JP3150446B2 - Image efficient coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-efficiency image coding method, and more particularly to a high-efficiency coding method for multi-view stereoscopic images using a parallax compensation prediction method.

FIG. 15 shows a conceptual configuration of a system disclosed in the NHK Technical Research Institute in 1992 for creating a multi-view stereoscopic image. First, a still or moving subject (octopus) 10 is Images are taken by a plurality of camera groups 11 whose positions are slightly shifted vertically and horizontally. Next, the image data obtained from the plurality of camera groups 11 is coded by the encoder 12 with high efficiency, multiplexed by the multiplexing unit 13 and then transmitted via the transmission path 14 or the like. After demultiplexing, decoding is performed and the result is displayed on the display 16.

As the display 16 on the output side, for example, a lenticular lens (when there is parallax only in the horizontal direction) or a fly's eye lens (when there is parallax in the vertical and horizontal directions) is used.

As an example, FIG. 16 shows an example in which the outputs from the respective cameras of five vertical eyes and five horizontal eyes when the octopus is set as a subject are displayed in an easy-to-understand manner. In this example,
There is also parallax in the vertical direction.

[0005] The use of a plurality of cameras whose positions are slightly shifted in the vertical and horizontal directions is such that the output from one camera is used as an input to one eye so that binocular parallax can be formed and stereoscopic vision can be performed. In addition, when a large number of cameras are used, even if a person who looks at the display 16 in the output system shakes his / her head, natural stereoscopic viewing can be performed. The development of a system for providing such a natural stereoscopic image is expected.

[0006]

2. Description of the Related Art When encoding such a stereoscopic image,
Conventionally, in a binocular stereoscopic video, a proposal has been made to realize high-efficiency coding by performing an image for one of the left and right eyes as a compression target image and performing parallax compensation using the other image as a reference image. High-efficiency coding in a form extended to multi-view stereoscopic video is being considered.

That is, as shown in FIG. 17, first, block matching with an input image on the left eye side is performed using an input image on the right eye side (an original image or a locally decoded image reproduced from an encoded image) as a reference image. To block matching unit 2
Take 1 Thereafter, the disparity compensation vector detection unit 22 detects the disparity compensation vector, the encoding unit 23 encodes the input image on the right eye side, and varies the right-eye-side coded corresponding block according to the disparity compensation vector. The difference signal from the input image on the left eye side is cut out by the delay unit 24, the obtained difference signal (prediction error signal) is coded by the coding unit 25, and the variable length coding unit (VLC) 26 , 27, and multiplexes them by a multiplexing unit 28 and sends them out to a transmission path.

FIG. 18 is a flowchart showing an example of a series of operations for the above-described block matching and parallax compensation vector detection.
In this example, as described above, the right eye input image F
As a reference image _R is for disparity compensation for left eye input image F _L.

Then, the search range is set to ± S with respect to the x direction.
± Sy for the x and y directions, Bx for the x direction, and By for the y direction (step S
0), an initial value SUM1 is obtained by summing the absolute value of the difference between the left and right input images as an absolute value error (or a square error) with respect to the entire block (step S1), and this is performed for each block. In this way (step S2), the parallax compensation vectors Vx and Vy are obtained by performing the comparison in step S3 and searching for the vector having the minimum value over the search range (step S4). ). This block matching is almost the same as the motion compensation prediction method in the high efficiency coding, and is a known method.

[0010]

In the above-described conventional parallax compensation method, the brightness of the (up / down) left and right cameras
It is not resistant to color changes (including noise and the like), and it is difficult to accurately detect a parallax compensation vector.

That is, since it is influenced by flicker, color and light reflection, and noise of each camera itself, it is difficult to accurately detect a parallax compensation vector. Also hindered efficient coding.

Accordingly, it is an object of the present invention to provide a high-efficiency image coding capable of detecting a parallax compensation vector as accurate as possible.

[0013]

According to the present invention, first, in order to solve the above problems, the present invention (part 1):
Obtaining as accurate a vector as possible with reference to surrounding disparity compensation vector information (two or more eyes), the present invention (part 2): also obtains information of disparity compensation vectors of other multiview images at the same time. Obtaining as accurate a vector as possible by reference (multiple eyes or more):

In other words, in the conventional method, in order to obtain an accurate disparity compensation vector, as an evaluation parameter,
Concept of using a disparity compensation vector of a block around the corresponding block, using a disparity compensation vector of a block at the same position in a past frame, or using a disparity compensation vector of another multiview image at the same time at the corresponding time There was no such thing.

