JPH11289540A - Image coder and its method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the image coder and its method where remarkably deterioration in image quality is avoided even when the motion of an image is conspicuous. SOLUTION: In the image coder 20 and its method where motion in 2-way is detected by comparing a received past retrieval object image with a current image and a future retrieval object image with the current image in the unit of prescribed data at a retrieval address AR set respectively to them and an input image D10 is compression-coded based on a motion vector detected by the motion detection, the retrieval object image and the retrieved address AR are set optionally depending on a motion of the input image D10 to surely detect a one-way motion at least in the case that the motion of the image is conspicuous and the image quality of the coded image is enhanced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[Table of Contents] The present invention will be described in the following order.

[0002] 2. Description of the Related Art [0004] Prior Art (FIGS. 12 to 14) Problems to be Solved by the Invention Means for Solving the Problems Embodiments of the Invention (FIGS.

[0003]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus and method, and more particularly to an image coding apparatus and method for detecting a motion vector of an input image and coding the input image based on the motion vector. It is suitable.

[0004]

2. Description of the Related Art Conventionally, for example, MPEG (Moving Picture)
2. Description of the Related Art In an image encoding apparatus that encodes a moving image based on the Experts Group) standard, for example, 15 frames of moving image data are encoded as one processing unit called a GOP (Group Of Pictures).

[0005] One GOP includes an I-Picture (Intra-Pic
ture: intra-frame coded image), P-picture (Predictiv
There are coding types in frame units called e-Picture: inter-frame forward predictive coded image) and B picture (Bidirectionally Predictive-Picture: bidirectional predictive coded image).

[0006] The I picture is for maintaining the independence of the GOP, and the entire picture is encoded.

[0007] The P picture divides the current frame (reference frame) to be coded into macroblocks of, for example, 16 × 16 pixels, and uses each of the divided macroblocks as a reference block. Then, a past I-picture or P-picture is used as a search frame to search within a predetermined search range within the search frame, and block matching with the reference block is performed in macro block units.
As a result, the distance of the block that matches the most is detected as a motion vector, and motion compensation is performed using the motion vector when encoding difference data from the search frame. Used for motion compensation. By the way, as a method of block matching,
A method is used in which all pixels in the reference block and the search block are subtracted, the sum of absolute values or the sum of squares is calculated, and the position where the calculation result is the minimum is matched to the block (full search block matching method). Have been.

[0008] The B-picture searches for past and future bidirectional I-pictures and P-pictures using each macroblock of the current frame (reference frame) to be encoded as a reference block. Then, block matching with the reference block is performed in macroblock units within a predetermined search range in the search frame. As a result, the distance of the block that matches the most is detected as a motion vector, and motion compensation is performed using the motion vector when encoding difference data from the search frame. Used for motion compensation.

As shown in FIG. 12, when performing bidirectional prediction, first, a P-picture (P6) is used as a reference frame, and an I-picture (I3) three frames away from the reference frame (P6) is used. A motion vector is detected only in the forward direction as a search frame. Next, a forward motion vector detection and a P-picture (P6) using the B-picture (B4) as a reference frame and the I-picture (I3) as a search frame.
Is performed as a search frame, and a backward motion vector is detected. And then, B picture (B
Using the reference frame 5) as a reference frame, bidirectional prediction is performed, which includes forward motion vector detection using an I-picture (I3) as a search frame and backward motion vector detection using a P-picture (P6) as a search frame.

FIG. 13 shows a motion vector detection order in such bidirectional predictive coding. In FIG. 13, input images B1, B2, I3, B4,... Each represent a picture type (I picture, B picture, P picture) by alphabet (I, B, P) and an encoding order by a frame number. ing.

FIG. 13 shows a case where two B-pictures for performing bidirectional prediction are used, and a frame marked with “○” in the figure indicates a frame to be currently encoded. The starting point of the arrow in the figure indicates a reference frame, and the frame indicated by the arrow indicates a search frame. Incidentally, the vertical axis in FIG. 13 represents the time in frame units.

In FIG. 13, for example, at time "3", bidirectional prediction is performed on a search frame (I3) for forward prediction and a search frame (P6) for backward prediction using the current frame (B4) as a reference macroblock. To obtain two motion vectors.

[0013]

When a motion vector is detected in both past and future directions, it may be difficult to detect one of the two-way motion detection if a scene change point exists in the middle. Atsuta. For example, FIG.
As shown in (2), when encoding is performed using two B pictures, one motion vector is detected in one frame and the other motion vector is detected in two frames. . In this case, if the motion of the image is large, 2
When the search range is insufficient in detecting the motion vector between frames, and when the motion of the image is further increased, it is difficult to detect the motion vector in both directions.

