a scan unit configured to scan DCT coefficients of a signal output from the compensation predictor and passed through the DCT block and the quantization block by using the output of the boundary checker. The motion compensation prediction block can be used in + to improve the motion compensation prediction error block and reduce the compressed data by effectively selecting a scanning method.

Description

Apparatus and method for high compression to consider edge direction of motion compensated prediction

The present invention relates to a high compression device and a method thereof considering the boundary direction of a motion compensation prediction block. In particular, a motion compensation prediction error block is improved by using a motion compensation prediction block in H.263 +, a standard video compression method, and a scanning method is provided. The present invention relates to a high compression device and a method thereof in consideration of the boundary directionality of a motion compensation prediction block suitable for effectively selecting and reducing compressed data.

In general, as ISDN was standardized by ITU-T (formerly CCITT) in the 80s, video telephony using ISDN network was considered. This videophone (H.320 terminal) is targeted at the primary interface (two 64 Kbps channels and one 16 Kbps channel) to the ISDN network in the home, or the primary group interface (up to 30 64 Kbps channels) used in the office. It has a bit rate of (P = 1 to 30). The international standard for video compression at this time is H.261.

In the early 90s, telecommunication companies such as ATT and MCI in the United States developed and distributed video telephones using telephone networks. However, the compatibility of communication devices is very important in nature, and there was no compatibility between these video phones. MPEG4 was originally started to meet this demand, but gradually expanded its scope, and eventually focused on multimedia database access and wireless multimedia communications.

Based on this background, ITU-T SG15 has developed international standard for video telephones operating within 64Kbps using telephone network. Because it aimed to complete in a short period of time, rather than accepting a new algorithm like MPEG4, it moved toward the H.320 videophone with the dog algorithm. This is the H.324 terminal.

The H.324 videophone is connected to the telephone network via a V.34 modem (up to 28.8Kbps at the highest speed for a telephone line modem), and video compression uses H.263, a significant improvement over H.261. Voice compression uses GLP, CELP.

The improvement of H.263 over H.261 is as follows. First, in order to encode a motion vector of each macro block, a more efficient method of taking the median of three vectors is used in consideration of high correlation with neighboring macro blocks. In this way, a data reduction of about 10% is obtained. In addition, motion estimation for each block that subdivides motions within one macroblock is possible.

For efficient quantization of the DCT transform coefficients, the quantizer is improved to save about 3% of data, and the variable length encoding of the quantized transform coefficients improves the two-dimensional code of H.261, JPEG, MPEG1, and MPEG2 to improve the in-screen and inter-screen. Use 3D code considering information. You can save about 5% of your data here.

The syntax of bit strings is also greatly simplified compared to H.261 to reduce the overhead of pure information bits.

In addition, it is a syntax-based adaptive arithmetic encoding and PB frame, which has improved performance but is complicated and left as an option. Syntax-based adaptive arithmetic coding is complex but saves 5-14%. PB frames can double the number of frames per second while allowing slightly more bits than other techniques to save bits, resulting in a much more stable and smooth video visually.

H.263, which has many factors above, is a very high performance algorithm and has been used as a reference algorithm for evaluating the performance of the proposed methods in the recent MPEG4 image quality evaluation.

1 is a block diagram of a conventional video encoder, FIG. 2 is a diagram showing motion estimation in FIG. 1, FIG. 3 is a diagram showing a relationship between a motion vector, a prediction and a prediction error macroblock in FIG. 1 illustrates zigzag scanning in FIG. 1.

Here, reference numeral 1 denotes a first combining block that adds an input video signal and a motion compensation predicted signal, and 2 denotes a DCT block that performs a discrete cosine transform (DCT) on the signal added in the first combining block 1. 3 is a quantization block for quantizing the signal of the DCT block 2, 4 is an inverse quantization block for inverse quantization of the signal quantized in the quantization block 3, and 5 is in the inverse quantization block 4 An inverse discrete cosine transform (IDCT) block for inverse discrete cosine transforming an inverse quantized signal.

Also, reference numeral 6 is a second combining block for combining the signals of the IDCT block 5 and the motion compensation prediction block 8, 7 is a motion estimation block for estimating motion in the input video signal, and 8 is the motion. A motion compensation prediction is performed on the signals combined in the second combining block 6 and output to the first and second combining blocks 1 and 6 according to the motion estimation value of the estimation block 7. Block.

