CN105930853A - Automatic image capturing device for content generation - Google Patents
Automatic image capturing device for content generation Download PDFInfo
- Publication number
- CN105930853A CN105930853A CN201610231049.7A CN201610231049A CN105930853A CN 105930853 A CN105930853 A CN 105930853A CN 201610231049 A CN201610231049 A CN 201610231049A CN 105930853 A CN105930853 A CN 105930853A
- Authority
- CN
- China
- Prior art keywords
- point
- image
- line segment
- module
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
Abstract
The invention discloses an automatic image capturing device for content generation. The device comprises an image preprocessing module, an image extreme point detection module, an image characteristic point positioning module, a main direction determining module, a feature extraction module and a content generation module, wherein the image characteristic point positioning module determines extreme points as characteristic points by rejecting noise sensitive low-contract points and instable edge points from different extreme points; and the main direction determining module connects two random adjacent peak values, in a gradient direction histogram related to the characteristic points, into a line to form multiple sub line segments, merges the adjacent sub line segments with similar gradients into line segments, and uses the direction of an optimal line segment among the multiple line segments as the main direction of the characteristic points. The device has the advantages that content generation is high in precision and speed.
Description
Technical field
The present invention relates to art of image analysis, be specifically related to a kind of automatic image capture device for generating content.
Background technology
In correlation technique, the most only image itself is identified, and does not generate its content.It is desirable that machine can not only
Target is detected, additionally it is possible to as people, image is understood.Additionally, in order to substantial amounts of view data is processed,
Need to improve analyzing and processing efficiency and precision.
Summary of the invention
For the problems referred to above, the present invention provides a kind of automatic image capture device for generating content.
The purpose of the present invention realizes by the following technical solutions:
Provide a kind of automatic image capture device for generating content, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring
When the value of 18 points that yardstick is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 of yardstick
When the value of 18 points that consecutive points are corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described difference of Gaussian chi
The reduced mechanical model in degree space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, positions for extreme point pinpoint first including be sequentially connected with
Submodule, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point,
Wherein:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalized successively to the image exported soon by image conversion submodule
Rear rejecting described low contrast point, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels higher than 128 of the gray value in image
Average, mLIt is the gray value average that is less than all pixels of 128, ψ (x, y) is the image after being processed by image filtering submodule,
T1For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed;
Preferably, the described automatic image capture device for generating content, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
Further, the sub-line section described in close slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3's
Span be (0,0.1].
The invention have the benefit that
1, the image pre-processing module arranged considers visual custom and the human eye perceptibility to different color with colouring intensity
Non-linear relation, it is possible to describe image the most accurately;
2, propose the reduced mechanical model of Gaussian difference scale space, decrease operand, improve arithmetic speed, Jin Erti
The high speed of graphical analysis;
3, the image characteristic point locating module arranged carries out low contrast point and the removal of mobile rim point to extreme point, it is ensured that special
Levy effectiveness a little, wherein the gray value of image is strengthened, it is possible to be greatly increased the stability of image, the most right
Low contrast point is removed, and then improves the accuracy of graphical analysis;
4, principal direction is set and determines module, it is proposed that the judgement formula of optimum line segment, with appointing in characteristic point gradient orientation histogram
The direction of the optimum line segment in the line segment that adjacent two peak value lines of anticipating are formed is as the principal direction of characteristic point, and line segment is relative to point more
Add stable so that the descriptor of image characteristic of correspondence point has repeatability, improves the accuracy of feature descriptor, and then
More fast and accurately image can be identified detection, there is the highest robustness.
Accompanying drawing explanation
The invention will be further described to utilize accompanying drawing, but the embodiment in accompanying drawing does not constitute any limitation of the invention, for
Those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to obtains the attached of other according to the following drawings
Figure.
Fig. 1 is the connection diagram of each module of the present invention.
Detailed description of the invention
The invention will be further described with the following Examples.