The present invention (No. 1): FIG. 1 In the image high efficiency coding (No. 1) according to the present invention,
As a solution to this, as shown in FIG. 1 in principle, one image is set as a compression target image between a pair of synchronized camera images, and the other image is used as a reference image. A disparity compensation vector detection unit 2 that detects a disparity compensation vector of an image block, a memory 3 that stores the disparity compensation vector,
And a correction vector detecting unit 4 for obtaining a correction vector having an optimum characteristic from a corresponding block and its surrounding blocks among the disparity compensation vectors stored in the memory 3. Of the local decoded image is variably delayed by the variable delay unit 5, a prediction error with respect to the input image to be compressed is obtained and coded, and further variably coded and multiplexed together with the coded data of the reference image, and then transmitted. It is configured to be sent to the road.

The operation of the configuration shown in FIG. 1 will be described. First, an input image to be compressed (for example, an input image for the left eye) and a reference image (for example, an input image for the right eye) are synchronized and input to the block matching unit 1 to block the image. Perform matching. Therefore,
As shown by a dotted line, encoding may be performed before input to the block matching unit 1, and a locally decoded image of the reference image may be used as the reference image. Also, in this case, it is usual to appropriately add a delay unit in order to ensure synchronization of both input images.

The block matching result output from the block matching unit 1 is input to the disparity compensation vector detection unit 2. The disparity compensation vector detection unit 2 detects a disparity compensation vector of each image block according to a known evaluation function.

This will be described with reference to FIG. 2. In this example, the disparity compensation vectors of each image block of the camera image K22 obtained by using the camera image (screen) K23 as a reference image are vectors "A" to "T". Becomes Note that the illustrated arrows indicate vectors.

The parallax compensation vector thus detected is stored in the memory 3.

The correction vector detecting section 4 detects the parallax compensation vector of the corresponding block from the memory 3 and a block around the corresponding block, and preferably stores the corresponding block at the same position in the past stored in another memory 6. And a parallax compensation vector of a block around the corresponding block are input, and an optimum characteristic, that is, an accurate parallax compensation vector is detected in accordance with the evaluation function.

That is, in the example of FIG.
In order to obtain the correction vector of the disparity compensation vector of the vector “f” of 22, the disparity compensation vectors “a”, “i”, “u”, “o”, “ki”, “ke”,
“U” and “S” will be used in the correction vector detection unit 4.

The detected correction vector is preferably input to the memory 6 and also to the variable delay unit 5. The variable delay unit 5 delays and outputs the reference image based on the input vector. Since the output result is a predicted value for the image to be compressed, the prediction error is encoded by taking the difference. After this coding, variable length coding (VLC) is performed, multiplexed and transmitted to the transmission path.

Although FIG. 1 shows a case where a two-lens image is used as an input image, as shown in FIG. 2, even when there are multi-view images K11 to K44, two images among them are displayed. Similar processing can be performed in between. For example, the camera image K44 is used as a reference image, and the obtained camera image K34 is similarly processed using the vector “this”.
In order to obtain the correction vector of the disparity compensation vector of, the disparity compensation vectors “O”, “KA”,
“Ki”, “ke”, “sa”, “su”, “se”, and “so” are used in the correction vector detection unit 4. However, it is basically a two-lens image processing method.

The present invention ( No. 2): FIG. 3 The present invention (No. 2) is, as shown in principle in FIG. 3, a plurality of adjacent sets of three or more synchronized multi-view camera images F1 to Fn. One of the camera image pairs is used as a compression target image, and the other image is used as a reference image to detect a disparity compensation vector of a corresponding block based on block matching results obtained from each of the block matching units 1-0 to 1-n. A parallax compensation vector detecting unit 2-0 to 2-n to be performed, a memory 3 for storing the parallax compensation vector at the same time obtained by each of the parallax compensation vector detecting units 2-0 to 2-n, and a parallax of the memory 3 A correction vector detection unit 4 for obtaining a correction vector having an optimum characteristic among the compensation vectors, and a local decode image of the coded data of the reference image is obtained by each of the variable delay units 5-0 to 5-n using each correction vector. Variable delay After encodes seeking prediction error between the compression target input image, and further after multiplexed by a variable length coding together with the encoded data of the reference image, and configured to deliver to the transmission path.