In addition, reference numeral 9 is a zigzag scanning block that performs zigzag scanning on the output signal of the quantization block 3, and 10 is a variable length coding that performs variable length coding on the signal of the zigzag scanning block 9. Block 11 stores a signal of the VLC block 10, transmits the signal to the quantization block 3, and outputs a compressed video signal.

Thus, each input video frame is divided into macro blocks having a pixel area of 16 x 16 and processed. All macro blocks are used in the process of motion estimation using the recovered previous frame as a reference frame. The motion estimation is to find the area most similar to the current macro block in a search window representing a certain area of the reference frame as shown in FIG. 2. Thus, the current macro block is a movement of the macro block at the same position in the previous frame and is made under the assumption that each pixel in the macro block also moves by the same amount. The most common matching criterion used for motion estimation is a mean absolute error (MAE), which is expressed by Equation 1 below.

Where (i, j) are motion vectors [-P, p] is the range of the search window. M and N have a value of 16 as the size of the macro block, k and l represent pixel positions in the macro block, and x and y represent macro block positions in the frame. In addition, C (x + k, y + l) represents the pixel position in the current macro block in the original frame, and R (x + k + i, y + l + j) represents the pixel position in the search window in the reference frame. Thus, the area most similar to the current macro block is the area within the search window of the reference frame with minimum mean absolute error (MMAE).

The process of finding the predictive macroblock, the block most similar to the current macroblock through motion estimation, is called motion compensated prediction. The reason for this process to be performed in the decoder in the encoder is that a prediction error macroblock, which is the difference between the current macroblock and the predictive macroblock, must be coded together with the motion vector. 3 is briefly arranged to facilitate understanding of the relationship between the motion vector, the prediction macro block, and the prediction error macro block.

To encode a prediction error macroblock, the 16x16 macroblock is divided into 8x8 blocks, and then transformed into a frequency domain using DCT. DCT is used to reduce spatial redundancy of pixels and has better performance than other transforms in energy compaction, which is expressed by Equation 2 below.

Where Cm, n is a two-dimensional 8x8 DCT coefficient and m, n is the position of the coefficient in the 8x8 two-dimensional block. for m, n = 0 ego, About to be. Bi, j are pixel values in an 8x8 block before DCT, and i, j are pixel positions in the block. The pixel values of the original 8x8 block can be recovered using the following equation 3 representing 8x8 IDCT.

After DCT is performed, the DCT coefficients are quantized. The determination of the quantization value is made in consideration of the characteristic that human vision is more sensitive to low frequency region than high frequency when viewing an image. For example, slow changes in brightness or color are noticeable while fast changes are not recognized or ignored. Here, the quantization equation is as shown in Equation 4 below.

here, Is a quantized result indicating that q is quantized and m, n are the positions of the DCT coefficients in the 8x8 block. In addition, Cm, n is a DCT coefficient before quantization, and Qm, n is a quantization value.

The reason for this quantization is to produce more efficient encoding results through the effect of zeroing the remaining small DCT coefficients that are not zero and the variance of the quantized coefficients less than the variance of the original DCT coefficients. To calculate.

The quantized DCT coefficients are rearranged in 2D to 1D and in the order of low to high frequencies according to the zigzag scanning sequence in FIG. 4. Since the DCT coefficients of many high frequency components have already been quantized to zero, the rear part of this one-dimensional array consists of zero values. Therefore, after the last nonzero value, the end-of-block (EOB) symbol is used to indicate that it now has a zero value.

Next to zigzag scanning, variable length coding (VLC) is used. This is because DCT coefficients have a large bias in occurrence probability unlike before DCT. Shorter codes are assigned to higher occurrence probability values and lower probability of occurrence. The average sign length is reduced by assigning a long sign to the value.

Examining the bit rate of the compressed video after the VLC is over, it can be seen that the frame has a different bit rate, and in some cases the channel bandwidth may be exceeded to prevent this. Use a buffer to do this. When the buffer reaches its limit, more DCT coefficients become zero by jumping over the frame or increasing the quantization value to reduce the bit rate.

Previously, the previous recovery frame is used as a reference frame for motion estimation, where the previous recovery frame is generated by combining the motion compensation prediction frame and the prediction error frame recovered through inverse quantization and inverse DCT after quantization of the DCT coefficients. .

