CN100487522C - Blurring correction method and imaging device - Google Patents

Abstract

The invention provides an image pickup device capable of calculating and correcting blur of an image with high accuracy by arithmetically processing motion vector in accordance with the reliability to blur of the image. An image pickup device 1 is provided with an image processor 3 for processing an image signal. The image processor 3 is provided with an image selector 3b for selecting an object image of an object for correcting blur of the image and a reference image of a reference of the correction, a pixel area divider 3c for dividing the object image into a plurality of pixel areas, a pixel block extractor 3d for extracting at least two object pixel blocks from the pixel areas, a motion vector arithmetic part 3e for calculating motion vector of the extracted pixel blocks, a weighting setter 3f for setting a weight coefficient of each object pixel block based on the reliability of the motion vector, a conversion coefficient calculator 3g for calculating conversion coefficient of an affine transformation corresponding to blur of the image, and an image correction calculator 3h for correcting blur by applying affine transformation to the object image based on the conversion coefficient.

Description

Blur correcting method and camera head

Technical field

The present invention relates to a kind of correcting image fuzzy (blur) blur correcting method and use this blur correcting method make a video recording shot object image and proofread and correct the fuzzy camera head of the image of being made a video recording, the particularly ambiguity correction of dynamic image and blur correcting method that is suitable for making a video recording and camera head.

Background technology

In the past, known have detect the offset comprise in being rotated in and carry out the position coalignment of harmonizing in the position of two images (for example, with reference to patent documentation 1).This position coalignment is divided into the benchmark image that becomes the benchmark of harmonizing the position a plurality of block of pixels and carries out the piece matching treatment, detection reference image and become offset in being rotated in comprising between the object images of position adjustment object is to harmonize in the position that two images are carried out on the basis with this testing result.And, this position coalignment is obtained the dispersion of the CONCENTRATION DISTRIBUTION of each block of pixels of being cut apart, be judged as the dispersion value of being obtained smaller or equal to the block of pixels of defined threshold in the result's that leads to errors of piece matching treatment possibility big, from the object of piece matching treatment, remove corresponding block of pixels, improve the accuracy of detection of offset.

11-No. 86003 communiques of [patent documentation 1] Japanese patent laid-open

6-No. 303490 communiques of [patent documentation 2] Japanese patent laid-open

[patent documentation 3] Jap.P. spy opens communique 2000-No. 161913

But, in the coalignment of above-mentioned existing position, only judge reliability at the piece matching treatment according to the dispersion of the CONCENTRATION DISTRIBUTION of each block of pixels, from process object, remove and be judged as the low block of pixels of degree of confidence, and utilize identical degree of confidence to handle, so exist the problem that not necessarily can detect two offsets between the image according to the subject that will make a video recording accurately to being judged as all high block of pixels of degree of confidence.

Summary of the invention

The present invention In view of the foregoing proposes, its purpose is to provide a kind of blur correcting method and camera head, it is based on the various characteristics value of the block of pixels cut apart, judge the degree of confidence of the mobile vector of each block of pixels of passing through institute's computings such as piece matching treatment, carry out calculation process corresponding to the mobile vector of this degree of confidence, two offsets between the image can be always detected accurately, and image blurring corresponding to this offset can be correctly proofreaied and correct.

In order to achieve the above object, one of the present invention's blur correcting method, comprise: image selection step from a plurality of images by the camera head shooting, is selected to object images and the benchmark image that becomes the benchmark of this correction to the image blurring object of proofreading and correct; The piece extraction step is from least two object pixel pieces of described object picture extraction; The vector calculus step, the computing mobile vector, described mobile vector is represented the amount of movement from the position on the described benchmark image corresponding with the position of each object pixel piece that extracts by described extraction step to the position of this each object pixel piece, according to proofread and correct the fuzzy of described object images by the mobile vector of described vector calculus step operation, it is characterized in that, described blur correcting method comprises: the weight setting step, with respect to described image blurring degree of confidence, set the weighting coefficient of described each object pixel piece based on each mobile vector by described vector calculus step operation; The coefficient calculation step, based on the true origin of regulation as the position vector of described each object pixel piece of benchmark, described each mobile vector and each weighting coefficient of setting by described weight setting step, computing is used for described object images is carried out the conversion coefficient of affined transformation (affine transformation); And the correction calculation step, based on the operation result of described coefficient calculation step, described object images is carried out affined transformation, to proofread and correct the fuzzy of this object images.

And the present invention's two blur correcting method is characterized in that, in foregoing invention, based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

And, the present invention's three blur correcting method, it is characterized in that, in foregoing invention, be divided in advance under the situation of image of a plurality of pixel regions described object images being divided into a plurality of pixel regions or described object images, described extraction step extracts at least two described object pixel pieces from described a plurality of pixel regions of being cut apart.

And, the present invention's four blur correcting method is characterized in that, in foregoing invention, described weight setting step based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

And, the present invention's five blur correcting method, it is characterized in that, in foregoing invention, described weight setting step handle is set to such an extent that ratio is big at the described weighting coefficient of the described object pixel piece of the substantial middle portion that is positioned at described object images at the described weighting coefficient of the described object pixel piece of the periphery that is positioned at described object images.

And, the present invention's six blur correcting method is characterized in that, in foregoing invention, described weight setting step is set the described weighting coefficient at described object pixel piece based on based on the direction of described mobile vector and the described degree of confidence of at least one side in the size.

And, the present invention's seven blur correcting method, it is characterized in that, in foregoing invention, the contrast of the image in the described object pixel piece of described weight setting step operation, based on described degree of confidence, set described weighting coefficient at described object pixel piece based on the operation result of this contrast.

And, the present invention's eight blur correcting method, it is characterized in that, in foregoing invention, described weight setting step operation is positioned near near the mobile vector of the block of pixels of described object pixel piece, with the described degree of confidence based on divergence is that the described weighting coefficient at described object pixel piece is set on the basis, and described divergence is represented near the extent of deviation of the mobile vector of described block of pixels with respect to the mobile vector of described object pixel piece.

And, the present invention's nine blur correcting method, it is characterized in that, in foregoing invention, described weight setting step is set described weighting coefficient at described object pixel piece according to the 1st weighting coefficient and divergence, the position of the described object pixel piece in described the 1st weighting coefficient and the described object images is corresponding, and described divergence represents to be positioned near near the extent of deviation of the mobile vector of the block of pixels the described object pixel piece with respect to the mobile vector of described object pixel piece.

And, the present invention's ten blur correcting method, it is characterized in that in foregoing invention, described coefficient calculation step is made as X respectively at handle described each position vector corresponding with described each object pixel piece, described each mobile vector and described each weighting coefficient _i(i=1,2 ..., n), V _i(i=1,2 ..., n) and w _i(i=1,2 ..., n), the summation of this each weighting coefficient is made as under the situation of w, according to X _c=(w ₁X _i+ w ₂X ₂+ ... + w _nX _n)/w comes computing to represent the position center of gravity vector X of the weighting center of gravity of described each position vector _c, according to V _c=(w ₁V ₁+ w ₂V ₂+ ... + w _nV _n)/w comes computing to represent the vector V of swerving the weight of the weighting center of gravity of described each mobile vector _c, come the rotationangle of the computing rotation matrix corresponding with the linear transformation composition of described conversion coefficient according to formula 1.

[formula 1]

$\mathrm{\φ}=\arctan \frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w}_i(X_i-X_c)\·{({-V}_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w}_i(X_i-X_c)\·\{(X_i-X_c)+({-V}_i+V_c)\}}$

Wherein, # represents the rotational transform of anticlockwise 90 degree.

And the present invention's 11 blur correcting method is characterized in that in foregoing invention, described coefficient calculation step is according to S=X _c-V _cCome the computing displacement vector S corresponding with the displacement composition of described conversion coefficient, described correction calculation step is middle mind-set counterclockwise just rotation described rotation angle with described object images with the weighting center of gravity of described position vector according to described rotation matrix, makes this object images displacement according to described displacement vector then.

And the present invention's 12 blur correcting method is characterized in that, in foregoing invention, described extraction step and described weight setting step are extracted described object pixel piece respectively and are set described weighting coefficient, so that described position center of gravity vector becomes zero vector.

And the present invention's 13 blur correcting method is characterized in that, in foregoing invention, described extraction step and described weight setting step are extracted described object pixel piece respectively and are set described weighting coefficient, so that described position center of gravity vector becomes zero vector.

And the present invention's 14 blur correcting method is characterized in that, in foregoing invention, described coefficient calculation step is being made as R (φ) to the described rotation matrix that counterclockwise carries out the rotational transform of described rotationangle just, according to S=R (φ) (X _c-V _c)-X _cCome the computing displacement vector S corresponding with the displacement composition of described conversion coefficient, after described correction calculation step makes described object images displacement according to described displacement vector, be that middle mind-set is counterclockwise rotated described rotation angle just with this object images with described true origin according to described rotation matrix.

And, the present invention's 15 blur correcting method is characterized in that, in foregoing invention, described coefficient calculation step is come the described rotationangle of computing according to formula 1 having been carried out the approximate formula 2 that obtains under the situation of size less than setting of described each mobile vector.

[formula 2]

$\mathrm{\φ}=\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w}_i(X_i-X_c)\·{({-V}_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w}_i(X_i-X_c)\·\{(X_i-X_c)+({-V}_i+V_c)\}}$

And, the present invention's 16 blur correcting method, it is characterized in that, in foregoing invention, described extraction step extracts described object pixel piece respectively from two these folding corner regions folding corner region, different in each bight of being arranged in described object images, described coefficient calculation step as described true origin, is made as X to the position of one of them the object pixel piece the object pixel piece that extracts from described two folding corner regions to described position vector, described mobile vector and the described weighting coefficient corresponding with the described object pixel piece that is positioned at this true origin respectively ₁, V ₁And w ₁By described weight setting step, at the described weighting coefficient of the object pixel piece that is positioned at described true origin with under distinguishing infinitely near 1 and 0 situation at the described weighting coefficient of another object pixel piece, come the described rotationangle of computing according to formula 1 having been carried out the approximate formula 3 that obtains.

