KR100295327B1 - Motion Estimation System Using Adaptive Threshold - Google Patents

Abstract

The present invention provides a video encoding apparatus for estimating motion and encoding an image according to an estimated motion vector, and in particular, a motion for enabling motion estimation with an adaptive threshold value according to a quantization interval. It relates to an estimation device.

Through block matching, a motion vector (MV) having a minimum matching error (SAD) is obtained and motion compensation is performed on an image. In the ultra low bit rate coding having a large quantization interval, DCT transforms a difference image and performs quantization. In this case, since all the coefficients become zero, a lot of cases occur. In this case, a better match in motion estimation does not lead to an improvement in image quality or a decrease in bit rate. .

Accordingly, in the present invention, the present invention sets the SAD for the DCT coefficient that becomes 0 after quantization to a threshold value (Th = 10QP), and when the SAD smaller than the threshold value is found, the search for more matching blocks is terminated. In this way, the motion estimation can be performed adaptively to the quantization interval QP, so that the computation amount can be reduced without deteriorating the image quality due to mismatch or reducing the coding efficiency.

Description

Motion Estimation System Using Adaptive Threshold

The present invention provides a video encoding apparatus for estimating motion and encoding an image according to an estimated motion vector, and in particular, a motion for enabling motion estimation with an adaptive threshold value according to a quantization interval. It relates to an estimation device.

H.261, H.263, MPEG1, MPEG2, and MPEG4, which are international standards for general video encoding, compress video signals by combining motion compensation and Discrete Cosine Transform (DCT).

Such image encoding removes the redundancy in the image sequence through the following two processes.

In the first process, a motion vector is estimated by using block matching in two adjacent temporal images, and a motion compensation image is generated by using the same, and the difference between the two images is compared with the previous image. To eliminate the temporal redundancy.

In the second process, spatial redundancy is eliminated by DCT transforming the difference image and performing quantization.

In this case, as the bit rate allocated to the video encoding is lowered, quantization must be performed at a larger interval, so that the amount of bits allocated to the difference video becomes smaller, and therefore, the importance of motion estimation and compensation becomes larger.

Therefore, efficient motion estimation is required for video coding such as H.263 and MPEG4, which are oriented to ultra low bit rates.

FIG. 1 is a block diagram illustrating a structure of a conventional conventional video encoding apparatus as described above. Referring to FIG. 1, the operation thereof will be described below.

A subtractor 2 for storing the original image in the field or frame unit in the image memory 1, obtaining a difference signal between the memory image and the reconstructed image, and a DCT converter 3 for performing DCT conversion of the difference signal. .

The DCT transformed result (DCT coefficient) is quantized by the quantizer 4, and the quantized signal is inversely quantized by the inverse quantizer 5 (Q ⁻¹ ) and inversely DCT-converted through the IDCT converter 6. Get the image.

That is, the current input image is restored.

This signal is added to the motion-compensated signal in the adder 7 and stored in the memory 8, and the image information stored in the memory 8 is in accordance with the motion vector MV estimated by the motion estimation 9. The motion compensation in the motion compensation unit 10 is passed back to the subtractor (2) and the adder (7),

The motion estimation unit 9 predicts the motion of the next image to be encoded, which is output from the image memory 1, and performs block matching between the current image and the image to be predicted. The best matched motion vector MV is obtained, and the motion compensator 10 outputs a motion-compensated, reconstructed image according to the motion vector MV obtained as described above.

On the other hand, the variable length coding unit (VLC) 11 performs variable length encoding on the quantized controlled final information and is input to the buffer 12 together with the motion vector information to the decoder (decoder) in the form of bitstream data. Is sent.

Here, the bit rate controller 13 controls the quantization rate of the quantizer 4 from information in consideration of the capacity of the output buffer (target memory) 12.

In the motion estimation using block matching, a sum of absolute difference (SAD) is mainly used as a matching error, and a motion vector (MV) is determined at the lowest position.

A conventional motion estimation technique is a full search method.

This technique finds the best matching position using all candidate positions in the search area.

However, this technique can find the most accurate motion vector, but it requires too much computation for real-time processing.

