JP2957851B2 - Image stabilization 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 camera shake correction method, and more particularly to a camera shake correction method used for, for example, a consumer camera-integrated VTR.

[0002]

2. Description of the Related Art An example of a method for detecting a shake component of an image pickup apparatus has been disclosed by Matsushita Electric Industrial Co., Ltd. at the 20th Conference on Image Engineering in 1989. This method
An image pickup apparatus is obtained from image information using a motion vector obtained from a representative point matching method described in Japanese Patent Application Laid-Open No. 61-201381 [H04N7 / 137] published on September 6, 1986. This is to detect the shake component of the image. In this presentation, an attenuation coefficient is introduced in a method of correcting camera shake based on a motion vector obtained from image information. Also, four detection areas are arranged on the screen, and thus four partial motion vectors are obtained from one screen.

Here, a method for correcting camera shake based on the obtained partial motion vector will be described. The average of the four partial motion vectors described above is used as the entire motion vector between the fields. Assuming that the entire motion vector is V _n , the integral vector _Sn is represented by Expression 5.

[0004]

S _n = k · S _n-1 + V _n where k is called an attenuation coefficient and is a small number smaller than 1. In this way, using the integral vector S _n thus obtained, for example, as shown in FIG. 6, by moving the cut-out position of the image, corrects the image blur based on the camera shake.

[0005]

[SUMMARY OF THE INVENTION] However, in this correction method, even when the whole motion vector V _n is 0, an integral vector S _n goes to center to be closer to 0 with time. That is, there is a problem that a phenomenon in which the image moves even when the camera is not moved appears, and the corrected image becomes difficult to see.

[0006] Therefore, a main object of the present invention is to provide a camera shake correction method capable of better correcting camera shake.

[0007]

SUMMARY OF THE INVENTION The first invention is a camera shake correction method for correcting the blur of the image according to the integration vector, the image based on the entire motion vector V _n of the image detected between fields integral vector as a correction pattern to determine the S _n,

[0008]

S _n = K ₁ · S _{n -1} + V _n S _n : n-th field integration vector S _n-1 : n-1 field integration vector V _n : n-1 field and n-th field A first correction pattern represented by a small number of attenuation coefficients K ₁ : 1 or less, and

[0009]

Equation 7] _{S n = S n-1 -K} 2 · | V n | K 2: which comprises using the second correction pattern shown with a small number of damping coefficient of 1 or less, an image stabilization process.

And,

Equation 8] _{S n = K 3 · S n} -1 + V n K 3: 1 decimal attenuation coefficients, however, is V _n, 1 pixel> | represented by S _n = S _n-1 when | V _n This is a camera shake correction method further using a third correction pattern to be used.

Further, when at least one of panning and tilting is detected in the first correction pattern, a transition is made to the second correction pattern, and when it is detected that there is little camera shake, a transition is made to the third correction pattern. 2
In the correction pattern, when it detects at least one of the end of panning and tilting, the larger the integral vector S _n transitions to the third correction pattern, a transition to the first correction pattern Smaller integral vector S _n, third
When at least one of panning and tilting is detected in the correction pattern, a transition is made to the second correction pattern, and when the integration vector _Sn is small, the first correction pattern is used.
This is a camera shake correction method that transitions to a correction pattern.

In the second invention, the direction of the entire motion vector V _n is

S _n = S _n-1 −K ₂ · | V _n | K ₂ : The direction of the whole motion vector V _n immediately before transition to the correction pattern represented by a small number of attenuation coefficients of 1 or less is obtained. This is a camera shake correction method for detecting the end of at least one of panning and tilting when a field continues for a third threshold or more.

[0013]

First, a first correction pattern is set as an initial state. In the first correction pattern, the integration vector S _n
Or transition by the overall motion vector V _n in the second correction pattern by detecting the panning or tilting,
Transitioning to a third correction pattern by detecting that there is little camera shake by the overall motion vector V _n.

[0014] In the second correction pattern, when detecting the end of panning or tilting by the overall motion vector V _n, the larger the integral vector S _n transitions to the third correction pattern, first the smaller the integral vector S _n Transition to the correction pattern. In a third correction pattern, by detecting the panning or tilting the integral vector S _n or the whole motion vector V _n transitions to the second correction pattern, a transition to the first correction pattern Smaller integral vector S _n.