[formula 3]

$\mathrm{\&phi;}=\arctan \frac{{-X}_2\&CenterDot;{({-\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="C200610001116E00381.png"\; state="\{\{state\}\}">V</figure-callout>}}_1+V_2)}^{\#}}{X_2\&CenterDot;\{X_2-({-\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="C200610001116E00381.png"\; state="\{\{state\}\}">V</figure-callout>}}_1+V_2)\}}$

And, the present invention's 17 blur correcting method, it is characterized in that, in foregoing invention, described a plurality of image is a dynamic image, described each object pixel piece is the piece that becomes the process object that the inter prediction that carries out handles in the compressed encoding of described dynamic image is handled, described each mobile vector in described inter prediction is handled by computing.

And, the present invention's 18 blur correcting method, it is characterized in that, in foregoing invention, described a plurality of image is a dynamic image, described each object pixel piece is the piece that becomes the process object that the inter prediction that carries out handles in the compressed encoding of described dynamic image is handled, described each mobile vector in described inter prediction is handled by computing.

The present invention's 19 camera head comprises: will assemble and the image pickup optical system of imaging shot object image from the light of subject; The make a video recording image unit of the image corresponding with described shot object image; The image selected cell, it is selected to the object images and the benchmark image that becomes the benchmark of this correction of the fuzzy object of correcting image from a plurality of images of being made a video recording by described image unit; The piece extraction unit, it is from least two object pixel pieces of described object picture extraction; The vector calculus unit, its computing mobile vector, this mobile vector is represented the amount of movement from the position on the described benchmark image corresponding with the position of each object pixel piece that extracts by described extraction unit to the position of this each object pixel piece, according to proofread and correct the fuzzy of described object images by the mobile vector of described vector calculus unitary operation, it is characterized in that, described camera head comprises: the weight setting unit, it with respect to described image blurring degree of confidence, sets the weighting coefficient of described each object pixel piece based on each mobile vector by described vector calculus unitary operation; The coefficient arithmetic element, it is the position vector of described each object pixel piece of benchmark, described each mobile vector and each weighting coefficient of setting by described weight setting unit according to the true origin with regulation, and computing is carried out the conversion coefficient that affined transformation is used to described object images; And correction computing unit, it carries out affined transformation based on the operation result of described coefficient arithmetic element to described object images, to proofread and correct the fuzzy of this object images.

And, the present invention's 20 camera head is characterized in that, in foregoing invention, described weight setting unit based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

And, the present invention's 21 camera head, it is characterized in that, in foregoing invention, described object images is divided into a plurality of pixel regions, perhaps described object images is the image that is split into a plurality of pixel regions in advance, and described extraction unit extracts at least two described object pixel pieces from divided described a plurality of pixel regions.

And, the present invention's 22 camera head is characterized in that, in foregoing invention, described weight setting unit based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

And, the present invention's 23 camera head is characterized in that, in foregoing invention, the described weighting coefficient at described object pixel piece is set based on based on the direction of described mobile vector and the described degree of confidence of at least one side in the size in described weight setting unit.

And, the present invention's 24 camera head is characterized in that, in foregoing invention, the described weighting coefficient at described object pixel piece is set based on the described degree of confidence based on the contrast of the image in the described object pixel piece in described weight setting unit.

And, the present invention's 25 camera head, it is characterized in that, in foregoing invention, described weight setting unitary operation is positioned near near the mobile vector of the block of pixels of described object pixel piece, with the described degree of confidence based on divergence is that the described weighting coefficient at described object pixel piece is set on the basis, and described divergence represents to be positioned near near the described mobile vector of block of pixels of described object pixel piece with respect to the extent of deviation of the mobile vector of described object pixel piece.

And, the present invention's 26 camera head, it is characterized in that, in foregoing invention, described weighting coefficient at described object pixel piece is set according to the 1st weighting coefficient and divergence in described weight setting unit, the position of the described object pixel piece in described the 1st weighting coefficient and the described object images is corresponding, and described divergence represents to be positioned near near the mobile vector of block of pixels of described object pixel piece with respect to the extent of deviation of the mobile vector of described object pixel piece.

And, the present invention's 27 camera head, it is characterized in that in foregoing invention, described coefficient arithmetic element is made as X respectively at handle described each position vector corresponding with described each object pixel piece, described each mobile vector and described each weighting coefficient _i(i=1,2 ..., n), V _i(i=1,2 ..., n) and w _i(i=1,2 ..., n), the summation of this each weighting coefficient is made as under the situation of w, according to X _c=(w ₁X ₁+ w ₂X ₂+ ... + w _nX _n)/w comes computing to represent the position center of gravity vector X of the weighting center of gravity of described each position vector _c, according to V _c=(w ₁V ₁+ w ₂V ₂+ ... + w _nV _n)/w comes computing to represent the vector V of swerving the weight of the weighting center of gravity of described each mobile vector _c, come the rotationangle of the computing rotation matrix corresponding with the linear transformation composition of described conversion coefficient according to formula 4.

[formula 4]

$\mathrm{\&phi;}=\arctan \frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{({-V}_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}$

Wherein, # represents the rotational transform of anticlockwise 90 degree.

And the present invention's 28 camera head is characterized in that in foregoing invention, described coefficient arithmetic element is according to S=X _c-V _cCome the computing displacement vector S corresponding with the displacement composition of described conversion coefficient, described correction computing unit is middle mind-set counterclockwise just rotation described rotation angle with described object images with the weighting center of gravity of described position vector according to described rotation matrix, makes this object images displacement according to described displacement vector then.

And the present invention's 29 camera head is characterized in that, in foregoing invention, described extraction unit and described weight setting unit extract described object pixel piece respectively, and set described weighting coefficient, so that described position center of gravity vector becomes zero vector.

And, the present invention's 30 camera head is characterized in that, in foregoing invention, described coefficient arithmetic element is come the described rotationangle of computing according to formula 4 having been carried out the approximate formula 5 that obtains under the situation of size less than setting of described each mobile vector.

[formula 5]

$\mathrm{\&phi;}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{({-V}_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}$

And, the present invention's 31 camera head, it is characterized in that, in foregoing invention, described extraction unit extracts described object pixel piece respectively from two different these folding corner regions of the folding corner region in each bight of being arranged in described object images, described coefficient arithmetic element as described true origin, is made as X to the position of one of them the object pixel piece the object pixel piece that extracts from described two folding corner regions to described position vector, described mobile vector and the described weighting coefficient corresponding with the described object pixel piece that is positioned at this true origin respectively ₁, V ₁And w ₁By described weight setting unit, at the described weighting coefficient of the object pixel piece that is made as described true origin with under distinguishing infinitely near 1 and 0 situation at the described weighting coefficient of another object pixel piece, come the described rotationangle of computing according to formula 4 having been carried out the approximate formula 6 that obtains.

[formula 6]

$\mathrm{\&phi;}=\arctan \frac{{-X}_2\&CenterDot;{({-\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="C200610001116E00381.png"\; state="\{\{state\}\}">V</figure-callout>}}_1+V_2)}^{\#}}{X_2\&CenterDot;\{X_2-({-\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="C200610001116E00381.png"\; state="\{\{state\}\}">V</figure-callout>}}_1+V_2)\}}$

And, the present invention's 32 blur correcting method, it is characterized in that, in foregoing invention, described extraction step extracts described object pixel piece respectively from two different these folding corner regions of the folding corner region in each bight of being arranged in described object images, described coefficient calculation step as described true origin, is made as V to the position of one of them the object pixel piece the object pixel piece that extracts from described two folding corner regions to corresponding described mobile vector ₁, described position vector corresponding with the one other pixel piece and described mobile vector are made as X ₂And V ₂, make the described weighting coefficient corresponding near 0 with described one other pixel piece, thereby according to X ₂(V ₁-V ₂) ^#(wherein, # represents the rotational transform of anticlockwise 90 degree) and X ₂{ X ₂-(V ₁-V ₂) rotation angle of coming the computing rotation matrix corresponding with the linear transformation composition of described conversion coefficient.

According to blur correcting method of the present invention and camera head, the various characteristics value of the block of pixels that has obtained to have carried out cutting apart, judge the degree of confidence of the mobile vector of each block of pixels of passing through institute's computings such as piece matching treatment, carry out the calculation process of the mobile vector corresponding with this degree of confidence, two offsets between the image can be always detected accurately, and the fuzzy of the image corresponding can be correctly proofreaied and correct with this offset.

Description of drawings

Fig. 1 is the block scheme of the structure of the camera head that relates to of expression embodiments of the present invention 1.

Fig. 2 is the stereographic map of the summary structure of the camera head that relates to of expression embodiments of the present invention 1.

Fig. 3-the 1st, the synoptic diagram of expression benchmark image, this benchmark image are a part of images that utilizes the dynamic image that camera head shown in Figure 1 makes a video recording.

Fig. 3-the 2nd, the synoptic diagram of indicated object image, this object images is to utilize a part of image of the dynamic image that camera head shown in Figure 1 makes a video recording.

Fig. 4 is the process flow diagram of the image blur correcting treatment step that carries out of expression image processing part shown in Figure 1.

Fig. 5 is that expression has been carried out object images to cut apart and the synoptic diagram of an example of the pixel region that obtains and object pixel piece.

Fig. 6 is that expression has been carried out object images to cut apart and the pixel region that obtains and another routine synoptic diagram of object pixel piece.

Fig. 7 is the process flow diagram that expression is extracted in the treatment step that occupies the high object pixel piece of ratio of image medium-high frequency composition.

Fig. 8 is that expression has been carried out object images to cut apart and another routine synoptic diagram of the pixel region that obtains.

To be expression set the process flow diagram of the treatment step that the weight setting of weighting coefficient handles according to the divergence of mobile vector to Fig. 9.

Figure 10 is the synoptic diagram of the extraction example of the expression adjacent pixel blocks adjacent with the object pixel piece.

To be expression set the process flow diagram of the treatment step that the weight setting of weighting coefficient handles according to picture contrast to Figure 11.