Embodiment 1
Seeing Fig. 1, the present embodiment is used for generating the automatic image capture device of content, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring
When the value of 18 points that yardstick is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 of yardstick
When the value of 18 points that consecutive points are corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described difference of Gaussian chi
The reduced mechanical model in degree space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, positions for extreme point pinpoint first including be sequentially connected with
Submodule, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point,
Wherein:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalized successively to the image exported soon by image conversion submodule
Rear rejecting described low contrast point, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels higher than 128 of the gray value in image
Average, mLIt is the gray value average that is less than all pixels of 128, ψ (x, y) is the image after being processed by image filtering submodule,
T1For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed;
Preferably, the described automatic image capture device for generating content, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
Further, the sub-line section described in close slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3's
Span be (0,0.1].
It is strong with color that the image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color
The non-linear relation of degree, it is possible to describe image the most accurately;Propose the reduced mechanical model of Gaussian difference scale space, subtract
Lack operand, improve arithmetic speed, and then improve the speed of graphical analysis;The image characteristic point locating module pair arranged
Extreme point carries out low contrast point and the removal of mobile rim point, it is ensured that the effectiveness of characteristic point, the wherein gray value to image
Strengthen, it is possible to be greatly increased the stability of image, the most accurate low contrast point is removed, and then improve image
The accuracy analyzed;Principal direction is set and determines module, it is proposed that the judgement formula of optimum line segment, with characteristic point gradient direction Nogata
The direction of the optimum line segment in the line segment that two peak value lines of the arbitrary neighborhood in figure are formed is as the principal direction of characteristic point, line segment phase
More stable for point so that the descriptor of image characteristic of correspondence point has repeatability, improves the accurate of feature descriptor
Property, and then can more fast and accurately image be identified detection, there is the highest robustness;The present embodiment takes threshold value
T1=0.01, T2=10, T3=0.1, the precision generating content improves 2%, and speed improves 1%.
Embodiment 2
Seeing Fig. 1, the present embodiment is used for generating the automatic image capture device of content, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring
When the value of 18 points that yardstick is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 of yardstick
When the value of 18 points that consecutive points are corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described difference of Gaussian chi
The reduced mechanical model in degree space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, positions for extreme point pinpoint first including be sequentially connected with
Submodule, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point,
Wherein:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalized successively to the image exported soon by image conversion submodule
Rear rejecting described low contrast point, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels higher than 128 of the gray value in image
Average, mLIt is the gray value average that is less than all pixels of 128, ψ (x, y) is the image after being processed by image filtering submodule,
T1For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed;
Preferably, the described automatic image capture device for generating content, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
Further, the sub-line section described in close slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3's
Span be (0,0.1].
It is strong with color that the image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color
The non-linear relation of degree, it is possible to describe image the most accurately;Propose the reduced mechanical model of Gaussian difference scale space, subtract
Lack operand, improve arithmetic speed, and then improve the speed of graphical analysis;The image characteristic point locating module pair arranged
Extreme point carries out low contrast point and the removal of mobile rim point, it is ensured that the effectiveness of characteristic point, the wherein gray value to image
Strengthen, it is possible to be greatly increased the stability of image, the most accurate low contrast point is removed, and then improve image
The accuracy analyzed;Principal direction is set and determines module, it is proposed that the judgement formula of optimum line segment, with characteristic point gradient direction Nogata
The direction of the optimum line segment in the line segment that two peak value lines of the arbitrary neighborhood in figure are formed is as the principal direction of characteristic point, line segment phase
More stable for point so that the descriptor of image characteristic of correspondence point has repeatability, improves the accurate of feature descriptor
Property, and then can more fast and accurately image be identified detection, there is the highest robustness;The present embodiment takes threshold value
T1=0.02, T2=11, T3=0.08, the precision generating content improves 1%, and speed improves 1.5%.
Embodiment 3
Seeing Fig. 1, the present embodiment is used for generating the automatic image capture device of content, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring
When the value of 18 points that yardstick is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 of yardstick
When the value of 18 points that consecutive points are corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described difference of Gaussian chi
The reduced mechanical model in degree space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, positions for extreme point pinpoint first including be sequentially connected with
Submodule, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point,
Wherein:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalized successively to the image exported soon by image conversion submodule
Rear rejecting described low contrast point, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels higher than 128 of the gray value in image
Average, mLIt is the gray value average that is less than all pixels of 128, ψ (x, y) is the image after being processed by image filtering submodule,
T1For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed;
Preferably, the described automatic image capture device for generating content, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
Further, the sub-line section described in close slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3's
Span be (0,0.1].