As such, in H.263 +, DCT and quantization are performed on 8x8 block motion compensated prediction error blocks, and the coefficients are rearranged using zigzag scanning.

However, according to the related art, unlike the distribution of the DCT coefficients of (a) in FIG. 6, it can be seen that the distribution of the coefficients in (b) and (c) is an arrangement not suitable for zigzag scanning. If you directly examine the distribution of DCT coefficients, you may choose the most appropriate one among various scanning methods such as vertical, horizontal, diagonal, and zigzag scanning. However, there is a problem of transmitting additional information to the decoder according to the selection of the scanning method. Will occur.

In addition, according to the related art, since the motion is mainly generated at the edge part in the video, the motion compensation prediction error also appears at the edge part. Therefore, since these errors occupy most of the data to be transmitted, there is a problem that the capacity of the transmission data increases according to the conventional technology.

Accordingly, the present invention has been proposed to solve the above-mentioned conventional problems, and an object of the present invention is to improve and scan a motion compensated prediction error block using a motion compensated prediction block in H.263 +, which is a standard video compression method. The present invention provides a high-compression apparatus and method that consider boundary boundaries of a motion compensation prediction block capable of effectively selecting a method and reducing compressed data.

1 is a block diagram of a conventional video encoder,

2 is a diagram illustrating motion estimation in FIG. 1,

3 is a diagram illustrating a relationship between a motion vector, a prediction, and a prediction error macroblock in FIG. 1,

4 is a diagram illustrating zigzag scanning in FIG. 1;

5 is a block diagram of a high compression device in consideration of the boundary direction of a motion compensation prediction block according to the present invention;

FIG. 6 is a diagram illustrating a relationship between a motion compensation prediction block, a motion compensation prediction error block, and quantized DCT coefficients in FIG. 5.

FIG. 7 is a diagram illustrating compensation for a straight edge portion of the motion compensation prediction block in FIG. 5,

8 is a flowchart illustrating a high compression method considering the boundary direction of a motion compensation prediction block according to the present invention;

9 is a flowchart illustrating a fourth step in detail in FIG. 8;

10 is a flowchart illustrating a fifth step in detail in FIG. 9;

FIG. 11 is a flowchart showing the seventh step in detail in FIG. 9;

12 is a view showing a projection of an arbitrary two-dimensional discrete function at an angle in FIG. 8,

FIG. 13 is a view illustrating angle measurement of straight edges using a pixel angle in FIG. 8;

14 is a view showing horizontal scanning and vertical scanning when scanning in FIG. 9,

FIG. 15 illustrates an example of blocks in which horizontal scanning is selected by FIG. 14;

FIG. 16 illustrates an example of blocks in which vertical scanning is selected by FIG. 14;

FIG. 17 is a diagram illustrating a linear edge value prediction to be compensated for in the motion compensation prediction block in FIG. 10.

18 is a diagram illustrating an example in which the step of FIG. 10 is applied to an actual image.

19 is a diagram comparing the experimental results according to the present invention with a table of the prior art,

FIG. 20 is a graph comparing the experimental result of FIG. 18 with the prior art. FIG.

Explanation of symbols on the main parts of the drawings

21: first coupling block 22: DCT block

23 quantization block 24 inverse quantization block

25: IDCT block 26: second combining block

27: motion estimation block 28: motion compensation prediction block

29: boundary check unit 30: compensation prediction unit

31: scan unit 32: VLC block

33: buffer block

Hereinafter, with reference to the accompanying drawings, an embodiment according to the present invention, a high-compression device considering the boundary direction of the motion compensation prediction block and the method according to the spirit of the present invention will be described.

First, in the present invention, the scanning method of DCT coefficients for the motion compensation prediction error block is selected and the prediction error value corresponding to the straight edge of the motion compensation prediction block is performed.

