JPH10247242A - Block matching searching method and device using the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the operation quantity of block matching and to make hardware compact. SOLUTION: A first matching part 5 operating the distortion quantity of rough precision between a frame being a processing object and a processing reference frame, a judgement part 6 narrowing a searching direction having high possibility for obtaining best matching based on the result, a second matching part 7 operating the distortion quantity of dense precision in a narrowed direction and a moving vector calculation part 8 calculating a moving vector by receiving the output are provided. The judgment part 6 narrows the searching direction from the correlation of matching positions by which the best result and a result next to it are obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for searching for a block matching, and more particularly to a method for performing matching on a block-by-block basis between a frame image to be processed and a frame image to be processed. It relates to the equipment used.

[0002]

2. Description of the Related Art In recent years, with the development of various video-related devices and communication technologies, there has been an increasing demand for transmitting or recording an image with a smaller code amount while maintaining higher image quality. From such a background, an image encoding and transmitting apparatus called a video codec sometimes employs image encoding using motion compensation. The motion compensation refers to a method of reducing the code amount of a frame image A by using a motion vector and a difference of image data with respect to a previously input frame image B when a certain frame image A is input. . When calculating a motion vector, block matching is often performed between the frame image A and the frame image B.

Japanese Patent Laid-Open Publication No. Hei 5-40828 proposes a search method for reducing the amount of calculation for block matching. In this method, in a first step, intermittent block matching inspection is performed over all data planes, and an optimum candidate block is searched. Next, in the second stage,
Detailed block matching is performed within a predetermined range surrounding the optimal candidate block. The optimal block data is finally obtained by the two-stage processing. As a result, compared to performing detailed block matching for all data planes from the beginning, the total calculation amount is reduced.

[0004]

In this block matching search method, the amount of calculation can be reduced by introducing the first stage, but in the second stage, a total of eight detailed steps are performed around the optimal candidate block. Block matching is required. In the second stage, for example, block matching is performed with an accuracy of half a pixel (half pel),
First, it is configured to calculate a reproduced image data value with half-pixel accuracy in eight surrounding directions and calculate an amount of distortion with respect to the reference frame image, that is, a sum of absolute differences of image data. Therefore, the amount of operation in eight directions is still enormous, and the work area of the memory required for the operation also becomes large. Hardware must also be designed at high speed.

[0005] The present invention has been made in view of the above points, and an object thereof is to further reduce the amount of calculation, make the hardware more compact, and secure hardware while ensuring the validity of the final result of block matching. It is an object of the present invention to provide a block matching search method capable of alleviating the required high-speed performance and an apparatus using the method.

[0006]

A block matching search method according to the present invention is a method for performing block matching between a frame image to be processed and a frame image to be a reference for processing. This method includes a first matching step of matching blocks of a frame image to be processed with coarse accuracy in a predetermined search range of a frame image to be a reference of the processing, and performing a best matching based on a result thereof. A step of narrowing down a search direction that is likely to be obtained is included. Further, the method includes a second matching step of matching the frame images to be processed with high precision in the narrowed search direction.

In the present invention, the determination step narrows down the search direction based on the mutual relationship between the positions of a plurality of blocks that showed good matching at the time of performing the first matching step.

At this time, the determination step connects the center of the position of the block showing the best matching at the time of performing the first matching step with the center of the position of the block showing the second best matching. The search direction is narrowed down with the direction as the main direction.

On the other hand, the block matching search device of the present invention is a device that performs block matching between a frame image to be processed and a frame image to be a reference for processing. This apparatus includes a first matching unit that matches blocks of a frame image to be processed with coarse accuracy in a predetermined search range of a frame image that is a reference of the processing, and performs a best matching based on a result thereof. Determining means for narrowing down a search direction having a high possibility of being obtained. The image processing apparatus further includes a second matching unit that matches the frame image to be processed with high accuracy in the narrowed search direction.

[0010] Further, the determining means detects a position of a block showing the best matching as a result of the matching by the first matching means, and a second best matching as a result of the matching by the first matching means. It includes second detection means for detecting the position of the indicated block, and range setting means for narrowing down the search direction based on the detection results of the first and second detection means.

Further, the first matching means performs matching with integer pixel precision, and the second matching means performs matching with half integer image precision.

Alternatively, the first matching means performs matching with coarse integer pixel precision obtained by thinning out search target points, and the second matching means performs matching with finer integer pixel precision than the first matching means. is there.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will be described with reference to the drawings as appropriate.