That is, as shown in FIG. 14, when there are two motion detection circuits each having a search range AR1 and AR2 of ± A pixels for forward prediction and backward prediction, for example, the video camera is panned at high speed. Assuming that (moving) and the actual motion is equivalent to B pixels (B> A) between one frame, for example, in forward prediction between one frame using B picture as a reference frame, FIG. As shown, the block to be matched is the search range AR.
Therefore, it is difficult to accurately detect a motion vector. In this case, as shown in FIG.
As a matter of course, in backward prediction between two frames, the block to be block-matched moves out of the search range AR2 by moving by 2B pixels from the reference position, and it becomes difficult to detect a motion vector accurately.

As a method for solving such a problem, as shown in FIG. 14C, a method of expanding the search range AR3 can be considered. There is a problem that it is inevitable that the detection becomes difficult and the image quality of the encoded image is deteriorated.

The present invention has been made in view of the above points, and it is an object of the present invention to propose an image coding apparatus and a method thereof capable of avoiding a significant deterioration of the image quality even when the motion of the image is large. .

[0017]

According to the present invention, in order to solve the above-mentioned problems, a predetermined data unit is set at a predetermined range of search locations set for an input past search target image and a future search target image. In the image coding apparatus and method for performing bidirectional motion detection by comparing the current image with the current image and compression-coding the input image based on the motion vector detected by the motion detection,
By arbitrarily setting the search target image and the search location according to the motion of the input image, even when the motion of the image is large, motion detection in at least one direction can be reliably performed. Can be improved.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings.

In FIG. 1, reference numeral 20 denotes an image encoding apparatus as a whole, which receives input image data D10 to a motion vector detecting section 21 and an arithmetic section 24. Motion vector detector 2
1 is 2 stored in a frame memory provided therein.
Vector information D21 between two frames (pictures)
Is supplied to the motion compensation unit 31.

At this time, the motion compensating unit 31 performs a motion compensation process using the motion vector information D21 on the reference image stored in the frame memory 30 to generate predicted image data D31. To supply.

The arithmetic unit 24 calculates a difference between the input image data D10 and the predicted image data D31, and calculates the difference data D
24 is supplied to a DCT (Discrete Cosine Transform) unit 25. The DCT unit 25 applies a DC to the difference data D24.
By performing T (discrete cosine transform) processing, the DCT
A coefficient sequence D25 is generated and supplied to the quantization unit 26. The quantization unit 26 generates quantized data D26 by quantizing the DCT coefficient sequence D25, and supplies this to the variable-length coding unit 33 and the inverse quantization unit 27, respectively.

The inverse quantization unit 27 restores the DCT coefficient sequence D27 by performing an inverse quantization process on the quantized data D26. The DCT coefficient sequence D27 is further supplied to an inverse DCT unit 28 and subjected to inverse DCT processing. Thus, the inverse DCT unit 28 restores the difference data D28 corresponding to the picture type, and outputs this to the arithmetic unit 29.

The arithmetic unit 29 generates reference image data D29 by adding the predicted image data D31 output from the motion compensating unit 31 to the difference data D28, and stores this in the frame memory 30.

The difference data quantized through the DCT unit 25 and the quantization unit 26 is restored as the difference data D 28 by the inverse quantization unit 27 and the inverse DCT unit 28, and is predicted by the arithmetic unit 29. The addition to D31 results in reference image data D29. As a result, a reference image for a subsequent frame (picture) is prepared in the frame memory 30.

Here, the motion vector detecting section 21 stores the input image data D10 in the frame memory 5 as shown in FIG.
1 based on the read control signal D52 output from the control determination unit 62, and reads out the image data of the reference frame to be coded in units of macroblocks. This is supplied to the detection units ME1 and ME2.

The frame memory 51 has a control judging unit 62
Based on a read control signal D52 output from the CPU, image data in a predetermined search range in a search frame is sequentially read out for each macroblock, and supplied to the motion detection units ME1 and ME2 as search block data D54.

In this case, the control determination section 62 supplies the search block data D54 of the search frame used for the forward prediction in the bidirectional prediction to one of the motion detecting sections ME1 (or ME2) and the backward predicting section. Is supplied to the other motion detection unit ME2 (or the motion detection unit ME1).