[0015] Thus by changing the correction patterns in accordance with the state obtains an integral vector S _n, for image stabilization.

[0016]

Effects of the Invention According to the present invention, by changing the method of calculating the integral vector S _n according to the state, Midzura of the image is eliminated by the centering after the panning.
For example, when centering is almost completed during panning, the centering period after panning can be shortened. Also, since the video camera is moving during panning, centering can be performed during panning without the user's notice. Therefore, camera shake correction can be performed better than before.

The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A video camera 10 of this embodiment shown in FIG.
Includes a solid-state imaging device 12 such as a CCD that converts a light signal from a subject (not shown) input from a lens 14 into an electric signal. An electric signal from the solid-state imaging device 12 is input to a camera circuit 16. As is well known, the camera circuit 16 includes a sample and hold circuit, and the solid-state imaging device 1
The sample and hold of the electric signal from 2 is performed. The level of the sampled and held electric signal is adjusted by the AGC, and a synchronizing signal is further added by a synchronizing signal adding circuit. Thus, the camera circuit 16 converts the image signal from the solid-state imaging device 12 into an analog video signal. This analog video signal further comprises A
The digital video signal is converted by the / D converter 18. The digital video signal is provided to the motion detection circuit 20. As the motion detection circuit 20, for example, an LSI “L7A0948” manufactured by Sanyo Electric Co., Ltd. is used.
Under the control of a memory control circuit 22 included in the same LSI constituting the motion detection circuit 20, digital video signals are written in a field memory 24 in a field sequence.

The motion detection circuit 20 uses the well-known representative point matching method, for example, to obtain the highest correlation (the minimum correlation value) for each of the four detection areas A, B, C and D shown in FIG. The position of one point and four points around the point and the respective correlation values are calculated. The position data and the correlation value data from the motion detection circuit 20 are provided to the microcomputer 26.

More specifically, referring to FIG. 2, motion detecting circuit 20 includes an input terminal 28 for receiving a digital video signal from A / D converter 18, and the digital video signal input from input terminal 28 is filtered. The signal is supplied to a representative point memory 32 and a subtraction circuit 34 through 30. The filter 30 is a kind of digital low-pass filter,
It is used to improve the S / N ratio and ensure sufficient detection accuracy with a small number of representative points. The representative point memory 32 extracts a plurality of representative points in each of the detection areas A to D shown in FIG.
Then, 30 representative points are extracted), and the position data and the luminance data are stored. Each detection area 42 (FIG. 4) formed by dividing into 30 is composed of, for example, 32 pixels × 16 rows.

The subtraction circuit 34 calculates the luminance data of the representative point of the previous field given from the representative point memory 32 and the input terminal 2.
8 is subtracted from the luminance data of all the pixels in the current field given by 8 and the absolute value is obtained. That is, a luminance difference is obtained between the luminance data of the current field and the luminance data of the previous field. The obtained luminance difference is given to the accumulation circuit 36. The accumulative addition circuit 36 accumulatively adds (30 in this embodiment) the luminance differences obtained for the pixels at the same position in each detection area 42 in the same detection area, and outputs correlation value data. The correlation value data is supplied to an arithmetic circuit 38, which calculates a minimum correlation value and an average correlation value for each of the detection areas A to D, and calculates the position data of the pixel indicating the minimum correlation value in each of the detection areas. Determined for each AD. The minimum correlation value, average correlation value and position data obtained in this way are output from the output terminal 40 to the microcomputer 2 described above.
6 given. However, the calculation of such a correlation value is
This is executed by the aforementioned LSI “L7A0948”.

In the microcomputer 26,
Based on the position data and the correlation value data, a motion vector of the entire screen, that is, the image field 44 (FIG. 3) (hereinafter, simply referred to as “entire motion vector”) is calculated. First, a shift of the pixel having the minimum correlation value with respect to the representative point is obtained based on the position data of the pixel having the minimum correlation value, and the shift is set as a partial motion vector. In order to improve the detection accuracy of the partial motion vector, interpolation is performed using the correlation values of four pixels around the pixel having the minimum correlation value,
The position data of the pixel having the minimum correlation value is calculated.