Embodiment 1 FIG. 1 is a diagram showing a configuration of a block matching search device according to Embodiment 1. In FIG. 1, reference numeral 1 denotes an input frame buffer into which frame image data to be processed is sequentially input, 2 denotes a frame buffer in which previously input frame image data serving as a reference for processing is stored, and 3 denotes an input frame buffer. An input data block output from the frame buffer 1, a reproduced image data block 4 output from the frame buffer 2, a first matching unit 1 for calculating a distortion amount with integer pixel precision, 1
0 is the amount of distortion output from the first matching unit 5 for each block within a predetermined search range (for example, a range close in position between both frames), and 6 is the best matching based on the matching result in the first matching unit 5. A determination unit that narrows down a search direction that is likely to be obtained, 7 is a second matching unit that calculates a distortion amount with half-pixel accuracy in the narrowed search direction, and 8 is an output of the second matching unit 7 Motion vector calculation unit for calculating a motion vector by using
2 is the calculated motion vector.

[0015] In the determination unit 6, the reference 11 detects the position of the reproduced image data block that gives the minimum distortion amount (ie, the best matching result) among the matching results of the first matching unit 5, and detects the input data block of the block. (Vector_m) indicating the relative position with respect to
1), 13 is a vector_min output from the first detection unit 11, and 12 is a first detection unit.
The position of the reproduced image data block that gives the second smallest distortion amount among the matching results of the matching unit 5 (that is, the second best matching result) is detected, and a vector (vector_second) indicating the relative position of the block with respect to the input data block is detected. The second detector 14 outputs the ve output from the second detector 12.
vector_min, 15 is vector_min
13 is a range setting unit that receives a vector_second 14 and sets a range in which a motion vector is searched by block matching with half-pixel accuracy.
5 is a search range of a motion vector (hereinafter simply referred to as a search range).

In the second matching unit 7, reference numeral 18 denotes an image reproducing unit for calculating reproduced image data with half-pixel accuracy in the search range 16, reference numeral 19 denotes reproduced image data reproduced with half-pixel accuracy, and reference numeral 20 denotes reproduced image data 19. An operation unit 21 for calculating the distortion amount with half-pixel accuracy by using the distortion amount obtained by the operation.

Next, the operation will be described. First, the first matching unit 5 inputs the input block data stored in the input frame buffer 1. At the same time, a reproduced image data block within a predetermined search range of a motion vector is read from the frame buffer 2, and an input data block and
The amount of distortion E (Vx, Vy) with the reproduced image data block at the search point with integer pixel precision is calculated based on Equation 1.
Here, the predetermined motion vector search range is ± 16 around the position of the input data block in the X-axis direction and ± 16 in the Y-axis direction.
It is assumed that the range is in the vicinity of 16.

[0018]

(Equation 1) Where A (x, y) is the pixel value of the input data block,
A ′ (x, y) is the pixel value of the reproduced image data block,
x and Vy are movement amounts (−16 ≦ Vx, Vy ≦ 16) based on the position of the input data block.

In the first detecting section 11, the first matching section 5
A vector (Vx_min, Vy_min) indicating the relative position of the reproduced image data block giving the E_min with respect to the input data block, finding the minimum distortion amount E_min among the plurality of distortion amounts E (Vx, Vy) output from the
That is, the vector_min13 is detected. on the other hand,
The second detector 12 obtains the second smallest distortion amount E_second among E (Vx, Vy), and calculates E_second.
Vector_sec indicating the relative position of the reproduced image data block with respect to the input data block
ond14 is output.

The range setting unit 15 sets the vector_min
A search range 16 is set based on the relationship between 13 and vector_second 14. A setting example will be described with reference to FIG. In the figure, ● indicates a position determined by vector_min13, that is, a position at which the best matching is obtained in the first matching unit 5 (hereinafter, referred to as “temporary best position”), and (a), (b), (c), (D), (e),
(F), (g), and (h) respectively correspond to the half-pixel positions of the upper left, the upper right, the upper right, the left, the right, the lower left, the lower right, and the lower right of the temporary best position.

As shown in FIG. 2, the periphery is divided into eight regions 1 to 8 around the temporary best position. Here, vector_second1 viewed from the temporary best position as the origin
4 (ie, vector_second)
If the position indicated by the vector determined by 14-vector_min13 and the “best possible position” hereinafter belongs to the area 1, the search range 16 is set to (b),
In the same manner, the search range 16 is set in accordance with the alphabetic characters assigned to each area in FIG.