Therefore, in case of selecting the scanning method of DCT coefficients for the motion compensated prediction error block, in H.263 +, DCT and quantization are performed for the 8x8 block motion compensated prediction error block, and the coefficients are rearranged using zigzag scanning. 6, unlike the distribution of the DCT coefficients of (a), it can be seen that the distribution of the coefficients in (b) and (c) is not suitable for zigzag scanning. If you directly examine the distribution of DCT coefficients, you may choose the most suitable one among various scanning methods such as vertical, horizontal, diagonal, and zigzag scanning, but there is a problem of transmitting additional information to the decoder according to the selection of the scanning method. Done. Therefore, the motion compensation prediction block and the motion compensation prediction error block shown in FIG. 6 show that the main directions of the blocks are similar. Therefore, a suitable scanning method can be obtained by predicting the distribution of quantized DCT coefficients from the motion compensation prediction block using this property. Make a choice.

Also, in the case of processing the prediction error value corresponding to the straight edge of the motion compensation prediction block, since the motion mainly occurs in the edge part in the video, the motion compensation prediction error also appears in such edge part as a result. Since these errors take up most of the data being transmitted, if you can reduce them, you can reduce a significant portion of the transmitted data. In the present invention, as shown in FIG. 7, the difference between the motion compensation prediction block and the current block due to such a motion is predicted by using the value around the edge, especially when the difference occurs in the straight edge portion. By reducing the motion compensation prediction error value is proposed.

5 is a block diagram of a high compression device in consideration of the boundary direction of the motion compensation prediction block according to the present invention.

As shown here, in a video encoder including a motion compensation prediction block 28, a DCT block 22, and a quantization block 23, boundary directionality in a signal output from the motion compensation prediction block 28 is shown. A boundary check unit 29 for finding a characteristic of the circuit and obtaining a difference value based on the found boundary line; A compensation predictor (30) for obtaining a new motion compensation prediction value by compensating with the difference value obtained by the boundary checker (29); A scan unit configured to scan DCT coefficients of a signal output from the compensation predictor 30 and passed through the DCT block 22 and the quantization block 23 by using the output of the boundary line checker 29. It comprises a 31.

In the above, the boundary checker 29 obtains the angle and the pixel value, and calculates the angle of the pixel using the difference between the obtained angle and the pixel value.

In the above, the compensation predictor 30 obtains a new compensation prediction value by obtaining an angle of the straight edge using the pixel angle obtained by the difference value.

The scan unit 31 performs horizontal, vertical and zigzag scanning according to the main direction of the prediction compensation.

In addition, reference numeral 21 is a first combining block for combining the input video signal and the compensation prediction value in the compensation predicting unit 30, and 22 is a DCT block for DCT the combined signal in the first combining block 21. , 23 is a quantization block for quantizing the signal of the DCT block 22, 24 is an inverse quantization block for inverse quantization of the signal quantized in the quantization block 23, 25 is inverse in the inverse quantization block 24 An IDCT block for inverse discrete cosine transforming a quantized signal.

In addition, reference numeral 26 is a second combining block for combining the signals of the IDCT block 25 and the compensation predictor 30, 27 is a motion estimation block for estimating motion in the input video signal, and 28 is the motion. The motion compensation prediction block performs motion compensation prediction on the signal coupled by the second combining block 26 according to the motion estimation value of the estimation block 27 and outputs the motion compensation prediction block to the boundary checker 29.

Also, reference numeral 32 denotes a VLC block that performs variable length coding on the output signal of the scan unit 31, 33 stores the signal of the VLC block 32, transmits the signal to the quantization block 23, and compresses the signal. The buffer block outputs the video signal.

8 is a flowchart illustrating a high compression method in consideration of the boundary direction of a motion compensation prediction block according to the present invention.

As shown therein, a first step ST11 of finding a characteristic of boundary directionality at the output of the motion compensation prediction block 28; A second step (ST12) of obtaining a difference value around the boundary line due to the boundary directionality found in the first step; A third step (ST13) of obtaining a new compensation prediction value by compensating for the motion compensation prediction value using the difference value of the second step; And a fourth step ST14 of compressing data using the boundary directionality.

9 is a flowchart illustrating a fourth step in detail in FIG. 8.

As shown therein, a fifth step (ST21) of reducing an error value of compensation prediction and modifying an inverse quantization equation using the compensation prediction image having a straight edge by the boundary direction; A sixth step (ST22) (ST23) for performing quantization with the DCT after performing the fifth step; After the sixth step, a seventh step ST24 is performed to scan the DCT coefficients using the main direction of the boundary direction.

10 is a flowchart illustrating a fifth step in detail in FIG. 9.