Figure 12 is the synoptic diagram of an example of the expression state that extracted two object pixel pieces from the folding corner region of object images.

Symbol description

1 camera head; 2 image pickup parts; The 2a image pickup optical system; The 2b imaging apparatus; 2c A/D converter section; 3 image processing parts; 3a Flame Image Process control part; 3b image selection portion; 3c pixel region cutting part; 3d block of pixels extraction unit; 3e mobile vector operational part; 3f weight setting portion; 3g conversion coefficient operational part; 3h image rectification operational part; The 3i frame memory; 4 input parts; The 4a switch of photographing; 5 display parts; The 5a display monitor; 6 sound input and output portions; 7 Department of Communication Forces; 8 storage parts; 9 portable recording mediums; The 9a card; A1～A4 pixel region; B1～B6, BO1, BO2, BO3, Bi, BS1, BS2, IB1～IB4, OB1～OB8 block of pixels; GO, G1, Gt object images; O1, O2, Os true origin; V ₁～V ₃, V _S1, V _S2Mobile vector; X, X ₁～X ₃Position vector.

Embodiment

Below, the preferred implementation of the camera head that present invention will be described in detail with reference to the accompanying relates to.In addition, be not to utilize this embodiment to limit the present invention.And, in the record of accompanying drawing, on same section, mark same numeral.

(embodiment 1)

At first, the camera head that embodiments of the present invention 1 relate to is described.Fig. 1 is the block scheme of the structure of the camera head that relates to of expression present embodiment 1.Fig. 2 is the stereographic map of the summary structure of the camera head that relates to of expression present embodiment 1.As shown in Figure 1, the camera head 1 that relates to of this embodiment 1 has: the shooting shot object image also generates the image pickup part 2 of picture signal; Handle the image processing part 3 of the picture signal of image pickup part 2 generations; Import the input part 4 of various indication informations; The display part 5 that shows various information; Carry out the sound input and output portion 6 of the input and output processing of acoustic information; And carry out the Department of Communication Force 7 of information communication between the external device (ED); The storage part 8 of storing various information; And carry out the portable recording medium 9 of the transmitting-receiving of data between the external device (ED); The processing of the integral body of control camera head 1 and the control part C of action.Image pickup part 2, image processing part 3, input part 4, display part 5, sound input and output portion 6, Department of Communication Force 7, storage part 8 and portable recording medium 9 are electrically connected with control part C.Control part C controls these each component parts.

Image pickup part 2 has image pickup optical system 2a, imaging apparatus 2b and A/D converter section 2c.Image pickup optical system 2a will be from the light optically focused of any subject, imaging shot object image on imaging apparatus 2b.Imaging apparatus 2b as image unit uses CCD (charge-coupled image sensor), CMOS solid-state imagers such as (complementary metal oxide semiconductor (CMOS)s) to realize, the light that image pickup optical system 2a assembled is detected as light signal, and the shooting image corresponding with shot object image, the image transitions of being made a video recording is exported for the electric signal as simulating signal.A/D converter section 2c is the analog signal conversion of imaging apparatus 2b output a digital signal, exports to image processing part 3.Image pickup optical system 2a utilizes the zoom lens can change focal length to realize, for example as shown in Figure 2, a plurality of lens are made up and constitutes.And image pickup optical system 2a makes part or all lens move at optical axis direction by not shown drive unit under the control of control part C, thus the change focal length, and carry out the focusing of the image that will make a video recording.In addition, image pickup optical system 2a can be the fixing single focal lense of focal length, also can constitute releasably to be installed on the housing, can replace mutually with other image pickup optical system.

Image processing part 3 has: Flame Image Process control part 3a, carry out the control at the various Flame Image Process of the picture signal of obtaining from image pickup part 2; Image selection portion 3b from the picture signal that is obtained from image pickup part 2, is selected to object images and the benchmark image that becomes the benchmark of this correction to the fuzzy object of proofreading and correct of image; Pixel region cutting part 3c is divided into a plurality of pixel regions to selected object images; Block of pixels extraction unit 3d extracts at least two object pixel pieces that do not comprise the border of each pixel region of cutting apart to some extent and do not exceed from each pixel region from object images; Mobile vector operational part 3e, computing mobile vector, this mobile vector are represented the amount of movement from the position on the benchmark image corresponding with the position of each object pixel piece that is extracted to the position of this each object pixel piece; The 3f of weight setting portion, based on each mobile vector of institute's computing degree of confidence fuzzy with respect to object images, the weighting coefficient of the weight of each object pixel piece that the setting expression is extracted; Conversion coefficient operational part 3g, as the position vector of each object pixel piece of benchmark, by each mobile vector of mobile vector operational part 3e computing and each weighting coefficient of setting by the 3f of weight setting portion, computing is carried out the conversion coefficient that affined transformation is used to object images based on a true origin of regulation; Image rectification operational part 3h based on this operation result, carries out affined transformation to object images, and proofreaies and correct the fuzzy of this object images; Frame memory 3i, interim store image information.These each component parts that image processing part 3 has, according to based on from the indication of control part C, from the indication of Flame Image Process control part 3a, handle picture signal, suitably exporting to control part C or storage part 8 as the image information of result, ambiguity correction information, operation result etc.3 pairs of picture signals of being obtained of image processing part also can be carried out various Flame Image Process such as γ proofreaies and correct, Y/C separation (luminance signal/color signal separates), colour switching.

Input part 4 receives the input of the indication information of various processing that camera head 1 carries out and action from the outside, and the indication information of being imported is exported to control part C.Comprise the setting etc. of the disposal route of the setting of setting, the image pickup mode of the focusing of starting/ends of camera head 1, the beginning/end of photographing, image pickup optical system 2a and zoom position, the picture signal of being made a video recording in the indication information that input part 4 receives.Input part 4 can utilize realizations such as various switches, enter key, touch pad such as button-type, conversion (toggle) type, for example as shown in Figure 2, has disposed the photography switch 4a of the beginning/end that is used to indicate photography.

Display part 5 has and uses for example display device of LCD, organic EL (Electroluminescence, electroluminescence) display, LED etc., shows various information according to the indication from control part C.The information that display part 5 shows comprises image information that image pickup part 2 shootings and the image information that generates, image processing part 3 handle, be used for notifying various processing that control part C controls and action the announcement information that begins and finishes, be used to notify in the error message of the mistake of various processing and action generation etc.Display part 5 has the display monitor 5a that the LCD utilized realizes in the back side of camera head 1 one for example as shown in Figure 2, almost shows the image that camera head 1 is made a video recording in real time.In addition, display monitor 5a also can show the various information such as image that are stored in the storage part 8.

Sound input and output portion 6 uses microphone and loudspeaker to realize, accepts the input of acoustic information from the outside, and the acoustic information of being imported is exported to control part C, and the acoustic information from control part C input is outputed to the outside.The information that sound input and output portion 6 carries out input and output has corresponding acoustic information of the video of photographing with image pickup part 2 etc.In addition, the output of the output of the notification voice that sound input and output portion 6 also can stipulate or the warning sound of regulation, the output of the notification voice of described regulation is used to notify the beginning and the end of the processing that each component part carries out, and the output of the warning sound of described regulation is used for notifying the mistake that produces in various processing and action.

Department of Communication Force 7 uses PERCOM peripheral communication such as RS232C, USB, IEEE1394 to wait with interface or based on the infrared communication interface of IrDA standard and realizes, according to indication, and carry out the various information communications of image information, acoustic information etc. between the external device (ED) from control part C.

Storage part 8 uses the ROM of various information such as the program of having stored the OS that is used to start regulation in advance, handling procedure and has stored the various processing parameters of each processing and input and output realize to the RAM of the various information of each component part etc.Storage part 8 canned datas have: the image information that image pickup part 2 is made a video recording, as the image information of the result of image processing part 3, control information, operation result etc., the computing parameter of in image processing part 3, using, with the corresponding acoustic information of video that image pickup part 2 is photographed, the various set informations of camera head 1 etc.

It is intelligent medium SM flash memory, DVD (Digital Versatile Disk such as (registered trademarks) that portable recording medium 9 uses nonvolatile memory, Digital video disc) optical storage media etc. such as, for example as shown in Figure 2, releasably be connected to camera head 1 by card 9a, in the time of on being inserted into camera head 1, can carry out the output or the record of various information such as view data.Portable recording medium 9 is according to the output and the record that carry out information from the indication of control part C.

The CPU of the handling procedure that control part C utilization execution storage part 8 is stored waits and realizes each component part of control camera head 1.Particularly control part C makes Flame Image Process control part 3a handle image and blur correction mode, makes display part 5 show the image information of being proofreaied and correct, and the image information of being proofreaied and correct is stored in the storage part 8.And the image information that control part C is made a video recording image pickup part 2 outputs to the outside by Department of Communication Force 7, perhaps is recorded in the portable recording medium 9.In addition, control part C also can output to the outside to the image information as the result of image processing part 3.

Below, the image blur correcting processing that image processing part 3 carries out is described.Fig. 3-1 and Fig. 3 the-the 2nd, and the image of the part of the dynamic image of being made a video recording by camera head 1 is respectively that expression becomes the benchmark image of the fuzzy benchmark of correcting image and becomes the figure of an example of object images of the object of blur correction mode.Shown in Fig. 3-1 and Fig. 3-2, object images G1 is with respect to benchmark image GO, produces corresponding image blurring of the action of the camera head 1 when photographing.This image blurringly generally can show by affined transformation, on the contrary, by object images G1 is carried out affined transformation, can correcting image fuzzy.Herein, affined transformation refers to based on linear transformation such as rotational transform and the parallel conversion of figure etc. of moving the combination that is shift transformation, the following numerical expression of general using (1) expression.

