JPH1175203A - Motion vector detector - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a motion vector detector by which a motion vector of a wavelet transformed video signal is detected through pattern classification. SOLUTION: The motion vector detector includes a current wavelet division section 410 that applies wavelet transformation to a current frame to generate M-sets of hierarchical current video images, a preceding wavelet division section 430 that applies wavelet transformation to a preceding frame to generate M-sets of hierarchical current video images, a retrieval block pattern classification section that classifies a retrieval block at a lowermost layer into any of prescribed patterns, an object block pattern classification section that classifies each object block corresponding to the retrieved block into any of prescribed patterns, and a motion estimate section 420 that retrieves an optimum object block based on the patterns to detect a displacement vector and obtains a displacement vector with respect to the retrieval block at the highermost layer based on the displacement vector so as to decide it as a motion vector.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a motion vector estimating apparatus, and more particularly, to a motion vector estimating apparatus for detecting a motion vector of a wavelet-transformed video signal by pattern classification.

[0002]

2. Description of the Related Art Discrete wavelet transform (discrete wa
The velet transform (DWT) technique has recently received considerable interest in the field of video processing due to its flexibility for representing unstable video signals and its properties for human visual characteristics. Wavelet representations provide a multi-resolution / multi-frequency representation of a video signal expressed as a function of time and frequency.

[0003] Such versatility is useful in video and video encoding. Because natural video and video have unstable properties, the wavelet transform splits the unstable signal into multi-scaled wavelets so that each component is relatively stable and suitable for coding. Also,
Since a variety of coding methods and parameters are applicable to the statistical properties of each wavelet, coding each stable component is more effective than coding a totally unstable signal. In addition, the wavelet representation is well tied to the spatially and frequency-modulated nature of human vision as reported in psychophysical and physiological studies.

In a typical wavelet segmentation technique, for example, as disclosed in US Pat. No. 5,477,272 granted to Ya-Qin Zhang on Dec. 19, 1995, video frames have different resolutions. It is divided into a plurality of layers, and each sub-image in the same layer corresponds to a different frequency band. FIG. 1 is a schematic diagram illustrating a normal wavelet division process. The current frame S ₁ is input to the first wavelet splitting unit 110, and each sub-image of layer 1 is
That is split into ^{_{S 2, W 2 1, W}} 2 2, W 2. afterwards,
Sub video S ₂ is input to the second wavelet decomposition unit 120, each sub-picture of the hierarchy 2, ^{_{i.e., S 4, W 4 1,}} W 4 2, W
Divided into ₄ ³ Subsequently, the sub-picture S ₄ is inputted to the third wavelet decomposition unit 130, each sub-picture of the hierarchy 2, i.e., is divided into ^{_{S 8, W 8 1, W}} 8 2, W 8 3.

[0005] These sub-pictures have a pyramid structure.
Thus, an image representation as shown in FIG. 2 is provided. Wavy
Current frame S converted₁Is two levels of resolution depth
And three sub-pictures and one row in each layer
It consists of 10 sub-images consisting of pass sub-images.
Sub-picture S_FourIs the sub-picture S₈And sub-screens in hierarchy 3
Statue W₈ ¹, W₈ ^Two, W₈ ^ThreeFormed in combination with
Video S_TwoIs the sub-picture S_FourAnd sub-pictures in layer 2
W_Four ¹, W_Four ^Two, W_Four ^ThreeFormed by the combination of
Laem S₁Is the sub-picture S_TwoAnd sub-pictures in layer 1
W_Two ¹, W_Two ^Two, W _Two ^ThreeIs formed in combination with

FIG. 3 shows a conventional multi-resolution motion estimation.
FIG. 3 is a schematic diagram illustrating a resolution motion estimation (MRME) method. First, a plurality of sub-images S ₈ the current frame S ₁ is ^{_{divided, W 8 1, W 8 2}} , W 8 3, W 4 1, W 4 2, W 3,
^{_{W 2 1, W 2 2,}} W 2 3 are generated, also the previous frame PS ₁ is divided a plurality of sub-images ^{_{PS 8, PW 8 1, PW}} 8 2, P
^{_{W 8 3, PW 4 1,}} PW 4 2, PW 4 3, PW 2 1, PW 2 2, PW
₂ ³ is generated. Here, the previous frame PS ₁ and the corresponding sub image ^{_{PS 8, PW 8 1, PW}} 8 2, PW 8 3, PW 4 1, P
^{_{W 4 2, PW 4 3,}} PW 2 1, PW 2 2, PW 2 3 are omitted for convenience of explanation.