Next, the operation of the configuration shown in FIG. 3 will be described. First, synchronized input images F1 to Fn are input to the respective block matching units 1-0 to 1-n. The reference images are F2 for F1, F3 for F2,.
It is Fn for Fn-1. The input image Fn is the last image not to be compressed.

Therefore, as in the case of FIG. 1, the input image F is not changed until it is input to each of the block matching units 1-0 to 1-n.
n, and a local decoded image of the input image Fn may be used as a reference screen. Further, the input image F1
To Fn-1 may be used as the terminal images. It is also common to insert a delay unit appropriately to ensure synchronization between the input images F1 to Fn-1 and Fn, as in FIG.

Each of the block matching units 1-0 to 1-
n, the block matching results output from the disparity compensation vector detection units 2-
The parallax compensation vector is input to each of 0 to 2-n and detected according to a known evaluation function. The detection results of these parallax compensation vector detection units 2-0 to 2-n are stored together in the memory 3.

This will be described with reference to FIG. 4. In this example, camera images K11 to K44 are input images, and camera images K12 for K11, K13,..., K43 for K12. The relationship is K44, and the parallax compensation vector of each camera image block is obtained in the same manner as in FIG. 2. However, the state shown in FIG.
, "P" of the second block from the left and only the second block from the top.
Note that the illustrated arrows indicate vectors.

The parallax compensation vector detected in this way is stored in the memory 3, and the correction vector detecting section 4 reads the parallax compensation vector of the corresponding block from the memory 3 and the surroundings of the corresponding block in the surrounding camera image. The disparity compensation vector of the block and the disparity compensation vector of the corresponding block at the same position in the past and the corresponding block in the surrounding camera image, which are preferably stored in another memory 6, are input, and are optimized according to the evaluation function. Characteristic, that is, accurate detection of the parallax compensation vector.

That is, in the example of FIG.
22 to obtain a correction vector of the disparity compensation vector “f”, the disparity compensation vectors “b”, “b” of the camera images K12, K32, and K42 in the same vertical direction with respect to the horizontal component.
Use “j” and “n” and use the parallax compensation vectors “e”, “g”, and “h” of the same horizontal camera images K21, K23, and K24 in the correction vector detection unit 4 for the vertical components. Becomes

The detected correction vector is preferably input to a memory 6 and a variable delay unit 5-0.
To 5-n. The variable delay unit 5 delays and outputs the reference image based on the input vector. Since the output result is a predicted value for the image to be compressed, the prediction error is encoded by taking the difference. After this coding, variable length coding (VLC) is performed, multiplexed and transmitted to the transmission path.

[0032]

5 to 14 show or explain embodiments of the correction vector detecting section 4 shown in FIGS. 1 and 3. Each embodiment will be described below. In the following embodiment, the correction vector detection unit 4 of FIG. 1 will be described unless otherwise specified. However, the case of FIG. 3 is basically the same.

Embodiment (Part 1): FIGS. 5 to 6 First, FIG. 5A shows the relationship between the block position in one camera image and the disparity compensation vector. , And the surrounding blocks are denoted by B11 to B33. According to the conventional method, the disparity compensation vectors obtained for the blocks B11 to B33 detected by the disparity compensation vector detection unit 2 in FIG. 1 are represented by SV11 (Vx11, Vy1).
1),..., SV33 (Vx33, Vy33).
Vx and Vy indicate a horizontal component and a vertical component, respectively.

FIG. 6 shows an embodiment of the correction vector detecting section 4 for detecting a correction vector for the image block shown in FIG. 5. In this embodiment, nine correction vectors are detected according to the flowchart shown in FIG. Horizontal component V of disparity compensation vector
The average value of x11 to Vx33 is taken. The circuit block diagram at this time is shown in FIG.

Although the vertical component Vy of the disparity compensation vector is not shown, it can be obtained in the same manner as in FIGS. 6A and 6B, and the disparity compensation vector Sv (Vx , Vy) are obtained.

When the above embodiment is applied to the present invention (part 2) in FIG. 3, blocks B11 to B33 are
(Blocks of vectors “a” to “p”). This will be described later with reference to FIGS.

Embodiment (Part 2): FIG. 7 This embodiment also shows an embodiment of the correction vector detecting section 4 for detecting a correction vector for the image block shown in FIG. 5, and this embodiment is also shown in FIG. As in the sixth embodiment, the horizontal components Vx11 to Vx33 of the nine parallax compensation vectors are set.
However, as compared with the embodiment of FIG. 6, a value Vx ′ obtained by averaging the respective parallax compensation vectors is obtained.
(Similarly for Vy ′) and the value of the
The difference is that weighting according to (0 <α <1) is added. A circuit block diagram at this time is shown in FIG.