X’＝T·X+S ……(1)

Herein, the position vector of the corresponding arbitrfary point of the figure before and after the expression affined transformation is made as X and X ' respectively, matrix as the linear transformation of representing affined transformation conversion coefficient partly is made as T, the vector as the shift transformation of representing affined transformation conversion coefficient partly is made as S.But, if shown in Fig. 3-2 by the displacement and the rotation constitute image blurring, then represent Matrix of Linear T can be replaced into the expression rotation angle be the rotation matrix R (φ) of the anticlockwise rotational transform of φ.In addition, a part of subject of object images G1 and benchmark image GO, shown in the relation of shot object image UO among the benchmark image GO and the shot object image U1 among the object images G1, position of the two and posture change are independent of image blurring, and this variation is not proofreaied and correct by affined transformation.

Image processing part 3 extracts object pixel piece B1～B3 from object images G1 for example shown in Fig. 3-2, the mobile vector V of each object pixel piece B1～B3 that is extracted by the computing of piece matching treatment ₁～V ₃And image processing part 3 is according to the mobile vector V of institute's computing ₁～V ₃, with mobile vector V ₁～V ₃At the weighting coefficient w of image blurring degree of confidence for basic each object pixel piece B1～B3 that sets ₁～w ₃, and indicate the position vector X of the center of each object pixel piece B1～B3 ₁～X ₃, the conversion coefficient of computing and the fuzzy corresponding affined transformation of object images G1 based on the conversion coefficient of institute's computing, carries out affined transformation to object images G1, correcting image fuzzy.Herein, said matching treatment is meant obtains each object pixel piece B1～B3 of being extracted processing apart from the parallel amount of movement of benchmark image GO, specifically, be on benchmark image GO, to detect respectively and the most consistent block of pixels BO1, BO2, the BO3 of object pixel piece B1～B3, obtain expression from the center of detected each block of pixels BO1, BO2, BO3 the mobile vector V to the amount of movement of the center of corresponding respectively object pixel piece B1～B3 ₁～V ₃Generally, the consistent degree between the block of pixels can use the fourier coefficient of phase cross correlation function, gray-scale value of absolute value, gray-scale value of difference of standard deviation of absolute value, gray-scale value of difference of absolute value sum, the average gray value of the difference gray-scale value of the pixel in the block of pixels to wait to estimate.In addition, become expression position vector X ₁～X ₃The true origin of benchmark, be the some O1 in the upper left bight of object images G1 in Fig. 3-2, but be not limited thereto, also can be the inside and outside arbitrfary point of object images G1.

And image processing part 3 utilizes least square method (least square method), and computing is used for the conversion coefficient of the affined transformation of calibration object image G1.That is, the conversion coefficient of the linear transformation composition of image processing part 3 computing affined transformations is the conversion coefficient of rotation matrix R (φ) and the shift transformation composition vector S that promptly is shifted, so that the value minimum of the evaluation function F of following numerical expression (2) expression.

[formula 7]

$F=\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i{\{(R\left(\mathrm{\&phi;}\right)\&CenterDot;X_i+S)-(X_i-V_i)\}}^2\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(2\right)$

Herein, with object pixel piece Bi (i=1,2 ..., n) corresponding position vector, mobile vector and weighting coefficient be made as X respectively _i(i=1,2 ..., n), V _i(i=1,2 ..., n) and w _i(i=1,2 ..., n), the object pixel piece number that extracts from object images G1 is made as n, make the formula vague generalization.In numerical expression (2), mobile vector V _iExpression is with position vector X _iThe amount of movement that is moved with respect to benchmark image GO.Therefore, from X _iDeduct V _iPosition vector (the X that obtains _i-V _i) become closely to X _iPosition vector under the situation of having carried out proofreading and correct.With respect to this, as numerical expression (1), as another (R (φ) X of numerical expression (2) _i+ S) represent by affined transformation position vector X _iPosition vector after having carried out proofreading and correct.Therefore, utilize weighting coefficient w in order to make _iTo vector (X _i-V _i) and vector (R (φ) X _iThe value of summation F after+S) square being weighted of absolute value of differential vector selects R (φ) and S to mean the best shift calibrating of least square method for minimum.Position alignment method based on the least square method of having considered weighting for example discloses in patent documentation 3.

For hour, rotation matrix R (φ) and displacement vector S satisfy following numerical expression (3) and numerical expression (4) in the value of evaluation function F, and utilize numerical expression respectively (5) and numerical expression (6) arithmetic shift vector S and rotation matrix R (φ).

[formula 8]

$\frac{\&PartialD;F}{\&PartialD;S}=2\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i\{R\left(\mathrm{\&phi;}\right)\&CenterDot;X_i+S-X_i+V_i\}=0\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(3\right)$

[formula 9]

$\frac{\&PartialD;F}{\&PartialD;\mathrm{\&phi;}}=2\underset{i}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i\{R\left(\mathrm{\&phi;}\right)\&CenterDot;X_i+S-X_i+V_i\}\&CenterDot;\{\frac{\&PartialD;R\left(\mathrm{\&phi;}\right)}{\&PartialD;\mathrm{\&phi;}}\&CenterDot;X_i\}=0\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(4\right)$

S＝X _c-V _c-R(φ)·X _c……(5)

[formula 10]

$\mathrm{\&phi;}=\arctan \frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{({-V}_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(6\right)$

Wherein, # represents the rotational transform of anticlockwise 90 degree.And, weighting coefficient w _iSummation be made as w, expression each position vector X _iThe position center of gravity vector of weighting center of gravity be made as X _c, then utilize numerical expression (7) to carry out computing, each mobile vector of expression V _iThe vector of swerving the weight of weighting center of gravity be made as V _c, utilize numerical expression (8) to carry out computing.

X _c＝(w _i·X ₁+w ₂·X ₂+…+w _n·X _n)/w ……(7)

V _c＝(w ₁·V ₁+w ₂·V ₂+…+w _n·V _n)/w ……(8)

And, utilize numerical expression (9) expression rotation matrix R (φ).

[formula 11]

$R\left(\mathrm{\&phi;}\right)=\left(\begin{array}{cc}\cos \mathrm{\&phi;}& -\sin \mathrm{\&phi;}\\ \sin \mathrm{\&phi;}& \cos \mathrm{\&phi;}\end{array}\right)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(9\right)$

In addition, learning rotationangle in advance hour, perhaps each mobile vector V _iSize during less than setting etc., also can utilize and carry out numerical expression (6) approximate and numerical expression (10) that obtain comes the computing rotationangle.Thus, image processing part 3 can be obtained the calculation process of rotation matrix R (φ) fast.

[formula 12]

$\mathrm{\&phi;}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{({-V}_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(10\right)$

Utilize rotation matrix R (φ) and the displacement vector S obtained like this, 3 couples of object images G1 of image processing part carry out the affined transformation shown in the following numerical expression (11), with bluring of correcting image.

X’＝R(φ)·X+{X _c-V _c-R(φ)·X _c} ……(11)

In this affined transformation, image processing part 3 is that middle mind-set is counterclockwise rotated rotationangle just with object images G1 with true origin O1, according to displacement vector S=X _c-V _c-R (φ) X _c, make postrotational object images G1 displacement.

In addition, numerical expression (11) can be deformed into following numerical expression (12).

X’＝(R(φ)·(X-X _c)+X _c)-V _c……(12)

That is, the affined transformation shown in the numerical expression (11) can be replaced into shown in numerical expression (12), with object images G1 with position center of gravity vector X _cRepresented each position vector X _iThe weighting center of gravity be middle mind-set rotation rotationangle just counterclockwise, according to vector (V _c) make the conversion of postrotational object images G1 displacement.Under this situation, compare with the computing of the affined transformation shown in the numerical expression (11), the computing of can simplified image handling part 3 carrying out can alleviate the processing load of image processing part 3, improves processing speed.

With reference to Fig. 4 the image blur correcting treatment step that image processing part 3 carries out is described herein.Fig. 4 is the process flow diagram of the image blur correcting treatment step controlled of the presentation video processing controls 3a of portion.In addition, flowchart illustration shown in Figure 4 at the ambiguity correction treatment step of an object images, for example when proofreading and correct dynamic image fuzzy, as long as repeat this treatment step continuously.

As shown in Figure 4, under the control of Flame Image Process control part 3a, image selection portion 3b selects the object images Gt and the benchmark image Go (step S101) of ambiguity correction from the image information of being obtained by image pickup part 2.Pixel region cutting part 3c selected object images Gt be divided into a plurality of pixel region As (s=1,2 ..., h, h is for cutting apart number) (step S103), block of pixels extraction unit 3d is divided into a plurality of block of pixels (step S105) again to each pixel region As of having cut apart, and carries out extracting at least two block of pixels and handle (step S107) as the piece extraction of object pixel piece Bi from all block of pixels of being cut apart.And mobile vector operational part 3e is by piece matching treatment, the mobile vector V of each object pixel piece of computing Bi _i(step S109).The 3f of weight setting portion is with each mobile vector V of institute's computing _iWith respect to image blurring degree of confidence is the basis, sets the weighting coefficient w of each object pixel piece Bi _iWeight setting handle (step S111).Then, conversion coefficient operational part 3g computing is used for the conversion coefficient (step S113) of the affined transformation of calibration object image Gt, image rectification operational part 3h is based on the conversion coefficient of institute's computing, object images Gt is carried out affined transformation, to proofread and correct fuzzy (step S115), finish a series of ambiguity correction and handle with respect to benchmark image Go.

At step S103, pixel region cutting part 3c is for example by shown in Figure 5, as pixel region As, object images Gt is divided into the medial region IA of the central portion that is positioned at object images Gt and is positioned at these two pixel regions of exterior lateral area OA of the periphery of object images Gt.Fig. 5 represents that specifically resolution is VGA (Video Graphics Array, Video Graphics Array), for the object images Gt of 640 * 480 pixels, establishes the example under the situation that medial region IA is 480 * 320 pixels.In addition, pixel region cutting part 3c also can be divided into more zone to object images Gt, for example, can come the All Ranges of cutting object image Gt according to each pixel region of 16 * 16 pixels, under this situation, object images Gt is split into 40 * 30 pixel regions.And the resolution of object images Gt is not limited to VGA, also can be XGA arbitrary resolutions such as (Extended GraphicsArray, XGA (Extended Graphics Array)).