Thereafter, each sub-picture in the current frame S ₁ is divided into a plurality of search blocks. The size of each search block in the sub-video belonging to the same hierarchy is the same. If the size of the search block in the sub-picture belonging to the highest hierarchy M is P × P, the size of the search block in the sub-picture belonging to the hierarchy m is P · 2 ^Mm × P
・ 2 ^Mm . Here, M, P and m are each a positive integer, and the values of M and P are typically 3 and 2, respectively.

[0008] Thereafter, each search block in each sub-picture is motion estimated based prior to the relevant sub-picture frames PS _1. For example, in FIG. 3, it is assumed that the search block 302 in the sub-image S ₈ is motion-estimated by a normal block matching algorithm. In this case, a search area corresponding to the search block 302 in the sub-image S ₈ is formed in the previous sub-image PS ₈ , and a plurality of candidate blocks are generated in the corresponding search area. Then, error values between the search block 302 and the candidate block in the sub-picture S ₈ is calculated. This error value is, for example, an average absolute error between the corresponding pixel values of the candidate block of the search block 302 in the sub-image S _8.

[0009] Of the calculated error value, the minimum error value is selected, the search in the sub-picture S ₈ block 302
And the difference between the best candidate block 304 that yields the smallest error value is detected. This difference value is determined as the motion vector MVS ₈ of the search block 302 in the sub-picture S ₈ .

This motion vector MVS₈Based on
Video W₈ ¹Search area corresponding to search block 306 in
Area is sub-picture PW₈ ¹The search block is formed by
Motion estimation for 306 is performed. To elaborate,
Image PW₈ ¹At the sub-picture W₈ ¹Search block in
The same position as the position 306 is detected, and the detected position is
Kutor MVS₈Is only displaced. After that, the sub video PW
₈ ¹The search area at is formed around the displacement position and
Video W₈ ¹The search block 306 in the sub-image S₈In
Motion estimation using the same method as the search block 302
Thus, the best candidate block 308 is
Is detected. Then, the sub video W₈ ^Two, W ₈ ^ThreeEach search in
The sub block W₈ ¹Search block 306 in
The motion is estimated in the same manner as

[0011] Based on the extended motion vector 2 MVS ₈ ,
By searching the area corresponding to the search block 310 in the sub-picture W ₄ ¹ are formed in the sub-picture PW ₄ ^1, motion estimation is performed for the search block 310. That is, in the sub image PW ₄ ^1, the same position and the position of the search block 310 in the sub-picture W ₄ ¹ is detected, the detection position is displaced by the motion vector 2MVS _8. Then, the search area in the sub video PW ₄ ¹ is formed around the displaced position,
Sub video W ₄ search block 310 is a sub video W ₈ ¹ in ¹
The optimal candidate block 312 is detected in the search area by performing motion estimation in the same manner as the search block 306 in. Then, even the search block in the sub image W ₄ ^2, W ₄ ^3, the search block in the sub-picture W ₄ ¹
Motion estimation is performed in the same way as 310.

[0012] Based on the extended motion vector 4MVS ₈ ,
By searching the area corresponding to the search block 314 in the sub-picture W ₂ ¹ is formed by the sub-picture PW ₂ ^1, motion estimation is performed for the search block 314. That is, in the sub image PW ₂ ^1, the same position and the position of the search block 314 in the sub-picture W ₂ ¹ is detected, the detection position is displaced by the motion vector 4MVS _8. Then, the search area in the sub video PW ₂ ¹ is formed around the displaced position,
Sub video W ₂ search block 314 in ^one sub-image W ₄ ¹
By performing motion estimation in the same manner as in the search block 310, the optimal candidate block 316 is detected in the search area. Then, even the search block in the sub image W _{² 2,} W ₂ ^3, the search block in the sub-picture W ₂ ¹
Motion estimation is performed in the same manner as in 314.