Embodiment (Part 3): FIGS. 8 to 10 FIG . 8A shows the relationship between the block position in one camera image and the disparity compensation vector. In this embodiment, FIG. More blocks are used than in the embodiment, the corresponding block is B33, and the surrounding blocks are B11 to B55. Then, according to the conventional method, the disparity compensation vectors obtained for the blocks B11 to B55 detected by the disparity compensation vector detection unit 2 in FIG. 1 are SV11 (Vx11, Vy
11),..., SV55 (Vx55, Vy55).

Further, since the input image (or its local decoded image) itself is input as shown by the dotted line in FIG. 1, parameters indicating the properties of the block (flat portion, edge portion, fine pattern portion) can be changed by a known method. It is determined by using technology such as dispersion.

That is, as shown in the table of FIG. 8B, when the variance is represented by H, if the value is small, the flat portion is obtained, if the value is average, the edge portion is obtained, and if the value is large, the value is large. It is a fine pattern part. That is, when H is small, in a stereoscopic image,
It is a large flat thing in the distance such as the sky or the like (or a very fine pattern in the distance is blurred), or a flat thing near. If H is an average value, it is a substantially patterned object or a near-far boundary. When H is large, it indicates a large pattern in the distance or a fine pattern in the vicinity. The variance H is represented by the equation shown in FIG.

Taking this into consideration, the correction vector detecting section 4 of this embodiment (part 3) will be described with reference to the flowchart shown in FIG. 9 and the circuit block diagram of FIG. , The variance H is an average value in many cases.

Therefore, first, the average value Bij 'is calculated from the original image Bij.
And the variance H (Hij) is calculated, and then the average value H of the variance H is calculated.
AVR is obtained (step S1). Next, the reciprocal of the value obtained by adding 1 to the absolute value of the difference between each variance H and the average value HAVR is defined as the variance H '(step S2). This variance H ′ is a kind of weight that becomes smaller as the value goes away from the average value HAVR. Then, in order to normalize this weight, the average value of each variance H ′ is set to WAVR (step S3).
/ WAVR is normalized to W (step S4).

Accordingly, the horizontal component Vx to be obtained is Vx1
1 to Vx55 multiplied by each W (Step S
5) The vertical component Vy is obtained by multiplying each of Vy11 to Vy55 by each W (step S6).

Embodiment ( No. 4): FIG. 11 In this embodiment, the disparity compensation vector as the correction vector two frames past is B2V, the disparity compensation vector as the correction vector one frame past is B1V, and the current correction vector is B2V. Is set as SV (step S11), and compared with the displacement of the past disparity compensation vector,
If the displacement of the current disparity compensation vector is significantly different, there is a possibility that there is a mismatch, and therefore, an attempt is made to correct this.

However, if any one of the disparity compensation vectors B1V, B2V, and SV is 0 (step S12), there is a possibility that the object has disappeared or newly appeared in the block. The process ends without adding (step S13).

In step 12, if any of the disparity compensation vectors is not 0, the displacement B2V-B1V of the past disparity compensation vector and the displacement B2 of the current disparity compensation vector
1V-SV is calculated (Step S14), and it is determined whether or not the absolute value of the difference between the absolute value of the displacement of the past disparity compensation vector and the current absolute value of the displacement of the disparity compensation vector is smaller than the threshold value TH ( Step S15) If the value is smaller, it is determined that the displacement is within the allowable displacement, and the process ends without making any change to the SV (Step S16).

Otherwise, the coefficient α is added to the value obtained by adding the past displacement and the disparity compensation vector one frame past.
, And the current disparity compensation vector is also weighted by multiplying by a coefficient (1−α), the two are added (step S17), and the disparity compensation vector at the time of mismatch is changed. .

Embodiment (Part 5): FIGS. 12 to 13 As described above, the above embodiments (Part 1) to (Part 4) are also applicable to the present invention (Part 2) of FIG. FIG. 12 shows an optimal application example. In this embodiment, depending on the arrangement of the cameras, the disparity compensation vectors Vx in the X direction of the respective blocks of the cameras in the vertical column are substantially the same, and the horizontal Based on the fact that the parallax compensation vectors Vy in the Y direction of each block of the camera are substantially the same, an attempt is made to detect a parallax compensation vector that is a correction vector.