At step S105, block of pixels extraction unit 3d according to the block of pixels of each 8 * 8 pixel to cutting apart in the object images Gt.Under this situation, object images Gt integral body is split into 80 * 60 block of pixels, and wherein medial region IA is split into 60 * 40 block of pixels.Herein, the size and dimension of the block of pixels of 8 * 8 pixels equates that with the size and dimension of the block of pixels of the process object that becomes the inter prediction processing it generally is to carry out that this inter prediction is handled in the compressed encoding of dynamic image is handled.Like this, in different disposal, by commonization of the block of pixels that realizes cutting apart, for example can handle the ambiguity correction that the mobile vector of each block of pixels of institute's computing in inter prediction is handled be used for this embodiment 1, can promote the rapid of the simplification of entire process and processing speed.In addition, the size of the block of pixels of 8 * 8 pixels that pixel region cutting part 3c considered pixel piece extraction unit 3d is cut apart is the pixel region As of image segmentation for the size of the integral multiple that is of a size of this block of pixels in length and breadth.

At step S107, block of pixels extraction unit 3d for example by shown in Figure 5, extracts a plurality of object pixel piece Bi from each pixel region IA, OA.And at step S111, the 3f of weight setting portion sets weighting coefficient w according to the pixel region As that includes object pixel piece Bi to each object pixel piece Bi _iAt this moment, the 3f of weight setting portion weighting coefficient w at the object pixel piece Bi of the periphery that is positioned at object images Gt _i, set greatlyyer than the object pixel piece Bi that is positioned near the central portion of object images Gt.That is, for a plurality of object pixel piece Bi shown in Figure 5, the 3f of weight setting portion is the weighting coefficient w that is positioned at the object pixel piece Bi of exterior lateral area OA _iSet than the weighting coefficient w of the object pixel piece Bi that is positioned at medial region IA _iGreatly.

Usually, for being accompanied by the image blurring of rotation, because the periphery in object images Gt produces bigger bluring, so be positioned at the mobile vector V of the object pixel piece Bi of exterior lateral area OA _iBecome bigger vector, accurately computing.On the other hand, the central portion of object images Gt by rotation cause fuzzy less, be positioned at the mobile vector V of the object pixel piece Bi of medial region IA _iDiminish.And it is more that general shot object image concentrates on the situation of central portion of object images Gt since with the mobile mobile vector V that makes of image blurring different subject _iChange easily, therefore be positioned at the mobile vector V of the object pixel piece Bi of medial region IA _iThe operational precision step-down.Therefore, be judged as mobile vector V _iFuzzy degree of confidence with respect to object images Gt is higher at exterior lateral area OA, and IA is lower in medial region, and the result is positioned at the weighting coefficient w of the object pixel piece Bi of exterior lateral area OA _iBe set than the weighting coefficient w of the object pixel piece Bi that is positioned at medial region IA _iGreatly.

Like this, the 3f of weight setting portion sets the weighting coefficient w of each object pixel piece Bi according to pixel region As _iThereby in the camera head 1 that this embodiment 1 relates to, that can alleviate subject and the image blurring irrelevant influence of moving, the conversion coefficient of computing affined transformation accurately is rotation matrix R (φ) and displacement vector S, correctly correcting image blurs.In addition, the 3f of weight setting portion is for a plurality of object pixel piece Bi that are positioned at same pixel zone As, can also set each weighting coefficient w according to the position judgment of each object pixel piece Bi with respect to image blurring degree of confidence _iThus, conversion coefficient operational part 3g more closely computing rotation matrix R (φ) and the displacement vector S.

In addition, the block of pixels that block of pixels extraction unit 3d is cut apart is not limited to the block of pixels of 8 * 8 pixels, also can be the block of pixels of arbitrary dimensions such as 16 * 16 pixels.And the object pixel piece Bi that block of pixels extraction unit 3d is extracted exceeds only otherwise from each pixel region As and gets final product, and can be one of pixel region As.In addition, object pixel piece Bi also can have the size and dimension that has nothing in common with each other.Block of pixels extraction unit 3d not necessarily extracts object pixel piece Bi from all pixel region As, for example also can only extract from a pixel region As.

Herein, another example of the object pixel piece Bi that extracted of Fig. 6 remarked pixel piece extraction unit 3d.As shown in Figure 6, as object pixel piece Bi, block of pixels extraction unit 3d extracts bigger object pixel piece IB from medial region IA ₁～IB ₄, extract less object pixel piece OB from exterior lateral area OA ₁～OB ₈Its reason and the 3f of weight setting portion are the weighting coefficient w that is positioned at the object pixel piece Bi of exterior lateral area OA _iSet than the weighting coefficient w of the object pixel piece Bi that is positioned at medial region IA _iIdentical greatly.That is, because object pixel piece OB ₁～OB ₈Mobile vector be bigger vector, so even reduce object pixel piece OB ₁～OB ₈Size, also computing mobile vector accurately by reducing this size, can alleviate the processing load of computing, improves processing speed.On the other hand, because object pixel piece IB ₁～IB ₄The operational precision of mobile vector lower, so increase the object pixel piece IB that will extract ₁～IB ₄Size so that can the bigger zone of computing utilization have carried out the mobile vector of equalization.

In addition, extract object pixel piece IB ₁～IB ₄And OB ₁～OB ₈Make them with respect to the center line in length and breadth of the central point O2 by object images Gt, in each pixel region IA, OA, be positioned at symmetrical position.Like this, by extracting the object pixel piece symmetrically, for example the object pixel piece Bi that is in the As of same pixel zone is set weighting coefficient w equably at the 3f of weight setting portion _iSituation under, can make and utilize position center of gravity vector X _cRepresented each position vector X _iThe weighting center of gravity consistent with central point O2.Under this situation,, can make position center of gravity vector X by the true origin with respect to object images Gt is set at central point O2 _cBecome zero vector, can shown in following numerical expression (13), simplify the computing of the affined transformation of numerical expression (12) expression.Thus, can further alleviate the processing load of image processing part 13, further improve processing speed.

X’＝R(φ)·X-V _c ……(13)

And, as shown in Figure 6, medial region IA, exterior lateral area OA, object pixel piece IB ₁～IB ₄With object pixel piece OB ₁～OB ₈Be the rectangular area, but can be arbitrary shape.In addition, image-region As and object pixel piece Bi can be predefined regulation zones, also can extract arbitrarily according to object images Gt, perhaps also can extract according to the various characteristics value of object images Gt.

Herein, pixels illustrated piece extraction unit 3d extracts an example of the processing of object pixel piece Bi according to the characteristic value of object images Gt.Fig. 7 be piece that expression will step S107 shown in Figure 4 extract the processing subroutinization the process flow diagram of treatment step.As shown in Figure 7, extract in the processing at this piece, block of pixels extraction unit 3d carries out DCT (Discrete Cosine Transform: discrete cosine transform) handle (step S201) to each block of pixels of cutting apart by step S105 under the control of Flame Image Process control part 3a.And, based on this DCT result, from each pixel region As of cutting apart by step S103, the high a plurality of block of pixels of ratio of occupying of the image medium-high frequency composition in block of pixels are extracted (step S203) as object pixel piece Bi, and return step S107.In step S203, block of pixels extraction unit 3d is the DCT result that the DCT coefficient is as D _Uv(u, v=0,1 ..., 7), use the evaluation of estimate C that determines by following numerical expression (14) _r(r=1,2 ..., h), be judged as evaluation of estimate C _rThe big more then image medium-high frequency composition in block of pixels to occupy ratio high more.

[formula 13]

$C_r=\underset{u+v\&GreaterEqual;10}{\underset{0\&le;u,v\&le;7}{\mathrm{\&Sigma;}}}\left|D_{\mathrm{uv}}\right|\&CenterDot;\frac{1}{D_{\mathrm{yn}}}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(14\right)$

Wherein, dynamic range is made as D _YnFor example utilize 8 rank to represent then this dynamic range D _YnBe D _Yn=2 ⁸=256.

Like this, occupy the high object pixel piece Bi of ratio, can be extracted in the high object pixel piece Bi of process sensitivity in the piece matching treatment, accurately the mobile vector V of each object pixel piece Bi of extracting of computing by what be extracted in image medium-high frequency composition _iSo, accurately computing rotation matrix R (φ) and the displacement vector S.In addition, the DCT of step S201 handles and not necessarily will carry out all block of pixels of being cut apart, for example also can extract 8 at each pixel region As inner evaluation value C _rTime point end process more than or equal to the block of pixels of setting.And, in this case, in order to extract object pixel piece Bi from object images Gt equably, for example as shown in Figure 8, can be being divided into central portion and 4 these 5 pixel region A1～A5 of bight in the object images Gt.In addition, the numerical value of the size of each pixel region of expression shown in Figure 8 A1～A5 is pixel region A1～A5 of an example set to(for) the object images Gt that resolution is made as VGA, and the size of each pixel region A1～A5 is not limited thereto.

Then, the variation of the processing of the weight setting shown in the description of step S111.In above-mentioned weight setting is handled,, set the weighting coefficient w of each object pixel piece Bi according to the pixel region As that includes each object pixel piece Bi _i, but also can be according to the mobile vector V of each object pixel piece Bi _iSize set weighting coefficient w _iUnder this situation, the 3f of weight setting portion for example can be according to w _i=V _iV _i=| V _i| ²Obtain the weighting coefficient w of each object pixel piece Bi _iIt is mobile vector V that this weight setting is handled _iThe big more mobile vector V that then is judged as _iWith respect to the high more processing of image blurring degree of confidence, for example more effective when the trickle object images Gt that moves more subject such as rocking of correcting captured leaf fuzzy.Under this situation, can reduce mobile vector V with the mobile same degree ground of subject _iThe weighting coefficient w of less object pixel piece Bi _i, can reduce the mobile arithmetic eror that causes by subject.