[0013] On the other hand, the search block in the sub-image W ₂ ¹
If the best candidate block 316 corresponding to 314 is detected as shown in FIG. 3, the search in the sub image W ₂ ¹ Block 31
Displacement from 4 to the best candidate block 316 is MVW ₂ ^1. Then, it is computed difference between the 4MVS ₈ and MVW ₂ ^1, is outputted as a motion vector differential MVDW ₂ ¹ of the search block 314 in the sub-picture W ₂ ^1.

In such an MRME method, since motion vectors for all sub-pictures are detected and transmitted,
The calculation process is complicated. For this reason, a motion estimation method applied to only the sub-video corresponding to the low frequency band has been proposed as shown in FIG. This new technique is based on the sub-picture S
₈ but it is only a size of about 1/64 the current frame S _1, includes a substantially large portion of the total energy present in the current frame S _1, resulting in a low frequency band than human vision high frequency band Based on the concept of being more sensitive to errors.

According to this technique, the sub-picture S₈In
Between the search block 318 and the optimal candidate block 320
Motion vector MV₈Is doubled and the sub-picture S_FourIn
Used as the initial motion vector for the search block 322
Motion vector MV_FourAnd motion vector difference MVD_Four
Will be detected. This motion vector difference MVD _Four
Is 2MV₈And MV_FourIs the difference between In addition,
Image S_FourSearch block 322 and the best candidate block
Motion vector MV between_FourIs doubled and the
Video S_TwoMotion vector of search block 326 in
Motion vector MV_TwoAnd motion vector
MVD_TwoWill be detected. This movement vector
MVD_TwoIs 2MV_FourAnd MV_TwoIs the difference between
You. Also, the sub video S_TwoAt search block 326 and
Motion vector MV between this candidate block 328 and_TwoIs twice
Extended, sub-picture S₁First of search block 330 in
Motion vector MV used as the initial motion vector₁Passing
And motion vector difference MVD₁Will be detected. This
Motion vector MV₁Is the search block 330 and the
This is the displacement between the lock 332 and the motion vector difference MV.
D₁Is 2MV_TwoAnd MV ₁Is the difference between

As described above, by estimating the motion of only the sub-picture corresponding to the low frequency band, the calculation process of the motion estimation can be further simplified. However, in this technique, since the motion vector of the search block in the sub-image S ₈ is detected and all of the candidate blocks that are generated by the search block corresponding search area by block matching, more simplified calculation process There is a need to.

[0017]

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a motion vector detecting device for detecting a motion vector of a wavelet transformed video signal by pattern classification.

[0018]

According to the present invention, there is provided a motion vector detecting apparatus for detecting a motion vector between a current frame represented by a video signal and a reference frame. A first wavelet transform unit for performing a wavelet transform on the current frame to generate M hierarchical current images, wherein a current image in a highest layer corresponds to the current frame and a current image in a higher layer Has a higher resolution than the current video in the lower hierarchy, each current video has the same number of search blocks,
Each of the search blocks in the lower layer corresponds to each of the search blocks in the higher layer, and the first wavelet transform means and the reference frame are wavelet transformed to generate M hierarchical reference images. A second wavelet transform means, wherein the reference image in the highest layer corresponds to the reference frame, the reference sub-image in one layer has a plurality of candidate blocks corresponding to each search block in the same layer, The size of each search block and the corresponding candidate block in the same layer is the same as each other. The second wavelet transform unit and the second classifier that classifies the search block in the lowest layer into one of predetermined patterns. 1 pattern classification means and each candidate block corresponding to the search block A second pattern classifying unit that classifies the search block and the optimum block based on the pattern related to the search block and the candidate block. An optimal candidate block searching means for detecting a displacement vector with respect to the search block, representing a displacement between the candidate blocks, and a displacement vector of the search block in the highest hierarchy based on the detected displacement vector; Is determined as a motion vector for the search block in the uppermost hierarchy.