The microcomputer 26 determines that the value obtained by dividing the average correlation value by the minimum correlation value is larger than a certain threshold value.
Asked whether and average correlation value is detected to or greater than a predetermined value for each of the detection areas A-D, false detection by an object such as a portion motion vector movement other than camera shake from the detection areas A-D Without detecting whether each detection area AD
Is determined as a valid area. If the value obtained by dividing the average correlation value by the minimum correlation value is greater than a certain threshold value and the average correlation value is equal to or greater than a predetermined value, the detection area is determined to be an effective area.

Specifically, whether or not an area is an effective area is determined as follows. First, when the contrast of the screen is low, since the luminance difference is small, the correlation value is small. For example, when the entire screen is white, the correlation value becomes small. In such a case, since the reliability is lost, the average correlation value ≧
It is determined to be valid when the value is a predetermined value. The predetermined value is determined by an experiment.

When there is a moving object in the detection area, the correlation value differs between the portion occupied by the moving object and the portion not occupied, and the portion occupied by the moving object takes various correlation values. (The correlation degree is low). Therefore, when there is a moving object in the detection area, there is a high possibility that the minimum correlation value increases, and there is a possibility that a partial motion vector in the detection area is erroneously detected. When the partial motion vector is erroneously detected, the entire motion vector is erroneously detected. However, when the average correlation value is large, it is reliable even if the minimum correlation value is somewhat large. On the other hand, when the average correlation value is small, the minimum correlation value cannot be relied on unless it is smaller. Therefore, specifically, (average correlation value)
It is determined to be valid when / (minimum correlation value)> 7, and a partial motion vector of a detection area that does not satisfy this condition is not used, thereby preventing the above-described adverse detection.

On the basis of these two conditions, it is determined whether the detection area is an effective area. Then, an average of partial motion vectors of valid regions to which the motion amount ie total motion vector V _n between fields. The overall motion vector indicates the amount of motion between fields and its direction. In the microcomputer 26, then the integration vector S _n is calculated. The integration vector _Sn indicates the amount of shift of the extraction area 48 from the center of the image field 44, that is, the amount of correction and its direction. Correction pattern to calculate the integral vector S _n is changed depending on the state. The state is shown in Table 1.
As shown in (1), there are four states, State 1 to State 4, which are determined by the state of the video camera 10 (for example, panning and tilting). The data in Table 1 is stored in a memory (not shown) of the microcomputer 26.

[0027]

[Table 1]

[0028] In Table 1, "state 1" is the same as the method for obtaining the conventional integral vector S _n (number 5). “State 2” is a mode in which the integral vector _Sn is brought close to 0 regardless of the direction of the overall motion vector V _n , that is, a mode in which so-called forced centering is performed. "State 3" and "State 4"
Is the same as obtaining the conventional integral vector S _n,
When the entire motion vector V _n is smaller than one pixel, the integration vector S _n the same as the previous field. That is, for example, the centering operation is stopped when the movement is small, such as being fixed to a tripod. Since the attenuation coefficient (0.992) in “state 4” is smaller than the attenuation coefficient (0.996) in “state 3”, centering is faster in “state 4”.

Here, in Table 1, 0.
Although 996 and 0.992 were used, it is sufficient if the attenuation coefficient is a small number of 1 or less. The attenuation coefficient has a frequency characteristic,
Further, by changing the damping coefficient, the speed of centering can be changed. Such integral vector S _n, which is determined in the given memory control circuit 22, the memory control circuit 22 determines the read start address of the field memory 24 on the basis thereof, stored from the address in the field memory 24 Read the digital video signal. That is, the memory control circuit 2
2, according to the integration vector S _n calculated by the microcomputer 26 moves the extraction area 48 which is formed by a digital video signal of the field memory 24.

However, if the digital video signal read from the field memory 24 is not used, the extraction area 4
8 cannot be moved, so the electronic zoom circuit 46 is used.
Referring to FIG. 5, an electronic zoom circuit 46 (FIG. 1) sets an extraction area 48 in which the image is enlarged according to the zoom magnification for the size of image field 44. The position of the extraction area 48 can be freely moved within the range of the image field 44 by changing the read start address of the field memory 24. And
In order to obtain a video signal of the entire image field 44 based on the extracted digital video signal, the image is enlarged using interpolation based on the digital video signal read from the field memory 24.