As shown therein, steps (ST31) (ST32) of obtaining a main direction of the prediction compensation by using projection and then determining whether the main direction is a horizontal direction; Determining whether a straight edge exists in the prediction compensation in the horizontal direction by rotating in the horizontal direction if the main direction is not the horizontal direction (ST33); If there is a straight edge in the prediction compensation, an average of the upper and lower pixels of the straight edge is obtained, and an average value of the straight edge portion is obtained (ST35).

FIG. 11 is a flowchart illustrating a seventh step in detail in FIG. 9.

As shown therein, determining the main directions of the DCT coefficients (ST41); If the main direction of the DCT coefficients is a horizontal direction, performing vertical scanning on the DCT coefficients (ST42); If the main direction of the DCT coefficients is vertical, performing horizontal scanning on the DCT coefficients (ST43); If the main direction of the DCT coefficients is a direction other than horizontal / vertical, performing a zigzag scanning on the DCT coefficients (ST44).

The operation of the high compression device and the method considering the boundary direction of the motion compensation prediction block constructed as described above will be described in detail with reference to the accompanying drawings.

First, FIG. 6 is a diagram illustrating a relationship between a motion compensation prediction block, a motion compensation prediction error block, and quantized DCT coefficients in FIG. 5, and FIG. 7 is a diagram showing compensation for a straight edge portion of the motion compensation prediction block in FIG. 5. .

Therefore, in H.263 +, DCT and quantization are performed in the DCT block 22 and the quantization block 23 for the 8x8 block motion compensation prediction error block, and the coefficients are rearranged using zigzag scanning. 6, unlike the distribution of the DCT coefficients of (a), it can be seen that the distribution of the coefficients in (b) and (c) is not suitable for zigzag scanning.

If the direct distribution of DCT coefficients is examined, the most appropriate one can be selected among various scanning methods such as vertical, horizontal, diagonal, and zigzag scanning. However, additional information according to the selection of the scanning method should be transmitted to the decoder. Looking at the motion compensated prediction block and the motion compensated prediction error block of FIG. 6, it can be seen that the main directions of the blocks are similar. This property can be used to predict the distribution of quantized DCT coefficients from the motion compensation prediction block to select an appropriate scanning method.

The prediction error value corresponding to the straight edge of the motion compensation prediction block is processed as follows.

That is, since motion occurs mainly at the edge part in the video, motion compensation prediction error also appears at such edge part. Since these errors take up most of the data being transmitted, if you can reduce them, you can reduce a significant portion of the transmitted data. In this case, as shown in FIG. 7, when the difference between the motion compensation prediction block and the current block has occurred due to such a motion, especially in the straight edge part, the value is estimated using the value around the edge and the motion is estimated using the predicted value. The compensation prediction error value is reduced.

FIG. 12 is a diagram illustrating a projection of an arbitrary two-dimensional discrete function at an angle in FIG. 8.

Thus, we describe the circumferential search of 8 x 8 blocks using projection.

Arbitrary two-dimensional discrete function as shown in FIG. With my axis The result of integrating in parallel radial directions with the angle of is a one-dimensional discrete function. And this function angle In This is called the projection of. And, and Can be expressed as Equation 5 below.

Here, m and n represent an axis of a projection direction, and mx, my represent an axis before projection.

Since m and n have integer values and mx and my have real values, the interpolation method must be used to find the value of the interpolation method. Commonly known interpolation methods include bilinear interpolation, cubic convolution interpolation, and the like. The projection of 8x8 blocks requires considerable time because not only the pixel values for the positions beyond the boundary are required, but also the multiple projections with changing the values.

Therefore, the angle of new straight edge is checked by using the pixel value of the motion compensation prediction block rather than the above projection method. The specific method is explained in FIG.

FIG. 13 is a view illustrating angle measurement of a straight edge using the pixel angle in FIG. 8.

Therefore, dx, dy, angle, and weight are obtained as in Equation 6 below.

dx = (j, i + 1)-(i, j)

dy = (j + 1, i)-(i, j)

angle = arctan (dx./ dy)

weight = | dx | + | dy |

In this way, when the weight is obtained from the angle and the pixel value, the weight is large because the difference of the pixel value is large near the edge. In the case of a block of straight edges, pixels with a large weight all have a similar angle, and thus an angle similar to the angle obtained by the projection can be obtained. By calculating the angle of the straight edge using the angle of the pixel in this way, the computation amount can be considerably reduced, which leads to a much shorter program execution time.