[0019]

Preferred embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 5 is a schematic block diagram of a motion vector detecting device 400 according to the present invention. As shown in FIG. 5, a current frame S ₁ is input to a current wavelet dividing unit 410 and a motion estimating unit 420 via a line L402. This current wavelet division
410 As is shown in FIGS. 1 and 2, by dividing the current frame S ₁ into a plurality of sub-images, current sub picture in the low frequency band (i.e., S _2, S ₄ and S ₈₎ each line L412, L
414 and L416 to the motion estimator 420.

On the other hand, the memory 490 sets the output as the previous frame PS ₁ via the line L492 and outputs the motion estimator 420,
It is supplied to the previous wavelet dividing section 430 and the motion compensating section 440. The front wavelet dividing unit 430 supplies the low frequency band front sub-images (ie, PS ₂ , PS ₄ and PS ₈ ) to the motion estimating unit 420 via lines L432, L434 and L436, respectively. Each of these sub-pictures PS ₂ , PS ₄ and PS
₈ is obtained by dividing a frame PS ₁ before it is supplied from the memory 490.

The motion estimator 420 detects a motion vector for each search block in the current frame S ₁ and supplies the motion vector to the motion compensator 440 and a transmitter (not shown). FIG.
FIGS. 7 and 8 are schematic diagrams for explaining the motion estimation process of the motion estimation unit 420 in detail. The motion compensation unit 440 receives the motion vector supplied from the motion estimation unit 420, and a corresponding best candidate block in the frame PS ₁ prior to being fed via a line L492 from the memory 490. This motion compensator
The 440 performs motion compensation on the optimal candidate block using the received motion vector, and supplies the motion-compensated optimal candidate block to the subtractor 450 and the adder 480.

The subtraction unit 450 subtracts the motion compensation optimum candidate block from the search block in the current frame S _{1 and} converts the error signal into a two-dimensional discrete cosine transform (DCT) & quantization.
(Q) Supply to section 460. The DCT & Q section 460 performs discrete cosine transform and quantization on the received error signal, and performs variable length coding (VL
C) Supply to the section 465 and the IDCT & IQ section 470. This V
LC section 465 is the quantization D received from DCT & Q section 460.
Variable length coding is performed on the set of CT coefficients, and a VLC signal is sent to the transmitter. IDCT & IQ unit 470 is DCT
An inverse discrete cosine transform and inverse quantization are performed on the set of quantized DCT coefficients from & Q section 460, and a restoration error signal is supplied to addition section 480. This adder 480 is an IDCT
& Restoration error signal received from the IQ section 470 and the motion compensation best candidate blocks received from the motion compensation unit 440 adds, to generate the reconstructed search blocks in the current frame S _1. The reconstruction search block in the current frame S ₁ is stored in the memory 490 as a previous frame for the next frame.
Is stored in

FIG. 6 is a detailed block diagram of the motion estimator 420 in FIG. The pattern block matching unit 422 receives the sub-image S ₈ through the line L416 and the sub-image PS ₈ through the line L436, and estimates the motion of each search block in the sub-image S ₈ by classifying the pattern of the pixel value distribution. Details of the motion estimation process of the pattern block matching unit 422 will be described with reference to FIG. Search block formation unit 422-1 receives the sub image S ₈ from the line L416, each predetermined size P × P (e.g., 2 ×
It is divided into a plurality of search blocks consisting of two pixels) and supplied to a search block pattern classification unit 422-2.