In the second matching section 7, first, the image reproducing section 18 calculates reproduced image data 19 with half-pixel accuracy corresponding to the search range 16. FIG. 3 is a diagram illustrating a calculation example. In the figure, if A to I are positions corresponding to the reproduced image data with integer pixel precision, and a to h are positions corresponding to the reproduced image data 19 with half pixel precision, the pixel values of a to h are expressed by Equation 2 respectively.
-9.

A = (A + B + D + E + 2) / 4 (Equation 2) b = (B + E + 1) / 2 (Equation 3) c = (B + C + E + F + 2) / 4 (Equation 4) d = (D + E + 1) / 2 (Equation 5) e = ( E + F + 1) / 2 (Equation 6) f = (D + E + G + H + 2) / 4 (Equation 7) g = (E + H + 1) / 2 (Equation 8) h = (E + F + H + I + 2) / 4 (Equation 9) The constant "2" is based on a software technique for simulating the subsequent rounding at this stage, and is the result of adding 0.5 for each character A-I.

FIG. 4 shows a reproduced image data block with half-pixel accuracy when the possible best position is directly above the provisional best position. In this case, for each point of the reproduced image data block (indicated by the bold frame 50 in the figure) at the tentative best position, the reproduced image data block (half-pixel accuracy) is calculated using Equation 3 in consideration of the search in the upward direction. (Indicated by a thin frame 51 in the figure). Since this calculation is performed in eight ways in the conventional technique, the calculation amount of this part is theoretically reduced to 1/8.

Subsequently, based on the reproduced image data block, the arithmetic unit 20 performs fine block matching. Here, the distortion amount Eh between the reproduced image data block (the thin frame in FIG. 4) at the best possible position and the input data block
(Vxh, Vyh) is calculated by Expression 10.

[0026]

(Equation 2) A ′ _h (x + Vxh, y + Vyh) indicates a reproduced image data block with half-pixel accuracy. Here, in order to consider the right above direction, calculations regarding Vxh = 0 and Vyh = 0.5 are performed. The result of this calculation is provided to the motion vector calculation unit 8 as the distortion amount 21. The calculation of Equation 10 can be performed using the method described in the related art, Vxh, Vyh = −0.5,
This is a portion that requires a total of eight calculations except for the case where Vxh = Vyh = 0 (that is, the provisional best position) assuming 0 and 0.5. In the present embodiment, the amount of calculation for Expression 10 is also theoretically 1
/ 8.

Finally, in the motion vector calculating section 8,
The amount of distortion 21 at the possible best position is compared with E_min at the tentative best position already obtained by the calculation of the first matching unit 5, and a motion vector is determined based on the smaller one. Therefore, the position that is finally considered to be the best is selected from the temporary best position and the possible best position. According to this selection, the motion vector 22 is output.

FIG. 5 is a flowchart showing the procedure of block matching according to the present embodiment. As shown in the figure, first, in S2, a distortion amount E (Vx, Vy) of a reproduced image data block within a predetermined search range of an input data block and a motion vector is calculated by Expression 1. S2 in S3
Is compared with E_min which is the minimum value of the distortion amount at that time, and E_mi is calculated.
If n> E (Vx, Vy), E_min, E_s in S4
econd, vector_min13, vector
_Second14 is updated. E (Vx, Vy) ≧
If E_min, the process proceeds to S5.

In S5, it is determined whether or not the calculation of the distortion amounts for all the reproduced image data blocks within the predetermined search range has been completed. If not completed, the reproduced image data blocks are updated in S6 and the process returns to S2. If it is determined in S5 that the process is to be ended, the process proceeds to S7. In S7, the search range 16 with half-pixel accuracy, that is, the best possible position is determined based on the relationship between the vector_min13 and the vector_second14. In S8, a reproduced image data block with half-pixel accuracy corresponding to the search range 16 is calculated using one of Expressions 2 to 9.