Based on the reference block data D53 and the search block data D54, each of the motion detection sections ME1 and ME2 supplies the distance between the reference block and the search block to the control determination section 62 as motion vector data D56 and D60, respectively. The difference between the image data of the reference block and the image data of the search block is supplied to the control determination unit 62 as residual data D57 and D61.

The control judging section 62 calculates the motion vector data D56 and the residual data D supplied from the motion detecting section ME1.
Based on 57, block matching within a predetermined search range is performed. In this block matching, the control judging section 62 searches for a block in which the sum of absolute values or the sum of squares of the result of subtraction between all pixels of the reference block and the search block is the smallest, and the result of the block matching is obtained. The obtained motion vector between the search block and the reference block is used as a motion vector in forward prediction (or backward prediction) together with control information such as a motion compensation mode and motion information D.
As 63, it is supplied to the motion compensator 31 (FIG. 1) and the variable length encoder 33 (FIG. 1).

In the same manner, the control determination section 62
Performs block matching within a predetermined search range based on the motion vector data D60 and the residual data D61 supplied from the motion detection unit ME2. In this block matching, the control judging section 62 searches for a block in which the sum of absolute values or the sum of squares of the result of subtraction between all pixels of the reference block and the search block is the smallest, and the result of the block matching is obtained. The obtained motion vector between the search block and the reference block is used as a motion vector in backward prediction (or forward prediction) together with control information such as a motion compensation mode and the like as motion information D63 as motion information D63.
(FIG. 1) and the variable length coding unit 33 (FIG. 1).

Here, the control determination unit 62 determines a search frame for performing bidirectional prediction and a search range in the search frame based on motion information such as a motion vector and a residual, and is determined by this. A search frame and a search range are designated by a read control signal D52, and the macroblock of the determined search frame and search range is read.

That is, FIG. 3 shows a basic motion prediction method in the motion vector detecting section 21 of this embodiment.
In FIG. 3, input images B1, B2, I3, B4,
.. Represent the picture types (I-picture, B-picture, P-picture) by alphabet (I, B, P) and the encoding order by the frame number. In the case of this embodiment, two B pictures for performing bidirectional prediction are used, and a frame marked with “○” in the figure indicates a frame to be currently encoded. The starting point of the arrow in the figure indicates the reference frame,
The frame indicated by the arrow indicates a search frame.
Incidentally, in FIG. 3, the vertical axis represents time in units of frames.

In FIG. 3, for example, at time “3”, when bidirectional prediction is performed using the current frame B picture (B4) as a reference frame, at time “0”, I picture (I3) is set as a search frame in advance. Forward prediction is performed, and at time “3”, the P picture (P
Post prediction is performed using 6) as a search frame.

For example, when performing bidirectional prediction using the B picture (B5), which is the current frame, as a reference frame at time "4", forward prediction using I picture (I3) as a search frame in advance at time "1" is performed. At time "4", post-prediction is performed using P picture (P6) as a search frame.

As described above, the control determining unit 62 (FIG. 2) of the motion vector detecting unit 21 performs one of the predictions in advance when performing the bidirectional prediction using the B picture as the reference frame. I have.

In FIG. 3, for example, when only one arrow is shown at the same time in the search for the I picture (I3) when the P picture (P6) at time "2" is used as the reference frame, The search frame (I) is sent to the two motion detectors ME1 and ME2 of the vector detector 21.
The search is performed by assigning different search ranges in 3).

In such a basic motion prediction method, the control determination unit 62 performs a search in one of the predictions performed in advance (for example, a search from the reference frame (B4) to the search frame (I3) at time “0” in FIG. 3). It is determined whether the range is sufficient.

That is, the motion detectors ME1 and ME2
Reads the reference block data D53 of the reference frame and the search block data D54 of the search frame supplied from the frame memory 51 for each field, obtains a motion vector and a residual for each field, and obtains the motion vector data D56. , D60 and the residual data D57, D61 to the control determination unit 62.

The control judging section 62 compares the residuals of the motion vector detection at a plurality of field intervals, and determines that the residual at the time of detecting the motion vector between the M fields is smaller than the residual between the N fields (M < N), it is determined that the search range is insufficient.

That is, in FIG. 3, when a motion vector is detected from the reference frame (B4) at time “0” to the search frame (I3), as shown in FIG.
There are four types of field motion vectors V1 to V4,
There are three types of field intervals: one field interval, two field intervals, and three field intervals.