In addition, can be corresponding to the mobile vector V of the pixel region As that includes each object pixel piece Bi and each object pixel piece Bi _iThis two side of size, for example according to following numerical expression (15), set weighting coefficient w _i

w _i＝w _area+(V _i·V _i)/k ² ……(15)

Wherein, the weighting coefficient corresponding to pixel region As is made as w _Area, being used for mobile vector V _iThe coefficient of size criteriaization (normalized) be made as k.Coefficient k is to be that mobile vector V supposes in unit with the pixel count _iThe peaked numerical value of size.For example, when violent sports etc. are moved in photography, mobile vector V _iSize to become big possibility higher, so set the value of coefficient k greatlyyer, when photography landscape etc., mobile vector V _iSize to become big possibility lower, so set the value of coefficient k lessly, can suitably change according to subject.

In addition, as another variation that the weight setting shown in the step S111 is handled, can be with the mobile vector V of object pixel piece Bi _iWith each adjacent pixel blocks Bij adjacent with object pixel piece Bi (j=1,2 ..., m) mobile vector V _Ij(j=1,2 ..., m) be the basis, mobile vector V is represented in computing _IjWith respect to mobile vector V _iThe divergence d of extent of deviation _i, according to this divergence d _iSet weighting coefficient w _iHerein, divergence d _iIt is characteristic value according to the computing of following numerical expression (16) institute.

[formula 14]

$d_i=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(V_i^{\#}\&CenterDot;V_{\mathrm{ij}})}^2\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(16\right)$

Wherein, # represents the rotational transform of anticlockwise 90 degree.And, the number of adjacent pixel blocks Bij is made as m.Divergence d according to this numerical expression (16) computing _i, at each mobile vector V _IjWith respect to mobile vector V _iThe deviation of directivity bigger value of indication when big, on the contrary, at mobile vector V _iWith each mobile vector V _IjDirection less value of indication when consistent.Particularly at mobile vector V _iWith each mobile vector V _IjDirection when all equating, divergence d _iValue be zero.

According to this divergence d _iSet weighting coefficient w _iWeight setting handle, the 3f of weight setting portion is at divergence d _iMore hour weighting coefficient w _iSet greatly more.That is near high more, the divergence d of unitarity of the moving direction in the zone, around object pixel piece Bi and its _iMore hour, the 3f of weight setting portion is judged as mobile vector V _iHigh more with respect to image blurring degree of confidence.If adopt this weight setting to handle, then can alleviate moving and the mobile vector V of the part that the noise of picture signal etc. causes by subject _iThe influence of arithmetic eror, computing rotation matrix R (φ) and displacement vector S accurately, correctly correcting image blurs.

Herein, illustrate according to divergence d _iSet weighting coefficient w _iAn example of the treatment step handled of weight setting.Fig. 9 is the process flow diagram of representing the treatment step after the weight setting processing subroutinization of step S111 shown in Figure 4.As shown in Figure 9, under the control of Flame Image Process control part 3a, the 3f of weight setting portion selects to set weighting coefficient w _iObject pixel piece Bi (step S301).And, extract the adjacent pixel blocks Bij (step S303) adjacent with selected object pixel piece Bi.The 3f of weight setting portion obtains the mobile vector V at the object pixel piece Bi of step S109 institute computing _i(step S305), the mobile vector V of each adjacent pixel blocks of computing Bij _Ij(step S307) is according to the divergence d of numerical expression (16) computing adjacent pixel blocks Bij with respect to object pixel piece Bi _i(step S309).Then, the 3f of weight setting portion is according to the set positions divergence d of object pixel piece Bi _iThreshold value d _Max(step S311), relatively divergence d _iWith threshold value d _MaxSize (step S313), at divergence d _iLess than threshold value d _MaxThe time (step S313 is a "Yes"), set the weighting coefficient w corresponding with the position of object pixel piece Bi _Area(step S315) is according to the weighting coefficient w of numerical expression (15) operand block of pixels Bi _i(step S317), and return step S111.On the other hand, at divergence d _iBe not less than threshold value d _MaxThe time (step S313 is a "No"), the 3f of weight setting portion sets the weighting coefficient w corresponding with the position of object pixel piece Bi _Area(step S319) is according to w _i=w _AreThe weighting coefficient w of/2 operand block of pixels Bi _i(step S321), and return step S111.

In step S303, the 3f of weight setting portion extracts adjacent pixel blocks Bij for example as shown in figure 10.Promptly, for object pixel piece B4, B5, B6 as object pixel piece Bi, with the neighboring of object images Gt non-conterminous object pixel piece B4 in, extract and 8 adjacent pixel blocks B41～B48 around the identical size of object pixel piece B4, in the object pixel piece B5 adjacent with the edge of object images Gt, 5 adjacent pixel blocks B51～B55 around extracting, the object pixel piece B6 in the bight that is arranged in object images Gt, 3 adjacent pixel blocks B61～B63 around extracting.In addition, the 3f of weight setting portion also can not extract adjacent pixel blocks Bij, though and extraction is positioned near block of pixels on every side as near block of pixels with object pixel piece Bi is non-conterminous.And Flame Image Process control part 3a also can make block of pixels extraction unit 3d extract adjacent pixel blocks Bij, and the 3f of weight setting portion obtains it and extracts the result.

In step S311, the 3f of weight setting portion is according to the number m of the adjacent pixel blocks Bij that extracts at step S303 and setting threshold d _Max, for example according to d _Max=m32 ²Setting threshold d _MaxSpecifically, for object pixel piece B4, B5, B6 shown in Figure 10, respectively threshold value d _MaxBe set at d _Max=832 ², 532 ², 332 ²In addition, threshold value d _MaxEstablishing method be not limited thereto, also can set according to other conditional corresponding with the number m of adjacent pixel blocks Bij, can also not use the information of adjacent pixel blocks Bij such as number m to set.And, also can suitably change threshold value d according to imaging conditions of object images etc. _MaxImpose a condition.

In addition, in step S313, the 3f of weight setting portion can also be according to divergence d _iWith threshold value d _MaxRelation, judge more branch condition.And, the weighting coefficient w that in step S317, S321, uses _iThe computing formula be not limited to above-mentioned computing formula, also can use other computing formula.In addition, in step S307, Flame Image Process control part 3a also can make mobile vector operational part 3e computing mobile vector V _Ij, the 3f of weight setting portion obtains its operation result.

And, as other variation that the weight setting shown in the step S111 is handled, also can be according to the picture contrast Con of object pixel piece Bi _iSet weighting coefficient w _iCan utilize the whole bag of tricks to come computing, evaluation map image contrast Con herein, _i, for example, the 3f of weight setting portion also can be according to (17) computing of following numerical expression and evaluation map image contrast Con _i

[formula 15]

${\mathrm{Con}}_i=\frac{1}{64}\&CenterDot;\underset{p=1}{\overset{8}{\mathrm{\&Sigma;}}}\underset{q=1}{\overset{7}{\mathrm{\&Sigma;}}}(|f_{q+1,p}-f_{q,p}|+|f_{p,q+1}-f_{p,q}|)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(17\right)$

Herein, object pixel piece Bi is the block of pixels of 8 * 8 pixels, and each pixel data of this object pixel piece is made as f _Pq(p, q=1,2 ..., 8).

And the 3f of weight setting portion also can utilize the evaluation of estimate C shown in the numerical expression (14) _rBe set as Con _i=C _r, come evaluation map image contrast Con _iUnder this situation, can be shared as DCT coefficient D the result that the DCT that carries out in general dynamic image compressed encodings such as MPEG4 dynamic image compression are handled handles _Uv, can promote the summary of bulk treatment and rapid.In addition, as evaluation of estimate C _rLike that, utilizing DCT coefficient D _UvThe represented evaluation of estimate of radio-frequency component be also used as the technology of the evaluation of estimate of picture contrast, for example in patent documentation 2, be disclosed.

In addition, the 3f of weight setting portion also can utilize the handled picture contrast of the autofocus mechanism that is had by existing camera head, comes the picture contrast Con of evaluation object image B i _iUnder this situation, also can be between different processing shared evaluation of estimate, can promote the summary of bulk treatment and rapid.

Like this according to picture contrast Con _iSet weighting coefficient w _iWeight setting handle, the 3f of weight setting portion is at the picture contrast Con of object pixel piece Bi _iWhen big more, weighting coefficient w _iSet greatly more.That is, the 3f of weight setting portion is at picture contrast Con _iBe judged as mobile vector V when big more _iHigh more with respect to image blurring degree of confidence.If use this weight setting to handle, then can be at the high more object pixel piece Bi of the process sensitivity in the piece matching treatment, weighting coefficient w _iSet greatly more, computing rotation matrix R (φ) and displacement vector S accurately, correctly correcting image blurs.

Herein, illustrate according to picture contrast Con _iSet weighting coefficient w _iAn example of the treatment step handled of weight setting.Figure 11 is the process flow diagram of representing the treatment step after the weight setting processing subroutinization of step S111 shown in Figure 4.As shown in figure 11, under the control of Flame Image Process control part 3a, the 3f of weight setting portion selects to set weighting coefficient w _iObject pixel piece Bi (step S401), the picture contrast Con of the selected object pixel piece of computing Bi _i(step S403) is according to w _i=gCon _iComputing weighting coefficient w _i(step S405), and return step S111.Herein, can coefficient g be set at be about g=1/4 greater than zero real number value.But, be not limited thereto, for example also can be that coefficient g is set on the basis with all characteristic values of object pixel piece Bi, also can be set at functional value to coefficient g.

And,, also can judge mobile vector V according to the importance degree of each object pixel piece Bi with respect to the operational precision of rotation matrix R (φ) and displacement vector S as other variation of handling at the weight setting shown in the step S111 _iWith respect to image blurring degree of confidence, be that weighting coefficient w is set on the basis with this degree of confidence _iHerein, as each mobile vector V that in computing, uses _iDegree of confidence when equating, then the space of Dui Ying each object pixel piece Bi is big more, the operational precision of the rotation matrix R (φ) and the vector S that is shifted is high more.For this reason, the big object pixel piece Bi in space is high more for the importance degree of operational precision, and on the contrary, the object pixel piece Bi that its importance degree is high is weighting coefficient w _iSet greatly more, computing rotation matrix R (φ) and displacement vector S accurately thus, correctly correcting image blurs.