This search block pattern classification unit 422-2
As shown in FIG. 8, classifies the pattern of each search block in the sub-images S _8. The size of the sub-image S ₈ is
Since there are 12 × 8 pixels and the size of the search block in the sub-picture S ₈ is 2 × 2 pixels, the number of search blocks is 24. For example, one search block 710 in the sub-picture S ₈ has four pixels, the values of the upper left pixel is 92, a value of the upper right pixel 124, the value of the lower left pixel 115 And the value of the lower right pixel is 109. Then, the search block pattern classification unit 422-2 compares the four pixel values with each other and assigns a priority to each pixel. Thereafter, the search block pattern classifying unit 422-2, the search block 710 in the search block 720 and the sub-picture S ₈ in the sub-image S ₈ assigned priority generates a pattern belongs. in this case,
The priority value of the search block 720 is, as shown in FIG.
Pixels 1, 2, 3, and 4 are assigned in the order of upper left pixel, lower right pixel, lower left pixel, and upper right pixel.
If two or more pixels have the same pixel value, the pixels are assigned the same priority.

Thereafter, information and pattern information for each search block in the sub-picture S ₈ are input to the first block matching unit 422-6. Here, the information search block represents the pixel value of each search block in the sub-picture S _8, the pattern information indicates whether each search block belongs to some pattern.

On the other hand, the candidate block forming section 422-3 in FIG.
Is the sub-picture PS via line L436₈Receive
And sub video PS₈Form a plurality of candidate blocks.
Referring to FIG. 8, the sub-image PS₈The size of 12 x 8
, The sub-picture S ₈Search block in
77 pixels having the same size as (i.e., 2 × 2 pixels)
Candidate block is sub-picture PS₈Is formed. That
Then, the candidate block forming unit 422-3 outputs the sub-picture PS₈In
Candidate block information representing the pixel value of each candidate block
This is supplied to the candidate block pattern classification unit 422-4.

This search block pattern classification unit 422-4
As shown in FIG. ₈Each candidate
Classify lock patterns. For example, sub video PS₈
The candidate block 740 at has four pixels,
The value of the pixel on the left is 120, and the value of the pixel on the top right is
112, and the value of the lower left pixel is 108,
The value of the pixel on the side is 98. And the candidate block putter
The classification unit 422-4 compares these four pixel values with each other.
Then, a priority is assigned to each pixel. Then, the candidate bro
The pattern classification unit 422-4 assigns priority.
Sub video PS ₈Block 740 and sub video in
PS₈And the pattern to which candidate block 740 belongs to
Occur. In this case, the priority value of candidate block 740
Is the lower right pixel and the lower left pixel as shown in FIG.
1, 2, right upper pixel, upper left pixel in order
3, 4 are assigned. If two or more pixels
Have the same pixel value, they have the same priority
Rankings are assigned.

Thereafter, information and pattern information for each candidate block in the sub-picture PS ₈ are input to the search area forming unit 422-5. Here, the pattern information indicates whether each candidate block in the sub-picture PS ₈ belongs to any pattern.

The search area formation section 422-5 forms a plurality of search regions in the sub video PS _8. In this case, the search area in the sub video PS ₈ is corresponds to one of the search block in the sub-image S _8. Referring again to FIG. 8, in the sub-image PS ₈ , the same position as the position of the search block 710 in the sub-image S ₈ is searched, and a search area 730 in the sub-image PS ₈ is formed around the position. Candidate block information for each candidate block in the sub video PS _8, the pattern information and the search area information is stored in the search area formation section 422-5. Here, the search area information indicates whether each search area in the sub-picture PS ₈ corresponds to some search block in the sub-picture S ₈ .

The first block matching unit 422-6 outputs the sub-picture S
_{In step 8} , motion estimation is performed for each search block. More specifically, the first block matching unit 422-6 is the time of motion estimation search block, located in the search area in the sub video PS ₈ corresponding to a search block in the sub-picture S _8, the search block in the sub-picture S ₈ A candidate block in the sub-picture PS ₈ having the same pattern as the pattern,
Extracted from the search area forming unit 422-5. Sub video PS having the same pattern as the search block in the sub-picture S ₈
If the candidate block at ₈ is not present, the sub video P
All candidate blocks in the sub video PS ₈ in the appropriate search block at S ₈ is taken out from the search area formation section 422-5.