Here, the residual between one field is MAE1, the residual between two fields is MAE2, the residual between three fields is MAE3, and the threshold value for determining whether or not motion compensation can be performed is Th. And The threshold value Th is not constant but may change depending on each image or each macroblock. As an example of the threshold Th, the amount of information predicted when macroblock data is transmitted without performing motion compensation can be considered.

Then, the control judgment section 62 calculates three residuals MAE1, MAE2, and M obtained at each field interval.
It is determined whether or not the search range is sufficient according to the value of AE3.

That is, the residuals MAE1, MAE2 and M
Each value of AE3 is represented by the following equation:

[0044]

(Equation 1)

When the above condition is satisfied, the control determination section 62 determines that the search range is sufficient.

Accordingly, in this case, as shown in FIGS. 5A and 5B, a search block having a block matching exists in a preset search range.
The control determination unit 62 detects each motion vector by performing forward prediction and backward prediction at times “0” and “3” in FIG.

In contrast, the following equation:

[0048]

(Equation 2)

Or the following formula:

[0050]

(Equation 3)

When the residual MAE1 between one field and the residual MAE2 between two fields is much smaller than the residual MAE3 between three fields as shown by In the motion detection between the two fields, it is determined that the search range is insufficient, whereas in the motion detection between the three fields, the search range is insufficient. Incidentally, in Expression (3), Th2 is a threshold value representing the ratio of the residual MAE1 between one field.

On the other hand, the following equation:

[0053]

(Equation 4)

Or the following formula:

[0055]

(Equation 5)

As shown in the following expression, the residual MAE1 between one field is smaller than the threshold value, and the residual ME1 between the two fields is
When the residual MAE3 between the AE2 and the three fields is larger or more than the threshold Th, the control determination unit 62 determines that only the motion detection for one field has a sufficient search range.

In this manner, when it is determined that the search range is sufficient in the motion detection for at least one field, the control determination unit 62 determines the maximum motion for one field from the motion vector distribution for which the search range is sufficient. The amount is estimated and the search range for the required field interval is predicted.

That is, the motion vector between one field is MV1 and the motion vector between two fields is MV2, 3
Assuming that the motion vector between the fields is MV3,

[0059]

(Equation 6)

[0060]

(Equation 7)

It is assumed that there is a relationship represented by

In a state where the search range is sufficient only in the motion detection for at least one field, FIG.
As shown in (A), a motion vector M for one field
Only V1 falls within the search range AR1, and the motion vector MV2 between the two fields and the motion vector MV3 between the three fields are expected to be outside the search range AR1.

Accordingly, at this time, the control judgment section 62
Based on the equations (6) and (7), the motion vector MV2 between the two fields and the motion vector M between the three fields
V3 is predicted, and the location of the search range AR1 is changed accordingly. As a result, as shown in FIG. 6B, a sufficient search range A for searching for the motion vectors MV2 and MV3 is obtained.
R1 'is obtained at a different place with the same area as the search range AR1. Incidentally, in FIG. 6, the search range AR1 'is obtained by focusing on only one motion vector. However, the present invention is not limited to this, and the search range AR1' is not limited to this.
The distribution of the motion vector MV1 between the fields is obtained, and many motion vectors MV2 and MV3 between the two fields and between the three fields are obtained based on the distribution.
Of the obtained motion vectors MV2 and MV3,
It is also possible to determine a search range in which as many vectors as possible are detected.

As described above, in the motion detection from the reference frame (B4) to the search frame (I3) at time "0" in FIG. 3, it is determined that the search range is insufficient for at least two fields, and a new search range is determined. When AR1 'is obtained, the control determination unit 62 does not perform the motion detection to the search frame (I3) two frames away from the reference frame (B5) at the time "1" in FIG.
As shown in (1), at time "1", the reference frame is again set to the B frame (B4) and 1
Search range AR with newly set motion detection between frames
1 '.

As a result, as shown in FIG. 8A, a motion vector is obtained using the newly set search range AR1 '. FIG. 8A shows a new search range AR1 'set when it is determined that there is a large movement in the horizontal right direction. Thus, in the second search of the reference frame (B4) at time “1” in FIG. 7, a motion vector from the reference frame (B4) to the search frame (I3) is detected.