In addition, in the weight setting shown in the step S111 was handled, the setting processing method that is not limited to herein illustrate also can use various disposal routes.And, can also make up various setting processing methods, judge each mobile vector V _iDegree of confidence, set weighting coefficient w _i

In the camera head 1 that this embodiment 1 of above explanation relates to, alternative image from a plurality of images, selected object images is divided into pixel region, extract a plurality of object pixel pieces from the pixel region of being cut apart, various characteristics value based on the object pixel piece that extracted, judge that mobile vector is with respect to image blurring degree of confidence, based on this degree of confidence, set the weighting coefficient of each block of pixels, carry out the calculation process of the mobile vector corresponding with degree of confidence, so conversion coefficient of computing and the fuzzy corresponding affined transformation of object images accurately always, can be based on the conversion coefficient of institute's computing, object images is carried out affined transformation, and correctly the calibration object image is fuzzy.And, because various characteristics value based on the object pixel pieces such as divergence of the mobile vector of the relevance of ratio, picture contrast, expression and near the block of pixels of the radio-frequency component of the size of the position of object pixel piece, mobile vector, image, carry out the judgement of the degree of confidence of the extraction of object pixel piece and mobile vector, so can be according to the various changes and the state of subject or shot object image, the fuzzy and correction of detected image reliably accurately always.

(embodiment 2)

Below, embodiments of the present invention 2 are described.In above-mentioned embodiment 1, extract a plurality of object pixel piece Bi from object images being cut apart each pixel region As that obtains, the conversion coefficient of computing and the fuzzy corresponding affined transformation of image accurately, but in this embodiment 2, extract two object pixel piece Bi from 4 folding corner regions of object images, can be simply and the conversion coefficient of computing affined transformation apace.

Figure 12 is the synoptic diagram of an example of the expression state that extracted two object pixel pieces from the folding corner region of object images.As shown in figure 12, in this embodiment 2, from 4 folding corner regions of object images Gt is bottom right and upper left folding corner region at figure, extracts object pixel piece BS1, BS2 as object pixel piece Bi respectively.And, the central point Os of the object pixel piece BS1 that is extracted is set at true origin with respect to this object images Gt, the position vector of indicated object block of pixels BS2 is made as X _S2, the mobile vector separately of object pixel piece BS1, BS2 is made as V _S1, V _S2In addition, the position vector of indicated object block of pixels BS1 is a zero vector.

When the weighting coefficient of object pixel piece BS1, BS2 is made as w respectively _S1, w _S2The time, the summation w of weighting coefficient is w=w _S1+ w _S2, position center of gravity vector X _cAccording to numerical expression (7) is X _c=w _S2X _S2/ w, the vector V of swerving the weight _cAccording to numerical expression (8) is V _c=(V _S1+ w _S2V _S2)/w.In weight setting was handled, the 3f of weight setting portion was each weighting coefficient w _S1, w _S2Size be made as

w _S2" 1, respectively infinitely near " 1 " and " 0 ".Under this situation, the summation w of weighting coefficient is approximately

Position center of gravity vector X _cBe approximately zero vector, the vector V of swerving the weight _cBe approximately

Thus, the rotationangle of rotation matrix R (φ) and displacement vector S utilize the following numerical expression (18) and the numerical expression (19) that have used numerical expression (5) and numerical expression (6) to represent respectively.

S＝V _S1 ……(18)

[formula 16]

$\mathrm{\&phi;}=\arctan \frac{{-X}_{S2}\&CenterDot;{({-\mathrm{<figure-callout\; id="1"\; label="V\; S"\; filenames="C200610001116E00381.png"\; state="\{\{state\}\}">V</figure-callout>}}_S+V_{S2})}^{\#}}{X_{S2}\&CenterDot;\{X_{S2}-({-\mathrm{<figure-callout\; id="1"\; label="V\; S"\; filenames="C200610001116E00381.png"\; state="\{\{state\}\}">V</figure-callout>}}_S+V_{S2})\}}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(19\right)$

Like this, in this embodiment 2, two folding corner regions that are in angular dependence from 4 folding corner regions of object images Gt extract object pixel piece BS1, BS2.And, the center of one of them object pixel piece BS1 is set at true origin, make the weighting coefficient w of object pixel piece BS1, BS2 _S1, w _S2Infinitely near " 1 " and " 0 ".Like this, compare with embodiment 1, can use the rotation matrix R (φ) of simple calculations formula and the calculation process of displacement vector S, the result can carry out calculation process more quickly.This ambiguity correction is fast handled, and for example the situation of the dynamic image in utilizing display monitor 5a observation shooting shown in Figure 2 is inferior, and the situation fuzzy for needs continuous correction in real time is very effective.

In addition, in the weight setting of this embodiment 2 was handled, the 3f of weight setting portion paid attention to setting for the object pixel piece BS1 of true origin, sets its importance degree with respect to calculation process higherly, according to this importance degree mobile vector V _S1Degree of confidence be judged as higherly, as its result, make each weighting coefficient w _S1, w _S2Infinitely near " 1 " and " 0 ".

And, object pixel piece BS1, BS2 shown in Figure 12 extracted from the bottom right that is in diagonal position in object images Gt and upper left folding corner region, but the extracting method of object pixel piece Bi is not limited thereto, and two folding corner regions arbitrarily also can 4 folding corner regions in object images Gt respectively extract an object pixel piece Bi.That is, two object pixel piece Bi that extracted are diagonal position not necessarily, also can be in horizontal direction or longitudinal direction position relation arranged side by side in object images Gt.

In addition, in above-mentioned embodiment 1 and 2, carried out utilizing the ambiguity correction of the image that camera head 1 made a video recording, but also can carry out from the image of outside input or be stored in the ambiguity correction of the image the storage part 8.And, also can be applied to the image processing apparatus such as computing machine different to blur correcting method of the present invention with camera head 1, carry out importing the ambiguity correction of the image of this image processing apparatus from the outside.Under this situation, for example for the image that utilizes camera head 1 to be made a video recording, also can use ambiguity correction under the situation of exporting to display monitor 5a handles, do not use ambiguity correction under the outside situation and handle outputing to by portable storage media 9 or Department of Communication Force 7, and be output as image to the image processing apparatus of external device (ED) on carry out ambiguity correction and handle.Under this situation, can alleviate the processing load of image processing part 3, make processing rapid.

And above-mentioned embodiment 1 and 2 various ambiguity corrections are handled also and can be selected switching at any time based on the model selection information of handling from the ambiguity correction of inputs such as input part 4.

Claims (32)

1. blur correcting method comprises:

Image selection step from a plurality of images by the camera head shooting, is selected to object images and the benchmark image that becomes the benchmark of this correction to the image blurring object of proofreading and correct;

The piece extraction step is from least two object pixel pieces of described object picture extraction;

Vector calculus step, computing mobile vector, described mobile vector are represented the amount of movement between the position of position on the described benchmark image corresponding with the position of the object pixel piece that extracts by described extraction step and this each object pixel piece,

According to proofreading and correct the fuzzy of described object images by the mobile vector of described vector calculus step operation,

It is characterized in that described blur correcting method comprises:

The weight setting step with respect to described image blurring degree of confidence, is set the weighting coefficient of described each object pixel piece based on the mobile vector by described vector calculus step operation;

The coefficient calculation step, based on the true origin of regulation as the position vector of described each object pixel piece of benchmark, described each mobile vector and each weighting coefficient of setting by described weight setting step, computing is used for described object images is carried out the conversion coefficient of affined transformation; And

The correction calculation step based on the operation result of described coefficient calculation step, is carried out affined transformation to described object images, to proofread and correct the fuzzy of this object images.

2. blur correcting method according to claim 1, it is characterized in that, described weight setting step based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

3. blur correcting method according to claim 1 is characterized in that, described object images is divided into a plurality of pixel regions, and perhaps described object images is the image that is split into a plurality of pixel regions in advance,

Described at least two described object pixel pieces of described a plurality of pixel regions extractions that extraction step has obtained from having carried out cutting apart.

4. blur correcting method according to claim 3, it is characterized in that, described weight setting step based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

5. blur correcting method according to claim 4, it is characterized in that described weight setting step handle is set to such an extent that ratio is big at the described weighting coefficient of the described object pixel piece of the substantial middle portion that is positioned at described object images at the described weighting coefficient of the described object pixel piece of the periphery that is positioned at described object images.

6. blur correcting method according to claim 1, it is characterized in that, described weight setting step is set the described weighting coefficient at described object pixel piece based on based on the direction of described mobile vector and the described degree of confidence of at least one side in the size.

7. blur correcting method according to claim 1, it is characterized in that, the contrast of the image in the described object pixel piece of described weight setting step operation, based on the described degree of confidence based on the operation result of this contrast, setting is at the described weighting coefficient of described object pixel piece.

8. blur correcting method according to claim 1, it is characterized in that, described weight setting step operation is positioned near near the mobile vector of the block of pixels of described object pixel piece, with the described degree of confidence based on divergence is that the described weighting coefficient at described object pixel piece is set on the basis, and described divergence is represented near the extent of deviation of the mobile vector of described block of pixels with respect to the mobile vector of described object pixel piece.

9. blur correcting method according to claim 2, it is characterized in that, described weight setting step is set described weighting coefficient at described object pixel piece according to the 1st weighting coefficient and divergence, the position of the described object pixel piece in described the 1st weighting coefficient and the described object images is corresponding, and described divergence represents to be positioned near near the mobile vector of block of pixels of described object pixel piece with respect to the extent of deviation of the mobile vector of described object pixel piece.

10. blur correcting method according to claim 1 is characterized in that, described coefficient calculation step is made as X respectively at handle described each position vector corresponding with described each object pixel piece, described each mobile vector and described each weighting coefficient _i, V _iAnd w _i, wherein, i=1,2 ..., n, the summation of this each weighting coefficient is made as under the situation of w, according to X _c=(w ₁X ₁+ w ₂X ₂+ ... + w _nX _n)/w comes computing to represent the position center of gravity vector X of the weighting center of gravity of described each position vector _c, according to V _c=(w ₁V ₁+ w ₂V ₂+ ... + w _nV _n)/w comes computing to represent the vector V of swerving the weight of the weighting center of gravity of described each mobile vector _c, come the rotationangle of the computing rotation matrix corresponding according to formula 1 with the linear transformation composition of described conversion coefficient,

[formula 1]

$\mathrm{\&phi;}=\arctan \frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{(-V_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}$

Wherein, # represents the rotational transform of anticlockwise 90 degree.