Thus, the image field 44
The electronic zoom circuit 46 (FIG. 1) electronically zooms an image of an arbitrary extraction area 48 within the area, thereby forming a correctable range 50 corresponding to the difference between the image field 44 and the extraction area 48. When camera shake occurs in the video camera 10 as shown in FIG. 6 according to the vibration of the hand of the person who operates the video camera 10, the video camera 10
When the target person exists at the lower left in the image field 44 (FIG. 6).
There are cases where the target person exists in the upper right of the image field 44 (lower in FIG. 6) and the like. Therefore,
By moving in accordance with the integral vector S _n of the extraction area 48 for each field was calculated by the microcomputer 26, the extraction area 4 as shown in FIG. 6 right
8 just fits the target person.

The digital video signal output from the electronic zoom circuit 46 is converted into an analog signal by the D / A converter 52 and output from the output terminal 54. Such a video camera 10 operates as shown in FIGS. 7 to 10 for each field. First, in step S1, state 1 is set as an initial value. For setting the status, for example, a status flag is used. Then
In step S3, the validity / invalidity of the partial motion vector of each detection area is determined based on the cumulative addition result. In step S5, it is determined whether there is a valid region, in step S7, if the effective region, an average of partial motion vectors of valid regions is it a whole motion vector V _n between fields. On the other hand, if it is determined in step S5 that there is no valid area, then in step S9, 1
The total motion vector of the previous field, the whole motion vector V _n of the current field. After the processing of step S7 and step S9, the process proceeds to step S11, and step S1
In step 1, it is determined whether the state is state 1. If the state is 1 in step S11, then in step S13, S _n =
0.996 * S _n-1 + V _n gives the integral vector S _n
Is calculated.

[0033] Then, in step S15 shown in FIG. 8, if the facing 30 fields the same direction is continuously total motion vector V _n, the state in step S17 1
To state 2. In step S15, if not oriented in the same direction 30 fields or more contiguous whole motion vector V _n, the process proceeds to step S19. Step S1
In 9, if the integration vector _Sn becomes large and becomes the correctable range 50 or more, and if the cutout position of the extraction area 48 is 48 fields or more continuously to the end of the image field 44, the state transits to the state 2 in step S 17. Then, the process proceeds to step S21, where the integration vector S _n
, The cutout position of the extraction area 48 is calculated.

In step S19, the extraction area 48
If the cutout position is not more than 48 fields continuously to the end of the field memory 44, the process proceeds to step S23. In step S23, the integral vector _Sn is 3
If it is larger than 0 pixels, the process proceeds to step S25. In step S25, the smaller field than one pixel total motion vector V _n is, in step S27 if continued for 10 fields in succession, and the transition from state 1 to state 3, the process proceeds to step S21.

In step S23, the integral vector S
_{If n} is 30 pixels or less, the process proceeds to step S29. If an integral vector S _n is 3 pixels or more in step S29, the process proceeds to step S31. In step S31, the correction amount V _n is smaller fields than one pixel, if continued for 10 fields in succession deviates from the state 1 to state 4 at step S33, the process proceeds to step S21. If the conditions of steps S25, S29, and S31 are not satisfied, the state is left at step S35, and the process proceeds to step S21.

If it is determined in step S11 that the state is not the state 1, the process proceeds to step S37 shown in FIG. Step S37
In the case of state 2, in step S39,
_{S n = S n-1 -} (1/4) · | V n | by calculating the integral vector S _n. Then, in step S41, in step S43 if the integral vector S _n is smaller than 0 pixels, as S _n = 0, the process proceeds to step S45. In step S41, the integration vector S _n is 0
If the number is equal to or larger than the pixel, the process directly proceeds to step S45. In step S45, the whole motion vector V _n is the field that the whole becomes a vector of the motion vector V _n in the opposite direction immediately before the state 2, if reached in succession 10 field,
Proceed to step S47. In step S47, the if the integration vector S _n is smaller than 3 pixels, step S49
Transitions from state 2 to state 1 in step S21.
Proceed to. When the condition of step S45 is not satisfied, the state is left at step S51, and the process proceeds to step S21. If the condition of step S47 is not satisfied, the state transits from state 2 to state 3 in step S53, and proceeds to step S21.