In order to compensate for the drawbacks, a three step search method and a hierarchical search have been published to reduce the complexity of motion estimation.

In addition, a method of setting a threshold and determining that the matching error is sufficiently small when the SAD is less than the threshold is also used.

Such an ultimate purpose of image coding is to maximize the image quality of the encoded image within the limits of the limited bit rate and the computation amount, and not to reduce the matching error.

For example, in ultra-low bit rate coding with a large quantization interval, when DCT transforms a difference image and quantizes, all coefficients in a block become 0. In this case, better matching in motion estimation is achieved. There is no need to find the best matching block with the lowest SDA since it does not lead to improvement of the data rate or reduction of the bit rate.

In the present invention, by using a property as described above, by setting a threshold for adaptively determining whether to perform the motion estimation according to the quantization interval, it is possible to perform the motion estimation, thereby reducing the amount of computation without reducing the coding efficiency It is to propose a motion estimation method.

1 is a block diagram illustrating a general video encoding apparatus.

2 is a block diagram illustrating a video encoding apparatus to which a motion estimation apparatus using an adaptive threshold according to the present invention is applied.

Figure 3 is a detailed block diagram of a motion estimation means to which the present invention is applied.

According to the present invention, an SAD threshold is adaptively set according to a changing quantization interval, and the SAD search is terminated when the SAD value is smaller than the set SAD threshold by comparing the set SAD threshold with the SAD value.

The configuration of the device of the present invention for implementing this,

A video encoding apparatus for performing motion estimation through block matching and compensating an image through motion estimation,

Block matching means for obtaining an image matching error through block matching and estimating a motion vector that minimizes the matching error;

Threshold setting means for adaptively setting the SAD threshold according to the quantization interval obtained from the quantization interval control means;

And a comparison means for controlling motion vector estimation execution by comparing the SAD threshold set by the threshold setting means with the matching error obtained by the block matching means.

The configuration of the motion estimation apparatus using the adaptive threshold value of the present invention having such a configuration will be described with reference to FIGS. 2 and 3.

FIG. 2 is a block diagram showing an embodiment of the apparatus of the present invention with the same reference numerals as in FIG. 1, and FIG. 3 is a block diagram showing the detailed configuration of the motion estimation in FIG. In the following description, redundant parts are omitted.

A video encoding apparatus for compressing / encoding an image by predicting a motion of a currently input image through a motion compensated image through block matching between a current input image and a previous image,

With the configuration of the motion estimation unit 9 ',

Estimating a motion vector (MV) that minimizes SAD by matching blocks of the current input image from the image memory 1 with blocks of the previous frame from the memory 8 Block matching unit 100,

A threshold setting unit 200 for adaptively setting the SAD threshold Th (10QP) according to the quantization interval QP output from the bit rate controller 13,

The SAD threshold Th set by the threshold setting unit 200 is compared with the SAD obtained from the block matching unit 100, and when the obtained SAD is larger than the SAD threshold Th, the estimation of the continuous motion vector MV is performed. If the obtained SAD is less than the SAD threshold Th, and the comparison unit 300 to stop the estimation of the motion vector (MV) and to select the motion vector (MV) corresponding to the current SAD is configured. .

The motion estimation apparatus of the present invention having such a configuration determines whether to terminate the search for the SAD obtained for motion estimation according to the quantization interval QP that is adaptively changed according to the bit rate of the buffer 12. It is characterized by setting the reference value, that is, the SAD threshold Th,

The block matching unit 100 receives an image frame currently input from the image memory 1 and a previous image frame from the memory 8 and matches these blocks, while the motion vector MV at which the SAD value is minimum is obtained. ),

In the process of searching for SAD as described above, the comparator 300 compares the SAD threshold Th output from the threshold setting unit 200 adaptively according to the SAD obtained during the search and the quantization interval QP. do.

When the SAD becomes smaller than the SAD threshold Th during the comparison process, the SAD search is no longer performed, and the motion vector MV corresponding to the SAD value at that time is obtained by the motion compensator 10 and the variable length coding unit. The output is (11).