11. blur correcting method according to claim 10 is characterized in that, described coefficient calculation step is according to S=X _c-V _cCome the computing displacement vector S corresponding with the displacement composition of described conversion coefficient, described correction calculation step is middle mind-set counterclockwise just rotation described rotation angle with described object images with the weighting center of gravity of described position vector according to described rotation matrix, makes this object images displacement according to described displacement vector then.

12. blur correcting method according to claim 10 is characterized in that, described extraction step and described weight setting step are extracted described object pixel piece respectively and are set described weighting coefficient, so that described position center of gravity vector becomes zero vector.

13. blur correcting method according to claim 11 is characterized in that, described extraction step and described weight setting step are extracted described object pixel piece respectively and are set described weighting coefficient, so that described position center of gravity vector becomes zero vector.

14. blur correcting method according to claim 10 is characterized in that, described coefficient calculation step is being made as R (φ) to the described rotation matrix that counterclockwise carries out the rotational transform of described rotationangle just, according to S=R (φ) (X _c-V _c)-X _cCome the computing displacement vector S corresponding with the displacement composition of described conversion coefficient,

After described correction calculation step makes described object images displacement according to described displacement vector, be that middle mind-set is counterclockwise rotated described rotation angle just with this object images with described true origin according to described rotation matrix.

15. blur correcting method according to claim 10 is characterized in that, described coefficient calculation step is come the described rotationangle of computing according to formula 1 having been carried out the approximate formula 2 that obtains under the situation of size less than setting of described each mobile vector,

[formula 2]

$\mathrm{\&phi;}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{(-V_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}.$

16. blur correcting method according to claim 10 is characterized in that, described extraction step extracts described object pixel piece respectively from two these folding corner regions folding corner region, different in each bight of being arranged in described object images,

Described coefficient calculation step as described true origin, is made as X to the position of one of them the object pixel piece the object pixel piece that extracts from described two folding corner regions to described position vector, described mobile vector and the described weighting coefficient corresponding with the described object pixel piece that is positioned at this true origin respectively ₁, V ₁And w ₁By described weight setting step, at the described weighting coefficient of the object pixel piece that is positioned at described true origin and at the described weighting coefficient of another object pixel piece unlimited respectively near 1 and 0 situation under, come the described rotationangle of computing according to formula 1 having been carried out the approximate formula 3 that obtains

[formula 3]

$\mathrm{\&phi;}=\arctan \frac{{-X}_2\&CenterDot;{(-V_1+V_2)}^{\#}}{X_2\&CenterDot;\{X_2-({-V}_1+V_2)\}}.$

17. blur correcting method according to claim 1 is characterized in that, described a plurality of images are dynamic images,

Described each object pixel piece is the piece that becomes the process object of the inter prediction processing of carrying out in the compressed encoding of described dynamic image is handled,

Described each mobile vector in described inter prediction is handled by computing.

18. blur correcting method according to claim 10 is characterized in that, described a plurality of images are dynamic images,

Described each object pixel piece is the piece that becomes the process object of the inter prediction processing of carrying out in the compressed encoding of described dynamic image is handled,

Described each mobile vector in described inter prediction is handled by computing.

19. blur correcting method according to claim 1 is characterized in that, described extraction step extracts described object pixel piece respectively from two different these folding corner regions of the folding corner region in each bight of being arranged in described object images,

Described coefficient calculation step is made as V to corresponding described mobile vector the initial point of the position of one of them the object pixel piece the object pixel piece that extracts from described two folding corner regions as rotation ₁, described position vector corresponding with the one other pixel piece and described mobile vector are made as X ₂And V ₂Situation under, according to X ₂(V ₁-V ₂) and X ₂{ X ₂-(V ₁-V ₂) rotation angle of coming the computing rotation matrix corresponding with the linear transformation composition of described conversion coefficient.

20. a camera head comprises:

To assemble the also image pickup optical system of imaging shot object image from the light of subject;

The make a video recording image unit of the image corresponding with described shot object image;

The image selected cell, it is selected to the object images and the benchmark image that becomes the benchmark of this correction of the fuzzy object of correcting image from a plurality of images of being made a video recording by described image unit;

The piece extraction unit, it is from least two object pixel pieces of described object picture extraction;

The amount of movement between the position of position on the described benchmark image corresponding with the position of the object pixel piece that extracts by described extraction unit and this object pixel piece is represented in vector calculus unit, its computing mobile vector, this mobile vector,

According to proofread and correct the fuzzy of described object images by the mobile vector of described vector calculus unitary operation, it is characterized in that described camera head comprises:

The weight setting unit, it with respect to described image blurring degree of confidence, sets the weighting coefficient of described each object pixel piece based on each mobile vector by described vector calculus unitary operation;

The coefficient arithmetic element, it is the position vector of described each object pixel piece of benchmark, described each mobile vector and each weighting coefficient of setting by described weight setting unit according to the true origin with regulation, and computing is carried out the conversion coefficient that affined transformation is used to described object images; And

Correction computing unit, it carries out affined transformation based on the operation result of described coefficient arithmetic element to described object images, to proofread and correct the fuzzy of this object images.

21. camera head according to claim 20, it is characterized in that, described weight setting unit based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

22. camera head according to claim 20 is characterized in that, described object images is divided into a plurality of pixel regions, perhaps described object images is the image that is split into a plurality of pixel regions in advance,

Described extraction unit extracts at least two described object pixel pieces from divided described a plurality of pixel regions.

23. camera head according to claim 22, it is characterized in that, described weight setting unit based on described object images in the corresponding described degree of confidence in position of described object pixel piece, set described weighting coefficient at described object pixel piece.

24. camera head according to claim 20 is characterized in that, the described weighting coefficient at described object pixel piece is set based on based on the direction of described mobile vector and the described degree of confidence of at least one side in the size in described weight setting unit.

25. camera head according to claim 20 is characterized in that, the described weighting coefficient at described object pixel piece is set based on the described degree of confidence based on the contrast of the image in the described object pixel piece in described weight setting unit.

26. camera head according to claim 20, it is characterized in that, described weight setting unitary operation is positioned near near the mobile vector of the block of pixels of described object pixel piece, with the described degree of confidence based on divergence is that the described weighting coefficient at described object pixel piece is set on the basis, and described divergence represents to be positioned near near the described mobile vector of block of pixels of described object pixel piece with respect to the extent of deviation of the mobile vector of described object pixel piece.

27. camera head according to claim 21, it is characterized in that, described weighting coefficient at described object pixel piece is set according to the 1st weighting coefficient and divergence in described weight setting unit, the position of the described object pixel piece in described the 1st weighting coefficient and the described object images is corresponding, and described divergence represents to be positioned near near the mobile vector of block of pixels of described object pixel piece with respect to the extent of deviation of the mobile vector of described object pixel piece.

28. camera head according to claim 20 is characterized in that, described coefficient arithmetic element is made as X respectively at handle described each position vector corresponding with described each object pixel piece, described each mobile vector and described each weighting coefficient _i, V _iAnd w _i, wherein, i=1,2 ..., n, the summation of this each weighting coefficient is made as under the situation of w, according to X _c=(w ₁X ₁+ w ₂X ₂+ ... + w _nX _n)/w comes computing to represent the position center of gravity vector X of the weighting center of gravity of described each position vector _c, according to V _c=(w ₁V ₁+ w ₂V ₂+ ... + w _nV _n)/w comes computing to represent the vector V of swerving the weight of the weighting center of gravity of described each mobile vector _c, come the rotationangle of the computing rotation matrix corresponding according to formula 4 with the linear transformation composition of described conversion coefficient,

[formula 4]

$\mathrm{\&phi;}=\arctan \frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{(-V_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}$

Wherein, # represents the rotational transform of anticlockwise 90 degree.

29. camera head according to claim 28 is characterized in that, described coefficient arithmetic element is according to S=X _c-V _cCome the computing displacement vector S corresponding with the displacement composition of described conversion coefficient,

Described correction computing unit is middle mind-set counterclockwise just rotation described rotation angle with described object images with the weighting center of gravity of described position vector according to described rotation matrix, makes this object images displacement according to described displacement vector then.

30. camera head according to claim 28 is characterized in that, described extraction unit and described weight setting unit extract described object pixel piece respectively and set described weighting coefficient, so that described position center of gravity vector becomes zero vector.

31. camera head according to claim 28 is characterized in that, described coefficient arithmetic element is come the described rotationangle of computing according to formula 4 having been carried out the approximate formula 5 that obtains under the situation of size less than setting of described each mobile vector,

[formula 5]

$\mathrm{\&phi;}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;{(-V_i+V_c)}^{\#}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i(X_i-X_c)\&CenterDot;\{(X_i-X_c)+({-V}_i+V_c)\}}.$

32. camera head according to claim 28 is characterized in that, described extraction unit extracts described object pixel piece respectively from two different these folding corner regions of the folding corner region in each bight of being arranged in described object images,

Described coefficient arithmetic element as described true origin, is made as X to the position of one of them the object pixel piece the object pixel piece that extracts from described two folding corner regions to described position vector, described mobile vector and the described weighting coefficient corresponding with the described object pixel piece that is positioned at this true origin respectively ₁, V ₁And w ₁By described weight setting unit, at the described weighting coefficient of the object pixel piece that is made as described true origin and at the described weighting coefficient of another object pixel piece unlimited respectively near 1 and 0 situation under, come the described rotationangle of computing according to formula 4 having been carried out the approximate formula 6 that obtains

[formula 6]

$\mathrm{\&phi;}=\arctan \frac{{-X}_2\&CenterDot;{(-V_1+V_2)}^{\#}}{X_2\&CenterDot;\{X_2-({-V}_1+V_2)\}}.$

Images