On the other hand, if it is not state 2 in step S37, the process proceeds to step S55. In step S55, if the state is 3, the process proceeds to step S57. In step S57, the smaller than one pixel total motion vector V _n, the process proceeds at step S59, the in step S61 as S _n = S _n-1. In step S57, the long overall motion vector V _n is 1 pixel or more, the step S
In 63, the integral vector _Sn is calculated according to S _n = 0.996 · S _n-1 + V _n , and the process proceeds to step S61. In step S61, the entire motion vector V
_{If n} continuously points in the same direction for 30 fields, the state transits from the state 3 to the state 2 in step S65, and proceeds to step S21. Does not satisfy the condition of step S61, in step S67, the integration vector S _n becomes large, the extraction position of the extraction area 48 is continuous with the edge of the image field 44 if 48 or more fields, the process proceeds to step S65. On the other hand, if not more than 48 fields continuously in step S67,
Proceed to step S69. In step S69, if the integration vector S _n is 30 pixels or less, steps S70
Proceed to. In step S70, the larger integral vector S _n is from 3 pixels, state 3 in step S71
To state 4 and proceed to step S21. In step S70, the long integration vector S _n is 3 or pixels or less, the transition from state 3 to state 1 at step S72, the process proceeds to step S21. In step S69,
If an integral vector S _n is larger than 30 pixels, in the state 3 in step S73, the process proceeds to step S21.

If the state is not the state 3 in the step S55, it is determined that the state is the state 4, and the process proceeds to the step S75 shown in FIG. In step S75, the overall motion vector V _n
Is smaller than one pixel, in step S77, S _n
= S _n-1 and the process proceeds to step S79. Step S7
In 5, if the whole motion vector V _n is 1 pixel or more, in step S81, S _n = 0.992 · S
seeking integral vector S _n by _n-1 + V _n, the process proceeds to step S79. In step S79, the whole motion vector V _n is, if directed to 30 fields the same direction continuously, a transition from state 4 to state 2 at step S83, the process proceeds to step S21.

On the other hand, if the condition in step S79 is not satisfied, the flow advances to step S85. In step S85, the integration vector S _n becomes large, the extraction area 48
If the cutout position is 48 fields or more continuously at the end of the image field 44, the state transits from the state 4 to the state 2 in step S83, and proceeds to step S21. If the condition of step S85 is not satisfied, the process proceeds to step S87. In step S87, if the integration vector S _n is 3 or pixels or less, a transition from state 4 to state 1 at step S89, the process proceeds to step S21.

In step S87, the integral vector S
_{If n} is larger than three pixels, the process proceeds to step S91. In step S91, the long is 30 pixels or more integral vector S _n, in step S93, the transition from state 4 to state 3, the process proceeds to step S21. Step S91
In, smaller than 30 pixels integral vector S _n, step in the state 4 at step S95 S
Proceed to 21.

In the video camera 10 operating as described above, as a result of the state transition, the state of the video camera 10 is set to “state 1” when the camera is simply shaken, to “state 2” when panning or tilting is performed, and to a tripod. "State 3" or "State 4" when the camera shake is small as fixed to
Transitions to. Therefore, the centering is forcibly performed during the panning or the tilting, and the difficulty in viewing the image due to the centering after the panning or the tilting, which is conventionally seen, can be eliminated. Also, even if the centering does not end during panning or tilting, the entire motion vector V _n
Is small, the state is changed to “state 3” or “state 4”, and the centering operation can be stopped with S _n = S _n−1 , so that it is possible to eliminate the difficulty in viewing the image due to the centering.

FIG. 1 shows the relationship between the number of fields and the amount of movement of the video camera 10 and the like when the video camera 10 is panned from a state of camera shake and then fixed.
It is shown in FIG. In FIG. 11, a solid line A indicates the video camera 10.
The dashed line B indicates the movement amount of the extraction area 48 after the camera shake correction by the video camera 10 of this embodiment, and the broken line C indicates the extraction area 48 after the camera shake correction by the conventional technique. Indicates the amount of movement.

First, when there is a camera shake as shown in a period a, the camera shake is corrected in the state 1. When the camera shake as shown in range b enters the panning, but initially is corrected, an integral vector S _n exceeds the correctable range, cut-out position of the extraction area 48 is on the edge of the image field 44. When this motion continues for 48 fields, the state is changed to state 2. Then, S _n = S _n-1 −
(1/4) · | V _n | corrected by the motion after the correction of the video camera 10 approaches rapidly to the movement of the video camera 10 itself (broken line B in the solid line A rapidly approaches). That is, forcible centering in which the cutout position of the extraction area 48 moves to the center of the image field 44 is performed quickly. Then, as shown in a period C, and the video camera 10 panning is completed is in a fixed state, because the integral vector S _n according to the centering is small, a transition to state 1 again. When the video camera 10 is fixed, the whole motion vector V _n is small, to stop centering the transition to state 4 after 10 field.