For this reason, as shown in FIG. 12, blocks at the same position in each of the camera images K11 to K55 (in the case of 25 eyes of 5 × 5) are respectively defined as KB11 to KB5.
5. The block disparity compensation vector SV11 (Vx11,
Vy11) to SV55 (Vx55, Vy55).

Then, the average value of Sxij in one column is represented by Sx,
The average value of Syij in one horizontal row is defined as Sy, and FIG.
As shown in (b), a disparity compensation vector is obtained as a correction vector obtained from (Sx, Sy).

According to FIGS. 13A and 13B, the disparity compensation vector SV obtained by the calculation by the correction vector detection unit 4 in FIG. 3 has the same vertical or horizontal vector.
Therefore, a common parallax compensation vector is provided between camera images in the same column or row in FIG. 12 and the variable delay unit 5-0 in FIG.
To 5-n, as shown in FIG.
A method of leaving some Syij information is also conceivable.

This is because when the position of the camera covers a wide range, the parallax compensation vector may change due to a change in the unevenness of the object.

Embodiment ( No. 6): FIG. 13 (c), ie, 0 <α <1, and Sx × α + Sxij × (1-α)
Is calculated in the X-direction parallax compensation vector, Sy × α + Sy
ij × (1−α) is obtained by calculating a disparity compensation vector in the Y direction,
Thus, all the parallax compensation vectors can be set to different values and provided to the variable delay units 5-0 to 5-n.

Embodiment ( No. 7): FIG. 14 In order to make the disparity compensation vector obtained by the above embodiment more complete, in this embodiment, an evaluation function is further provided, and a correct disparity compensation vector ( The correction vector according to the present invention) is compared with a disparity compensation vector obtained by a conventional method (for example, the disparity compensation vector “f” in the camera image K22 in FIG. 2), and an image having a better characteristic is selected. is there. This is effective in the case of an orthogonal transform system in which encoding itself dislikes high frequency components (impulses and the like).

In FIG. 14, first, a prediction error block which is a difference between a prediction block obtained from an original disparity compensation vector and a block to be encoded is represented by YB1, and a disparity compensation vector obtained based on the present invention is used. A prediction error block which is a difference between the obtained prediction block and a block to be encoded is set to YB2 (step S21).

For each pixel in the prediction error block YB1, the number of pixels whose absolute value exceeds the threshold value TH is K1, and for each pixel in the prediction error block YB2, the absolute value of the pixel exceeds the threshold value TH. K
2 (step S22).

If K1> K2 (step S23), the disparity compensation vector obtained according to the present invention is used (step S24), otherwise, the disparity compensation vector obtained by the conventional method is used.

[0058]

As described above, according to the image efficient coding method according to the present invention, the information of the surrounding disparity compensation vector or the information of the disparity compensation vector of another multi-view image at the same time is also referred to. In order to obtain the most accurate disparity compensation vector as a correction vector, it is not only necessary to perform block matching between the encoded screen and the reference screen as in the past, but also to obtain information about the surroundings and images from other cameras. , A relatively accurate disparity compensation vector can be obtained, and therefore, coding efficiency and visual characteristics can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a principle configuration (No. 1) of an image efficient coding system according to the present invention.

FIG. 2 is a camera image diagram for explaining the operation of the principle configuration (No. 1) of the image efficient coding system according to the present invention.

FIG. 3 is a block diagram showing the principle configuration (No. 2) of the image efficient coding system according to the present invention.

FIG. 4 is a camera image diagram for explaining the operation of the principle configuration (No. 2) of the image efficient coding system according to the present invention.

FIG. 5 is a diagram for explaining a relationship between a block position of a camera image and a disparity compensation vector used in the present invention.

FIG. 6 is a diagram showing an embodiment (part 1) of a correction vector detection unit used in the present invention.

FIG. 7 is a diagram showing an embodiment (part 2) of a correction vector detection unit used in the present invention.

FIG. 8 is a diagram for explaining a relationship between a block position of a camera image used in the present invention, a disparity compensation vector, and properties thereof.

FIG. 9 is a flowchart of an embodiment (part 3) of a correction vector detection unit used in the present invention.

FIG. 10 is a circuit block diagram of an embodiment (part 3) of a correction vector detection unit used in the present invention.

FIG. 11 is a flowchart illustrating an example (part 4) of a correction vector detection unit used in the present invention.

FIG. 12 is a diagram for explaining a relationship between a multi-view camera image used in the present invention (part 2), a disparity compensation vector, and properties thereof.