In S9, the amount of distortion Eh (Vx) between the input data block and the reproduced image data block obtained in S8 is determined.
h, Vyh) is calculated based on Equation 10. In S10,
The distortion amount Eh (Vxh, Vyh) at the best possible position calculated in S9 is compared with the distortion amount E_min at the temporary best position, and E_min> Eh (Vx
h, Vyh), the update of E_min and v
ector_min13 is updated. Eh (Vxh,
Vyh) ≧ E_min, the process proceeds to S12, and the vector
_Min13 is output. After this, the final vector
The motion vector 22 is determined based on r_min13.

As described above, according to this embodiment,
The search direction in which the best matching can be obtained is narrowed down based on the mutual relationship between the positions of the plurality of blocks showing good matching in the first matching unit 5, so that the amount of calculation is greatly reduced and the processing time is reduced. Can be shortened. As a result, hardware such as a memory necessary for the operation can be made compact.

In the present embodiment, in FIG. 2, only one position is determined as the search range 16 as a possible best position, but there are naturally other methods. For example, vector_second1 viewed from the temporary best position
When 4 exists in the following areas, the three directions shown on the right may be set in the search range 16 as possible best positions, and the search may have a certain width.

Area 1 ... (a), (b), (c) Area 3 ... (c), (e), (h) Area 5 ... (f), (g), (h) Area 7... (A), (d), (f) As for the relationship between the region and the search direction, FIG. 6 may be used instead of FIG. 2. In the case of FIG.
Areas 1 and 6 correspond to one of the four quadrants, respectively.
, Y> 0, x = 0, region 3 corresponds to x> 0, y = 0, region 5 corresponds to y <0, x = 0, and region 7 corresponds to x <0, y = 0. Regarding which of FIG. 2 and FIG. 6 is adopted,
It may be determined appropriately by experiment or the like.

Embodiment 2 In the first embodiment, the search range 16 by the second matching unit 7 is determined by vec
tor_min13 and vector_second14
Was used. Here, the motion vector vec corresponding to the reproduced image data block having the third smallest distortion is further added.
A case where tor_third is also considered will be described.

FIG. 7 is a block diagram showing a configuration of the block matching search device according to the second embodiment. In the figure, reference numeral 30 denotes a determination unit that determines the search range 16 in consideration of vector_third, and reference numeral 40 denotes vector_t.
The third detector 41 for detecting the third is a vector_
third and 42 are range setting units for searching for a vector with half-pixel accuracy. Other configurations are the same as those in FIG. 1, and the same reference numerals as those in FIG.

The operation of this device will be described. Embodiment 1
Similarly, in the first matching unit 5, the distortion amount E (Vx, Vy)
However, E_min in the first detection unit 11 is
In E_second, E_second is detected. Further, the third detection unit 40 obtains the third smallest distortion amount E_third of E (Vx, Vy), and outputs a vector vector_third 41 indicating the relative position of the reproduced image data block to which the E_third is provided with respect to the input data block.

In the range setting unit 42, the vector_mi
n13, vector_second14, and vector
The search range 16 with half-pixel accuracy is obtained based on the relationship with r_third41. The notation in FIG. 2 is used for the relationship between the area and the search direction, and the search range 16 can be set, for example, according to FIGS.

FIG. 8 shows a case where the temporary best position indicated by the vector_min13 is set as the origin and the vector_sec
9 shows an example of setting the search range 16 when ond14 is included in the areas 2, 4, 6, and 8. Similarly, FIG.
An example of setting the search range 16 when ond14 is included in the regions 1, 3, 5, and 7 is shown.

As shown in FIG. 8, the vector_second
When d14 is included in regions 2, 4, 6, and 8, vect
or_third41 is non-reference (don't care),
The search range 16 is (c), (h), (f),
(A) is determined. On the other hand, vector_second1
If 4 is included in regions 1, 3, 5, and 7, vector
_Third41 is also referred to, for example, vector_
second14 is area 1, vector_third
If 41 is included in the area 3, the search range is only (c). The procedure of block matching according to the present embodiment is as follows.
In S4 of FIG. 5, E_third and vec
It is sufficient to update tor_third.
Can be left as is.

As described above, according to this embodiment, the vector
Since the search range is set in addition to the or_third 41, there is an effect of increasing the validity of the set search range.
Conversely, it is easy to narrow the final search range to one as shown in FIGS. 8 and 9, and the amount of calculation and hardware can be greatly reduced. Note that, also in this embodiment, another relationship such as FIG. 6 can be used instead of FIG. 2 for the region and the search direction.