Therefore, in the prior art, centering was observed after the completion of panning (see the two-dot chain line C). However, in the video camera 10 of this embodiment, the centering was almost completed during the panning, and the centering was not completed. However, since it is detected that the camera shake is small, such as being fixed to a tripod, according to this embodiment, it becomes difficult to view the image due to centering after panning.

The motion vector may be obtained by using, for example, an angular velocity sensor instead of using the representative point matching method. In the above-described embodiment, the case where the video camera 10 operates for each field has been described. However, the present invention is not limited to this, and the video camera 10 may operate for each frame.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention.

FIG. 2 is a block diagram illustrating a motion detection circuit.

FIG. 3 is an illustrative view showing a principle of electronic zoom and showing a detection area in an image field;

FIG. 4 is an illustrative view showing a principle of electronic zoom and showing a representative point and a sampling point in a detection area.

FIG. 5 is an illustrative view showing a principle of camera shake correction;

FIG. 6 is an illustrative view showing each block in an image field to which a representative point matching method is applied;

FIG. 7 is a flowchart showing the operation of this embodiment.

FIG. 8 is a flowchart showing an operation subsequent to FIG. 7;

FIG. 9 is a flowchart showing an operation continued from FIG. 8;

FIG. 10 is a flowchart showing an operation continued from FIG. 9;

FIG. 11 is a graph showing a state of camera shake correction in comparison with the related art.

[Description of Signs] 10 ... Video camera 20 ... Motion detection circuit 22 ... Memory control circuit 24 ... Field memory 26 ... Microcomputer 42 ... Detection area 44 ... Image field 46 ... Electronic zoom circuit 48 ... Extraction area

Continuation of front page (58) Field surveyed (Int.Cl. ⁶ , DB name) H04N 5/232

Claims (6)

(57) [Claims] An image stabilization method for correcting image blur according to an integral vector, comprising: a motion vector V _{n of} an image detected between fields;
To the correction pattern for determining the integral vector S _n of the image based on, Equation 1] _{S n = K 1 · S n} -1 + V n S n: n th field of an integral vector S _n-1: n-1 th field V _n : the whole motion vector of the image between the (n−1) th and n-th fields K ₁ : a first correction pattern represented by a small number of attenuation coefficients of 1 or less, and S _n = _{S n-1 -K 2 · |} V n | K 2: which comprises using the second correction pattern shown with a small number of damping coefficient of 1 or less, image stabilization method. 2. A Equation 3] _{S n = K 3 · S n} -1 + V n K 3: 1 decimal attenuation coefficients, however, is V _n, 1 _{pixel> | V n | S n =} S when
_The method according to claim 1, further comprising using a third correction pattern indicated by _n-1 . 3. A transition to the second correction pattern when at least one of panning and tilting is detected in the first correction pattern, and a transition to the third correction pattern when it is detected that the camera shake is small. , in the second correction pattern, when it detects at least one of the end of panning and tilting, a transition to the third correction pattern larger the integral vector S _n,
Smaller integral vector S _n transitions to the first correction pattern, in the third correction pattern, a transition to the second correction pattern when detecting at least one of panning and tilting, when the integration vector S _n is small The method according to claim 2, wherein a transition is made to the first correction pattern. Wherein when the integral vector S _n is equal to or higher than the correctable range field fields oriented in the same direction the whole motion vector V _n or images when a continuous first threshold value or more is continuously smaller than the second threshold value, panning and tilting The camera shake correction method according to claim 3, wherein at least one of the touching is detected. 5. The orientation of the whole motion vector V _n is Equation 4] _{S n = S n-1 -K} 2 · | V n | K 2: transition to the correction pattern shown with a small number of damping coefficient of 1 or less when direction and in the opposite field of the whole motion vector V _n just before the consecutive third threshold value or more, for detecting at least one end of the panning and tilting, image stabilization method. 6. When the entire field motion vector V _n is small lasted continuously fourth threshold value or more, it is detected that the camera shake is small, claim 3 shake correction method according.