Meanwhile, a scanning method of DCT coefficients using the main direction of the motion compensation prediction block will be described.

First, FIG. 14 is a diagram illustrating horizontal scanning and vertical scanning when scanning in FIG. 9, FIG. 15 is a view illustrating examples of blocks in which horizontal scanning is selected by FIG. 14, and FIG. 16 is a block in which vertical scanning is selected by FIG. 14. The figure shows an example.

The first method proposed here obtains the main direction of the motion compensated prediction block that is already decoded by the block angle and assumes that it is the main direction of the motion compensation prediction error block to be decoded accordingly. To determine the appropriate scanning method.

This way you don't have to send any information to the decoder, but it could be misleading. In FIG. 6, since the motion compensation prediction block and the motion compensation prediction error block have similar directionality, it may not be considered that the motion between the previous frame and the current frame is considerably large or the scene that does not exist in the previous frame is not present.

So the scanning method added by the present invention is horizontal and vertical scanning. When the main direction obtained from the motion compensation prediction block is the horizontal direction as shown in FIG. 6 (b), the vertical scanning method is selected since the DCT coefficients show the vertical distribution, and as shown in (c) of FIG. 6, the main axis is the vertical direction. In this case, the DCT coefficients represent a horizontal distribution, so the horizontal scanning method is selected. As shown in Fig. 6 (a), the conventional zigzag scanning is used for directions other than horizontal and vertical. Vertical and horizontal scanning methods are shown in FIG. 14 and actual examples are shown in FIGS. 15 and 16.

On the other hand, the reduction of the motion compensation prediction error value and the deformation of the inverse quantization equation using the motion compensated prediction image with the straight edge are as follows.

FIG. 17 is a diagram illustrating a straight edge value prediction to be compensated for in the motion compensation prediction block in FIG. 10. Thus, when the motion compensation prediction block has a straight edge, the error value of the straight edge portion is predicted using the values around the straight edges.

So, we use the projection (pixel angle) described in the calculation of the block angle to find the main direction of the motion compensation prediction block, and if it is not the horizontal direction, rotate the horizontal direction to make it easier to detect the straight edge and use the value around the straight edge. Let's do it.

Then, it is checked whether there is a straight edge in the rotated motion compensation prediction block.

If there is a straight edge, the average of the upper pixels and the lower pixels of the straight edge is averaged, and the values are A and B, respectively.

The prediction value of the straight edge portion is obtained as in Equation 7 below.

Where QP is the quantization value. If the QP is less than or equal to 10, since the blocks with small motion compensation prediction error values increase, the predicted straight edge value needs to be small.

In addition, when the QP is greater than or equal to 10, since the blocks having a large value of the motion compensation prediction error value increase, it is necessary to increase the predicted straight edge value.

In addition, the predicted straight edge value is signed, which is determined by comparison with the current block, and one bit is allocated and passed to the decoder.

After the new motion compensated prediction error block is obtained using the motion compensated prediction block with linear edge compensation, DCT, quantization, and VLC are performed.

FIG. 18 is a diagram illustrating an example in which the step of FIG. 10 is applied to an actual image. When the above process is implemented with the actual image, the following procedure is performed.

If the above method is used, there is a possibility that the predicted value of the straight edge part may be in error, but by modifying the inverse quantization equation used in H.263 +, the bit rate is reduced and the PSNR (Peak Signal to Noise Ratio) is kept low. Could be achieved. The quantization equation and the inverse quantization equation used are as follows and the value of p was modified from 1 to 0.5.

Quantization: | LEVEL | = (| COF | -QP / 2) / (2QP)

Inverse Quantization: | REC | = QP · (2 · | LEVEL | + p) if QP = "odd"

| REC | = QP · (2 · | LEVEL | + p) -1 if QP = "even"

Where LEVEL is the quantized DCT coefficient, COF is the DCT coefficient to be quantized, and REC is the recovered DCT coefficient.

19 is a diagram comparing the experimental results according to the present invention with a table of the prior art, and FIG. 20 is a graph comparing the experimental results of FIG. 18 with the prior art.

In this experiment, five images of QCIF format carphone, claire, foreman, suzie, and trevor with 176 x 144 resolution were used, and the quantization values were 4, 5, 7, 10, 15, and 25.