At this time, the comparison unit 300 compares the SAD threshold value (Th) 10QP output from the threshold setting unit 200 and the SAD obtained from the block matching unit 100,

As such, the threshold 10QP as the reference value obtained through the threshold setting unit 200 is obtained through the following process.

As mentioned above for the conventional motion estimation, since the SAD is used as a measure of the motion estimation error, when the motion is estimated for the 8 × 8 block, the equation of SAD can be expressed as Equation 1 below.

The current image created from the previous image using the motion vectors (mvx, mvy) obtained after the motion estimation is called a motion compensation image, and the difference between the motion compensation image and the original current image is called a difference image.

The difference image f '(i, j) at this time may be represented by Equation 2 below.

f ′ (i, j) = f (i, j, t) -f (i-mvx, j-mvy, t-1)

The image encoder obtains a DCT coefficient by performing DCT on the difference image, quantizes it with a quantization interval (QP), and performs variable length encoding. The DCT number of the difference image is expressed as in Equation 3 below.

with u, v, i, j = 0,1,2, ......, 7

where i, j = spatial coordinates in the pixel region,

(spatial coordinates in the pixel domain)

u, v = coordinates in the transform domain,

(coordinates in the transform domain)

for u = 0, otherwise 1,

for v = 0, otherwise 1.

When the absolute value is taken at both sides in Equation 3, an inequality as in Equation 4 below is established.

At this time, in Equation 4 Since can be expressed as in Equation 5 below,

According to Equation 5, Equation 4 may be expressed as Equation 6 below.

An equation for quantizing such DCT coefficients in an inter macro block is shown in Equation 7 below.

As shown in Equation 7, the inequality as in Equation 8 below must be satisfied in order for the level of quantizing the DCT coefficient to be zero.

Solving Equation 8, the result of Equation 8 is expressed as Equation 9 below.

If the result is applied to Equation 6, it is expressed as Equation 10 below.

Here, when C (u) and C (v) become 1, the minimum value can be obtained, and an inequality such as the following Equation 11 can be obtained.

SAD (mvx, mvy) <10QP

As shown in Equation 11, when the matching error SAD is less than 10 QP, the DCT coefficients are all zero when all DCT coefficients in the block are quantized.

Therefore, in the present invention, when the SAD is less than 10QP, all coefficients of the DCT become zero after quantization, so that even if a SAD smaller than this value is not obtained, the degradation of the image quality or the variation of the bit amount does not occur. Therefore, if the threshold value is set to 10QP and smaller than this, no further motion estimation is performed.

In the above example, the motion estimation is performed using 8 × 8 blocks. However, in the case of motion estimation using 16 × 16 blocks, SAD is divided into 8 × 8 blocks for each block as described above. It is sufficient to determine whether or not the reference SAD threshold is satisfied to execute search end control of the SAD.

As described above, the present invention is applicable to various image coding algorithms for performing motion estimation / compensation through block matching.

As described above, the control of the SAD is controlled by adaptively setting a reference value according to the quantization interval (QP) that is changed for bit rate control, thereby reducing the image quality and reducing the coding efficiency. The amount of calculation can be reduced.

Claims (3)

A video encoding apparatus for performing motion estimation through block matching and compensating an image through motion estimation, Block matching means for obtaining an image matching error through block matching and estimating a motion vector that minimizes the matching error; Threshold setting means for adaptively setting the SAD threshold according to the quantization interval obtained from the quantization interval control means; And a comparing means for comparing the matching error threshold set by the threshold setting means with the matching error obtained by the block matching means to control the motion vector estimation execution. 2. The method of claim 1, wherein when performing motion estimation through 8x8 block matching, the matching error threshold set by the threshold setting means is a relationship between matching errors such that all DCT coefficients are zero after quantization. SAD (mvx, mvy) <10QP Set to 10QP according to the quantization interval (QP) using And the comparison means stops the motion vector search if a matching error smaller than the set 10QP appears. The method of claim 2, wherein when the motion estimation unit estimates the motion through matching of 8x8 blocks or more, the comparing unit divides the blocks by 8x8 units and compares the matching error and the threshold value for each block to determine the threshold value for each block. Motion estimation apparatus using an adaptive threshold characterized in that to find a smaller matching error.