Embodiment 3 In the first and second embodiments, the search in the first matching unit 5 is performed with integer pixel precision, and the search in the second matching unit 7 is performed with half pixel precision. Here, the case where the former is performed with coarse integer pixel precision and the latter is performed with finer integer pixel precision will be described.

The configuration of the block matching search apparatus according to this embodiment is the same as that of FIG. However, unlike the first embodiment, the first matching unit 5 calculates the amount of distortion with coarse integer pixel accuracy. The first detection unit 11 of the determination unit 6 includes:
V corresponding to the minimum distortion amount in the first matching unit 5
ector_min13, and the second detector 1
2 is vector_s corresponding to the second smallest distortion amount
econd14 is detected. The range setting unit 15 determines these 2
A search range is set from the two vectors in the same manner as in the first embodiment. The image reproducing unit 18 of the second matching unit 7 calculates the reproduced image data 19 with a fine integer pixel precision in the set search range. The calculation unit 20 calculates the distortion amount 21 with fine integer pixel precision. The motion vector calculator 8 calculates
The motion vector 22 is output with reference to the calculation result.

Next, the operation will be described. First, the first matching unit 5 inputs the input block data stored in the input frame buffer 1. At the same time, a reproduced image data block within a predetermined search range of the motion vector is read from the frame buffer 2, and a distortion amount E (Vx, Vy) between the input data block and the reproduced image data block at the search point with coarse integer pixel precision is calculated by an equation. 11 is calculated. The predetermined search range is a range of ± 16 in the X-axis direction and ± 16 in the Y-axis direction centered on the position of the input data block, and a search with coarse integer precision is performed every other pixel.

[0044]

(Equation 3) Here, A (x, y) is an input data block, A "
(X, y) is a reproduced image data block corresponding to a search point of coarse accuracy, and Vx and Vy are movement amounts (Vx = 2Nx, Vy = 2) based on coordinates indicating the position of the input data block.
Ny, −8 ≦ Nx, Ny ≦ 8). Therefore, the first matching unit 5 calculates E (-16, -16), E (-16, -14),... E (-16,14), E (-16,16), E (-14, -16), E (-14, -14), ....... E (-14,16),: E (16, -16) , E (16, -14),... E (16,16) are calculated in order.

Hereinafter, the processing of the present embodiment is realized by reading the processing of the integer pixel precision and the half-integer pixel precision of the first embodiment into the processing of the coarse integer pixel precision and the processing of the finer integer pixel precision, respectively. For example, the distortion amount Em (Vx
m, Vym) is calculated by the following equation instead of equation 10.

[0046]

(Equation 4) Here, A ″ _h (x + Vxm, y + Vym) indicates a reproduced image data block with dense integer pixel precision. Vxm and Vy
m takes one of -1, 0 and 1 according to the search direction.

As described above, according to this embodiment,
The same effect as in the first embodiment can be realized by another process, that is, two-stage matching between coarse integer pixel precision and denser integer pixel precision. According to this embodiment, the number of calculations performed by the first matching unit 5 is smaller than in the first embodiment. Therefore, Embodiment 1 provides higher image quality and Embodiment 3
May be used for purposes that require less computation and processing time.

In this embodiment, similarly to the second embodiment, the vector_th showing the third best result is shown.
ird41 may be added. In that case, Embodiment 2
The processing of integer pixel precision and half integer pixel precision may be read as coarse integer pixel precision and denser integer pixel precision processing, respectively.

In the above embodiment, an example of the block matching calculation based on the sum of absolute differences has been described. However, other differences such as a sum of squares can be used. In addition, although the case where the present invention is applied to, for example, motion-compensated inter-frame coding has been described here, the present invention can also be applied to a case where block data most similar to each other is searched for in the same frame image. In this case, the frame image to be processed is the same as the frame image to be the reference for processing.

Further, here, the two-stage matching is “integer pixel precision → half integer pixel precision” or “coarse integer pixel precision → dense integer pixel precision”, but these are combined to “coarse integer pixel precision → A modification example of “half-integer pixel accuracy” is also possible.

[0051]

According to the block matching search method of the present invention, since the search direction which is likely to obtain the best matching is narrowed down before the matching with high precision, the calculation amount is reduced and the hardware is reduced. Simplification of wear,
Shortening of processing time is realized.

When the search direction is narrowed down based on the mutual relationship between the positions of a plurality of blocks showing good matching with coarse accuracy, the narrowed-down result has validity.