Thereafter, the error value between each of the extracted candidate blocks in the sub-picture PS ₈ and the search block in the sub-picture S ₈ is calculated. The value of this error is, for example, an average absolute error between the corresponding pixel values of preparative issued candidate block in the pixel value and the sub image PS ₈ of the search block in the sub-image S _8. Among the candidate blocks issued collected in the sub-video PS _8, candidate block which yields a minimum error is determined as the optimum candidate block in the sub video PS _8, the displacement from the best candidate block to the search block in the sub-picture S ₈ , The motion vector MV ₈ for the search block in the sub-picture S ₈ . The motion vector MV ₈ thus determined is supplied to the second block matching unit 424 in FIG.

The second block matching section 424 is connected to the line L414
Upper sub-image S ₄ , sub-image PS ₄ on line L434, and sub-image S from first block matching unit 422-6 in FIG.
Receiving a motion vector MV ₈ to the search block at _8. Each of the sub-images S ₄ has a predetermined size of 2p × 2
It is divided into a plurality of search blocks having p (for example, 4 × 4 pixels). Thereafter, the search region corresponding to a search block in the sub-image S ₄ is an enlarged motion vector of the corresponding search block in the sub-image S ₈ (e.g.,
2MV ₈ ) and is formed in the sub-picture PS ₄ .

[0033] More specifically, to detect the same position as the position of the search block in the sub video S ₄ in the sub video PS _4, to generate a displaced position by displacing the detected position only 2 MV _8. Thereafter, a search area in the sub video PS ₄ around its displaced position, motion estimation search block in the sub-image S ₄ in this search area. afterwards,
In the search area in the sub-picture S _4, to detect a displacement resulting in a minimum error, and determines the displacement as the motion vector MV ₄ to the search block in the sub-image S _4. The motion vector MV ₄ thus determined is supplied to the third block matching unit 426.

The third block matching unit 426 is located on the line L412.
Sub-picture S_Two, Sub-picture PS on line L432_Two, 2nd bu
2MV from lock alignment part 424_FourBased on the sub-picture
S_TwoMotion vector MV for search block in_Two
Is determined and supplied to the fourth block matching unit 428. This second
4-block matching unit 428 is the current frame on line L402
S₁ , Previous frame PS on line L492₁, Third block
2MV from matching unit 426_TwoBased on the current frame S₁
Motion vector MV for search block in₁Decide
Set. These motion vectors MV_TwoAnd MV₁Ask for
Is the motion vector MV_FourIs similar to the process of finding
You. That is, the third block matching unit 426 and the fourth block
The combining unit 428 performs the same function as the second block matching unit 424.
Is the motion vector MV_TwoAnd MV₁Ask for.

After that, the last obtained current frame S ₁
Is the motion vector MV ₁ for the search block in
It is sent to the motion reward section 440 and the transmitter. In the above,
Having described the preferred embodiments of the invention, those skilled in the art will be able to make various modifications without departing from the scope of the invention.

[0036]

Therefore, according to the present invention, the sub-picture S ₈
The motion estimation for the search block in
_Since the search is performed between the search block in _{8 and} the candidate block in the sub-picture PS ₈ having the same pattern as the search block, the calculation process can be further simplified.

[Brief description of the drawings]

FIG. 1 is a schematic diagram illustrating a normal wavelet division process.

FIG. 2 is a schematic diagram showing a sub-image having a pyramid structure.

FIG. 3 is a schematic diagram illustrating a general multi-resolution motion estimation (MRME) method.

FIG. 4 is a schematic diagram illustrating a method of estimating a motion of only a sub video corresponding to a low frequency band.

FIG. 5 is a schematic block diagram of a motion vector detecting device according to the present invention.

FIG. 6 is a detailed block diagram of a motion estimator in FIG. 5;

FIG. 7 is a detailed block diagram of a pattern block matching unit in FIG. 6;

FIG. 8 is a schematic diagram illustrating in detail a motion estimation process of a motion estimation unit according to the present invention.