When the motion vector of the reference frame (B4) is detected at time "1" in FIG.
The reference frame to be executed at time “3” (B
No search is performed from 4) to the search frame (P6). Instead, a search is performed to the search frame (P6) one frame behind the new reference frame (B5). At this time, as shown in FIG. 9A, if there is no motion vector to be detected in the search range AR2, at the subsequent time “4”, as shown in FIG. 3, the reference frame (B8)
It is clear that even if the search frame (P6) two frames ahead of the search range is searched, the search range AR2 is insufficient as shown in FIG. 9B. Therefore, at this time, the control determination unit 62
Represents the reference frame (B) as shown at time “4” in FIG.
The search for the search frame (P6) one frame away from 5) and the search for the search frame (P6) one frame away from the reference frame (B7) are performed.

As a result, as shown in FIG. 9C, a new search range AR2 'can be set to obtain a motion vector.

Incidentally, the motion vector in the direction in which no search is performed at all is, for example, forcibly set as a zero vector so that the image quality of the still image portion is not reduced by the operation.

Thus, at times "1", "3",
B picture (B4,
The search using B5, B7 and B8) as a reference frame is performed between one frame using the newly set search range. At this time, even if the search range is insufficient in the search between two frames, one frame is searched. A motion vector can be reliably detected by a search between the two.

FIG. 10 shows a search processing procedure in the control determination section 62, and the control determination section 62 executes step SP0.
In step SP1, the motion vector and the residual for two fields constituting the reference frame are received from the motion detector ME1 in step SP1, and the motion vectors for two fields constituting the search frame are received from the motion detector ME2. And receive the residual.

Then, the control judging section 62 proceeds to step SP2, and based on the motion vectors and residuals for the four fields received from the motion detecting sections ME1 and ME2, the residual MAE1 between one field and the residual MAE2 between two fields. And the residual MAE3 between the three fields is determined, and the determination of the threshold value Th is performed based on the above-described equations (1) to (5).

If the result of the determination in step SP2 satisfies the above expression (1) (condition a), the control determining unit 62 proceeds to step SP3 to set the search range of the normal bidirectional prediction shown in FIG. The read control signal D52 (FIG. 2) is set in the frame memory 51 in step SP7 based on the set search range and search frame.
(FIG. 2), motion detection is performed by the motion detection units ME1 and ME2.

On the other hand, if the determination in step SP2 satisfies the above expression (2) or (3) (condition b), the control determination section 62 proceeds to step SP4 and executes the one-field described above with reference to FIG. The search range is determined using the motion vector MV1 between the two fields and the motion vector MV2 between the two fields, and is set so as to perform only one-way prediction for one frame as described above with reference to FIG. Thus, the process proceeds to step SP7, and motion detection is performed using the set search range.

On the other hand, if the judgment in step SP2 satisfies the above-mentioned expression (4) or (5) (condition c), the control judgment section 62 proceeds to step SP5 and executes the one-field operation described in FIG. The search range is determined using the inter-motion vector MV1, and the setting is made such that only unidirectional prediction for one frame is performed as described above with reference to FIG.
Thus, the process proceeds to step SP7, and motion detection is performed using the set search range.

On the other hand, the judgment in step SP2 is as follows:

[0076]

(Equation 8)

If the condition is satisfied (condition d), this means that the search range between all the fields is insufficient. At this time, the control judgment unit 62 proceeds to step SP6 and executes the normal bidirectional prediction. The motion detection is performed in step SP7 using the search range in.

At time "7" in FIG. 7, the control determining section 62 determines that each residual has become sufficiently small.
The processing returns to the normal processing shown in FIG.

In the above configuration, the control determination section 62
With respect to the B picture for which bidirectional prediction is performed, motion prediction is performed on the search frame at one time (forward or backward) at different times (FIG. 3). At this time, for example, in the search for one of the motion vectors first performed at time “0” in FIG. 3, if the residual in the search range set at this time is large and a motion vector to be detected cannot be found, control determination is performed. The unit 62 predicts a search place where a motion vector between two fields can be searched based on a motion vector between one field where motion can be detected at the shortest interval with respect to the search frame at this time, and newly predicts this. Set as search range.

Then, in the bidirectional prediction shown in FIG. 3, a prediction two frames apart is changed to a prediction between one frame, and a motion vector search is performed. As described above, even when the motion of the moving image is large, it is predicted that the motion vector in at least one field is accurately obtained, and it is difficult to search for the motion vector between two frames when the motion is large. Focusing on a certain point, the prediction between two frames in bidirectional prediction is changed between one frame, and the search range (location) is changed to search for a motion vector between one frame, so that the motion vector can be surely determined. Vectors can be detected.