In the case where the search direction is narrowed down by using the direction connecting the center of the position of the block showing the best matching with coarse accuracy and the center of the position of the block showing the second best matching as the main direction, it is difficult to narrow down the search direction. Natural embossing based on the rule is realized.

On the other hand, according to the block matching search apparatus of the present invention, since the search direction which is likely to obtain the best matching is narrowed down before the matching with high precision, the calculation amount is reduced and the hardware is reduced. This simplifies the wear and shortens the processing time.

When a first detecting means for detecting a position of a block showing the best matching as a result of the coarse precision matching and a second detecting means for similarly detecting a position of a block showing the second best matching are included. In addition, it is possible to make the search direction narrow down.

When performing two-step processing of matching with integer pixel accuracy and matching with half-integer image accuracy, the amount of calculation can be reduced while maintaining high image quality.

When performing two-stage processing of matching with coarse integer pixel precision and matching with finer integer pixel precision,
Further reduction in the amount of calculation is realized.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of a block matching search device according to a first embodiment.

FIG. 2 is a diagram showing a vector_mi in the first embodiment.
vector_sec based on the position indicated by n
It is a figure showing the relation of the field to which ond belongs, and the search range set up.

FIG. 3 is a diagram illustrating an example of calculation of reproduced image data by an image reproducing unit according to the first embodiment.

FIG. 4 is a diagram illustrating a reproduced image data block with half-pixel accuracy when a possible best position is immediately above a provisional best position in the first embodiment.

FIG. 5 is a flowchart showing a procedure of block matching according to the first embodiment.

FIG. 6 is a diagram illustrating a configuration of the first embodiment.
vector_sec based on the position indicated by n
FIG. 14 is a diagram illustrating another relationship between an area to which ond belongs and a set search range.

FIG. 7 is a block diagram showing a configuration of a block matching search device according to a second embodiment.

FIG. 8 is a diagram showing a search range set in the second embodiment.

FIG. 9 is a diagram showing a search range set in the second embodiment.

[Explanation of symbols]

1 input frame buffer, 2 frame buffer, 3
Input data block, 4 playback image data block,
5 first matching unit, 6 determination unit, 7 second matching unit, 8 motion vector calculation unit, 10 distortion amount, 11
1st detection part, 12 2nd detection part, 13 vector_
min, 14 vector_second, 15 range setting section, 16 search range, 18 image playback section, 19
Reconstructed image data, 20 operation unit, 21 distortion amount, 22
Motion vector, 30 determining unit, 40 third detecting unit, 41
vector_third, 42 range setting unit, 50
The reproduced image data block at the temporary best position, 51 reproduced image data blocks with half-pixel accuracy.

Claims (7)

[Claims] 1. A method for performing block matching between a frame image to be processed and a frame image to be a reference for processing, the method comprising the steps of: A first matching step of matching blocks of a target frame image; a determination step of narrowing down a search direction that is likely to obtain the best matching based on the result; A second matching step of matching the frame images to be processed with a high precision. 2. The method according to claim 1, wherein the determining step narrows down the search direction based on the mutual relationship between the positions of the plurality of blocks that showed good matching at the time of performing the first matching step.
The method described in. 3. The method according to claim 1, wherein the determining step is a direction connecting the center of the position of the block showing the best matching at the time of performing the first matching step and the center of the position of the block showing the second best matching. 3. The method according to claim 2, wherein the search direction is narrowed down by setting the main direction to. 4. An apparatus for performing block matching between a frame image to be processed and a frame image to be a reference for processing, the apparatus comprising: First matching means for matching blocks of a target frame image, determination means for narrowing down a search direction that is likely to obtain the best matching based on the result, And a second matching means for matching the frame image to be processed with high precision. 5. The determination means comprises: a first detection means for detecting a position of a block showing the best matching as a result of the matching by the first matching means; and a second best matching as a result of the matching by the first matching means. 5. The apparatus according to claim 4, further comprising: a second detection unit configured to detect a position of the block indicating the following; and a range setting unit configured to narrow down a search direction using detection results obtained by the first and second detection units. 6. 6. The apparatus according to claim 4, wherein said first matching means performs matching with integer pixel precision, and said second matching means performs matching with half integer image precision. 7. The first matching means performs matching with a coarse integer pixel precision obtained by thinning out search target points, and the second matching means performs matching with a finer integer pixel precision than the first matching means. The apparatus according to any one of claims 4 to 6.