[Explanation of symbols]

110, 120, 130 Wavelet splitter 302, 306, 310, 314, 318, 322, 326, 330 Search block 304, 308, 312, 316, 320, 324, 328, 332 Optimal candidate block 400 Motion vector detector 410 Current Wavelet dividing unit 420 Motion estimating unit 422 Pattern block matching unit 424 Second block matching unit 426 Third block matching unit 428 Fourth block matching unit 422-1 Search block forming unit 422-2 Search block pattern classification unit 422-3 Candidate block Formation section 422-4 Candidate block pattern classification section 422-5 Search area formation section 422-6 First block matching section 430 Pre-wavelet division section 440 Motion compensation section 450 Subtraction section 460 Discrete cosine transform (DCT) & quantization (Q ) Section 465 Variable length coding (VLC) section 470 Inverse discrete cosine transform (IDCT) & inverse quantization (IQ) section 480 Addition section 490 Memory 710, 720 Search block 730 Search area in sub-picture PS ₈ 740, 750 candidates Block S ₁ current frame S ₂ , _{S 4, S 8, W 8} 1, W 8 2, W 8 3, W 4 1, W 4 2, W 4 3, W
^{_{2 1, W 2 2, W}} 2 3 sub video in the current frame

Claims (9)

[Claims] 1. A motion vector detecting apparatus for detecting a motion vector between a current frame represented in a video signal and a reference frame, the wavelet transform being performed on the current frame to generate M hierarchical current videos. A first wavelet transform means for generating, wherein a current image in a highest layer corresponds to the current frame, a current image in a higher layer has a higher resolution than a current image in a lower layer, and each current image is A first wavelet transform unit having the same number of search blocks, wherein each search block in the lower hierarchy corresponds to each search block in the higher hierarchy; A second wavelet transform means for generating a hierarchical reference image of The reference sub-image in one layer has a plurality of candidate blocks corresponding to each search block in the same layer, and the size of each search block and the corresponding candidate block in the same layer is the same. Second
A wavelet transform unit; a first pattern classifying unit that classifies a search block in the lowest hierarchy into one of predetermined patterns; and each candidate block corresponding to the search block, A second pattern classifying unit that classifies the search block and the candidate block based on the pattern related to the search block and the candidate block; An optimal candidate block searching means for detecting a displacement vector for the search block, the displacement vector for the search block in the highest hierarchical layer based on the detected displacement vector, The motion vector for the search block in the upper layer is And a motion vector determining means for determining the motion vector. 2. The motion vector detecting device according to claim 1, wherein the size of each search block in one hierarchy is the same. 3. When P and Q are integers greater than 1,
The size of each search block in the lowest hierarchy is P ×
The motion vector detecting device according to claim 2, wherein Q is Q. 4. The first and second pattern classification means,
4. The motion vector detecting device according to claim 3, further comprising a block pattern determining means for determining a pattern of one block based on P × Q pixel values contained therein. 5. The block pattern deciding unit decides a pattern of the block by substituting each of the P × Q pixels in the block with a number representing the order of magnitude of a corresponding pixel value. The motion vector detecting device according to claim 4, wherein 6. The optimum candidate block searching means, the candidate block selecting means for selecting a plurality of candidate blocks having the same pattern as the pattern of the search block, and the optimum candidate block among the plurality of candidate blocks. The motion vector detecting device according to claim 5, further comprising: an optimal candidate block determining unit for determining. 7. The displacement vector determination means, based on a displacement vector for a search block in the next lower hierarchy, motion-estimates a search block in one hierarchy for the corresponding candidate block, and Motion estimation means for searching for a candidate block; displacement vector determination means for determining a displacement vector for the search block, representing a displacement between the optimal candidate block and the search block in the hierarchy; 7. The method according to claim 6, further comprising: iteratively estimating a search block in a next higher layer repeatedly to determine a corresponding displacement vector until the displacement vector for the search block is obtained. Motion vector detection device. 8. The motion vector detecting device according to claim 7, wherein the size of the search block in one hierarchy is 2 × 2 times larger than the size of the search block in the next lower hierarchy. 9. The motion vector detecting device according to claim 8, wherein the values of M, P, and Q are 4, 2, and 2, respectively.