The total number of frames used is 300 frames for each image, and the frame rate is 10 fps because it skips 2 frames. The average peak signal-to-noise ratio (PSNR) and average bit rate are used for performance measurement.

In the result table, since the bit rate is good, the case where result is "-" is improved, and the case of PSNR is case where result is "+".

Thus, as shown, it can be seen that the bit rate of about 3% is improved in the same PSNR on average.

In the graph, proposed scheme 1 + 2 means the simulation result of the sum of the compensation of the prediction block and the change of the scanning method.

As described above, the present invention improves the motion compensation prediction error block by using the motion compensation prediction block in H.263 +, which is a standard video compression method, and reduces the compressed data by effectively selecting the scanning method.

Although the preferred embodiment of the present invention has been described above, the present invention may use various changes, modifications, and equivalents. It is clear that the present invention can be applied in the same manner by appropriately modifying the above embodiments. Accordingly, the above description does not limit the scope of the invention as defined by the limitations of the following claims.

As described above, the high compression device and its method considering the boundary direction of the motion compensation prediction block according to the present invention improve the motion compensation prediction error block by using the motion compensation prediction block in H.263 + which is a standard video compression method. By effectively selecting the scanning method to reduce the compressed data, it can be used as one of the standard image compression methods, or it can be used to increase the video compression efficiency by adopting its own mode.

In addition, the present invention can be usefully used for wired and wireless video communication or video storage, and can provide a good image quality at a low bit rate, so that in the case of an office surveillance system, it is possible to achieve an efficient reduction of storage space. do.

Claims (8)

In a video encoder comprising a motion compensation prediction block, a DCT block and a quantization block, A boundary line checker which finds a property of boundary directionality in the signal output from the motion compensation prediction block and obtains a difference value based on the found boundary line; A compensation prediction unit obtaining a new motion compensation prediction value by compensating with the difference value obtained by the boundary checker; And a scan unit configured to scan DCT coefficients of the signal output from the compensation predictor and passed through the DCT block and the quantization block by using the output of the boundary checker. High compression device considering directionality. The method of claim 1, wherein the boundary checker, A high-compression device considering boundary direction of a motion compensation prediction block, wherein the angle and the pixel value are obtained and the angle of the pixel is obtained using the difference between the obtained angle and the pixel value. The method of claim 1, wherein the compensation prediction unit, A high-compression device considering boundary direction of a motion compensation prediction block, wherein a new compensation prediction value is obtained by obtaining an angle of a straight edge with an angle of a pixel obtained by a difference value. The method of claim 1, wherein the scan unit, A high-compression device considering boundary direction of a motion compensation prediction block, wherein horizontal, vertical, and zigzag scanning is performed according to the main direction of the prediction compensation. Finding a characteristic of the boundary directionality at the output of the motion compensation prediction block; A second step of obtaining a difference value around the boundary line due to the boundary directionality found in the first step; Obtaining a new compensation prediction value by compensating for the motion compensation prediction value using the difference value of the second step; And a fourth step of compressing the data using the boundary directionality. The method of claim 5, wherein the fourth step, A fifth step of reducing an error value of compensation prediction and modifying an inverse quantization equation by using the compensation prediction image having a straight edge by the boundary directionality; A sixth step of performing quantization with the DCT after performing the fifth step; And a seventh step of scanning the DCT coefficients by using the principal direction of the boundary directionality after the sixth step. The method of claim 6, wherein the fifth step, Obtaining a main direction of the prediction compensation by using the projection and then determining whether the main direction is a horizontal direction; If the main direction is not the horizontal direction, rotating in the horizontal direction to determine whether there is a straight edge in the prediction compensation in the horizontal direction; If the prediction compensation has a straight edge, calculating the average of the upper and lower pixels of the straight edge and calculating the average value of the straight edge portion; . The method of claim 6, wherein the seventh step, Determining a principal direction of the DCT coefficients; If the main direction of the DCT coefficients is a horizontal direction, performing vertical scanning on the DCT coefficients; If the main direction of the DCT coefficients is vertical, performing horizontal scanning on the DCT coefficients; If the main direction of the DCT coefficients is a direction other than horizontal / vertical, performing zigzag scanning on the DCT coefficients.