Thus, according to the above arrangement, at least a unidirectional prediction motion vector can be detected even when the image motion is large in encoding for bidirectional prediction, and as a result, encoding efficiency can be improved. In addition, the image quality of the encoded image can be improved.

In the above-described embodiment, as described above with reference to FIG.
Although a case has been described in which motion prediction between one frame is performed using E1 and ME2, the present invention is not limited to this. For example, as shown in FIG.
2, different search ranges may be searched at the same time. In this case, as shown in FIGS. 8D and 9D, a search range of the two motion detection units ME1 and ME2 is set.

In the above embodiment, the case where the number of B pictures is two (M = 3) has been described. However, the present invention is not limited to this, and the number of B pictures is one or three or more. It may be hot.

In the above-described embodiment, a case has been described in which block matching is performed in units of macro blocks. However, the present invention is not limited to this, and search may be performed in other various data units. .

[0085]

As described above, according to the present invention, the present image is searched for in a predetermined data unit at a predetermined range of search locations set for the input past search target image and future search target image. In an image coding apparatus and method for performing bidirectional motion detection by performing comparison and compression-coding an input image based on a motion vector detected by the motion detection, a search is performed according to the motion of the input image. By arbitrarily setting the target image and the search location, even when the motion of the image is large, motion detection in at least one direction can be reliably performed, and the image quality of the encoded image can be improved by this amount.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of an image encoding device according to the present invention.

FIG. 2 is a block diagram showing a configuration of a motion vector detection unit of the image encoding device according to the present invention.

FIG. 3 is a schematic diagram illustrating a motion detection procedure in a normal state.

FIG. 4 is a schematic diagram used for explaining motion detection between fields.

FIG. 5 is a schematic diagram showing a state in which a search range is sufficient.

FIG. 6 is a schematic diagram for explaining a method of determining a search range.

FIG. 7 is a schematic diagram illustrating a motion detection procedure when a search range is insufficient.

FIG. 8 is a schematic diagram for explaining the resetting of a search range.

FIG. 9 is a schematic diagram for explaining the resetting of a search range.

FIG. 10 is a flowchart showing a motion detection processing procedure according to the present invention.

FIG. 11 is a schematic diagram illustrating a motion detection procedure according to another embodiment.

FIG. 12 is a schematic diagram for explaining motion detection in bidirectional prediction.

FIG. 13 is a schematic diagram illustrating a conventional motion detection procedure.

FIG. 14 is a schematic diagram for explaining a conventional problem.

[Explanation of symbols]

20 image encoding device, 21 motion vector detecting unit, 24, 29 arithmetic unit, 25 DCT unit, 26
... Quantization unit, 27... Inverse quantization unit, 28.
, 30, 51 ... frame memory, 31 ... motion compensation unit, 33 ... variable length coding unit, 62 ... control judgment unit, M
E1, ME2 ... Motion detecting unit.

Claims (6)

[Claims] 1. Bidirectional motion detection by comparing an input past search target image and a future search target image with a current image in predetermined data units at predetermined search locations set respectively. And an image encoding apparatus for compressing and encoding the input image based on the motion vector detected by the motion detection, wherein the search target image and the search location are arbitrarily set according to the motion of the input image. An image coding apparatus, comprising: 2. A search control unit comprising: a motion detection unit that performs the motion detection in two parts; a search target image and a search location in a first one of the two motion detections; 2. The image coding apparatus according to claim 1, further comprising setting means for setting the search target image and the search location in the second motion detection according to the motion of the input image. 3. The search control means according to claim 1, wherein said search location is determined in accordance with a difference of said predetermined data unit when a motion vector is detected among a plurality of fields based on said search target image and a distribution of said detected motion vector. The image encoding apparatus according to claim 1, wherein 4. Bidirectional motion detection by comparing the inputted past search target image and future search target image with a current image in a predetermined data unit at a predetermined range of search locations. And an image encoding method for compressing and encoding the input image based on the motion vector detected by the motion detection, wherein the search target image and the search location are arbitrarily set according to the motion of the input image. Image encoding method. 5. The image encoding method according to claim 1, wherein the motion detection is performed twice, and the search target image and the search location in the first motion detection of the two motion detections and the second are detected. The image encoding method according to claim 4, wherein the search target image and the search location in the motion detection are set according to the motion of the input image. 6. The image encoding method according to claim 1, wherein the search is performed in accordance with a difference of the predetermined data unit and a distribution of the detected motion vector when a motion vector is detected among a plurality of fields based on the search target image. The image encoding method according to claim 4, wherein a location is set.