FIG. 13 is a flowchart of an embodiment (parts 5 and 6) of the correction vector detection unit used in the present invention.

FIG. 14 is a flowchart of an example (part 7) of the correction vector detection unit used in the present invention.

FIG. 15 is a block diagram showing a general configuration of a multiview stereoscopic video system.

FIG. 16 is a diagram showing an example of a multi-view (5 × 5) camera output.

FIG. 17 is a block diagram showing a conventional parallax compensation method.

FIG. 18 is a flowchart illustrating block matching and parallax compensation vector detection.

[Explanation of symbols]

 1 Block Matching 2, 2-0 to 2-n Parallax Compensation Vector Detector 3, 6 Memory 4 Correction Vector Detector 5, 5-0 to 5-n Variable Delay Unit .

Continuing from the front page (72) Inventor Kiichi Matsuda 1015 Kamikodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Prefecture Inside Fujitsu Limited (56) References JP-A-4-196998 (JP, A) JP-A-4-178095 (JP, A) JP-A-2-131697 (JP, A) JP-A-2-13094 (JP, A) JP-A-2-100571 (JP, A) JP-A-2-100590 (JP, A) JP-A-2 -50689 (JP, A) JP-A-1-202093 (JP, A) JP-A-1-114283 (JP, A) JP-A-64-64489 (JP, A) JP-A-61-206395 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) H04N 13/00-15/00

Claims (7)

(57) [Claims] 1. A parallax compensation vector of each image block based on a block matching result obtained from a block matching unit (1) using one image as a compression target image and the other image as a reference image between a pair of synchronized camera images. A disparity compensation vector detection unit (2) for performing detection, a memory (3) for storing the disparity compensation vector, and a disparity compensation vector stored in the memory (3). A correction vector detecting section (4) for obtaining a correction vector having an optimum characteristic, wherein a local decoding image of the encoded data of the reference image is variably delayed by the variable delay section (5) using the correction vector, Find and encode the prediction error with the input image,
Further, a high-efficiency image coding method characterized in that variable-length coding is performed together with the coded data of the reference image, multiplexed, and then transmitted to a transmission path. 2. Synchronized three or more multi-view camera images (F1
~ Fn), a block obtained from each block matching unit (1-0 to 1-n) using one image as an image to be compressed and the other image as a reference image between a plurality of pairs of camera image pairs adjacent to each other. The disparity compensation vector detection unit (2-0 to 2-n) that detects the disparity compensation vector of the corresponding block based on the matching result, and the same time obtained by each disparity compensation vector detection unit (2-0 to 2-n) A memory (3) for storing a disparity compensation vector of the same type, and a correction vector detector (4) for obtaining a correction vector having an optimum characteristic among the disparity compensation vectors of the memory (3). After locally delaying the locally decoded image of the encoded data of the image in each of the variable delay units (5-0 to 5-n), a prediction error with respect to the input image to be compressed is obtained and encoded, and the code of the reference image is further encoded. After multiplexing with variable-length coding together with the encoded data, Image High efficiency coding scheme, wherein. 3. The high-efficiency image coding method according to claim 1, wherein said correction vector detecting section uses an average value of the respective disparity compensation vectors as the characteristic. 4. The high-efficiency image coding method according to claim 1, wherein said correction vector detecting section uses an average value of variance as said characteristic. 5. In addition to the memory (3), another memory (6) for storing a past disparity compensation vector is provided, and the correction vector detecting unit (5) also stores the past disparity compensation vector. The high-efficiency image coding method according to claim 1, wherein the image is included in a correction vector detection target. 6. The correction vector detecting section (4) determines whether the past correction vector and the current disparity compensation vector are not “0”, and the absolute value of the displacement of the past disparity compensation vector and the current disparity compensation vector are 6. The method according to claim 5, wherein when the absolute value of the difference from the absolute value of the displacement is larger than the threshold value, the past correction vector and the correction vector that takes into account the amount of the displacement are used, and the remaining correction vector is the current disparity compensation vector. Described image high efficiency coding method. 7. When the correction vector detection unit (4) calculates the horizontal component of the disparity compensation vector to detect the correction vector, the correction vector detection unit (4) calculates the horizontal disparity compensation vector of the camera image in the vertical direction of the corresponding block. The image according to any one of claims 1 to 6, wherein when calculating a vertical component of the disparity compensation vector, information of a vertical disparity compensation vector of a camera image in a horizontal direction of the corresponding block is used. High efficiency coding method.