It is strong with color that the image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color
The non-linear relation of degree, it is possible to describe image the most accurately;Propose the reduced mechanical model of Gaussian difference scale space, subtract
Lack operand, improve arithmetic speed, and then improve the speed of graphical analysis;The image characteristic point locating module pair arranged
Extreme point carries out low contrast point and the removal of mobile rim point, it is ensured that the effectiveness of characteristic point, the wherein gray value to image
Strengthen, it is possible to be greatly increased the stability of image, the most accurate low contrast point is removed, and then improve image
The accuracy analyzed;Principal direction is set and determines module, it is proposed that the judgement formula of optimum line segment, with characteristic point gradient direction Nogata
The direction of the optimum line segment in the line segment that two peak value lines of the arbitrary neighborhood in figure are formed is as the principal direction of characteristic point, line segment phase
More stable for point so that the descriptor of image characteristic of correspondence point has repeatability, improves the accurate of feature descriptor
Property, and then can more fast and accurately image be identified detection, there is the highest robustness;The present embodiment takes threshold value
T1=0.03, T2=12, T3=0.06, the precision generating content improves 2.5%, and speed improves 3%.
Embodiment 4
Seeing Fig. 1, the present embodiment is used for generating the automatic image capture device of content, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring
When the value of 18 points that yardstick is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 of yardstick
When the value of 18 points that consecutive points are corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described difference of Gaussian chi
The reduced mechanical model in degree space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, positions for extreme point pinpoint first including be sequentially connected with
Submodule, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point,
Wherein:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalized successively to the image exported soon by image conversion submodule
Rear rejecting described low contrast point, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels higher than 128 of the gray value in image
Average, mLIt is the gray value average that is less than all pixels of 128, ψ (x, y) is the image after being processed by image filtering submodule,
T1For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed;
Preferably, the described automatic image capture device for generating content, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
Further, the sub-line section described in close slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3's
Span be (0,0.1].
It is strong with color that the image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color
The non-linear relation of degree, it is possible to describe image the most accurately;Propose the reduced mechanical model of Gaussian difference scale space, subtract
Lack operand, improve arithmetic speed, and then improve the speed of graphical analysis;The image characteristic point locating module pair arranged
Extreme point carries out low contrast point and the removal of mobile rim point, it is ensured that the effectiveness of characteristic point, the wherein gray value to image
Strengthen, it is possible to be greatly increased the stability of image, the most accurate low contrast point is removed, and then improve image
The accuracy analyzed;Principal direction is set and determines module, it is proposed that the judgement formula of optimum line segment, with characteristic point gradient direction Nogata
The direction of the optimum line segment in the line segment that two peak value lines of the arbitrary neighborhood in figure are formed is as the principal direction of characteristic point, line segment phase
More stable for point so that the descriptor of image characteristic of correspondence point has repeatability, improves the accurate of feature descriptor
Property, and then can more fast and accurately image be identified detection, there is the highest robustness;The present embodiment takes threshold value
T1=0.04, T2=13, T3=0.04, the precision generating content improves 1.5%, and speed improves 2%.
Embodiment 5
Seeing Fig. 1, the present embodiment is used for generating the automatic image capture device of content, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring
When the value of 18 points that yardstick is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 of yardstick
When the value of 18 points that consecutive points are corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described difference of Gaussian chi
The reduced mechanical model in degree space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, positions for extreme point pinpoint first including be sequentially connected with
Submodule, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point,
Wherein:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalized successively to the image exported soon by image conversion submodule
Rear rejecting described low contrast point, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels higher than 128 of the gray value in image
Average, mLIt is the gray value average that is less than all pixels of 128, ψ (x, y) is the image after being processed by image filtering submodule,
T1For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed;
Preferably, the described automatic image capture device for generating content, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
Further, the sub-line section described in close slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3's
Span be (0,0.1].
It is strong with color that the image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color
The non-linear relation of degree, it is possible to describe image the most accurately;Propose the reduced mechanical model of Gaussian difference scale space, subtract
Lack operand, improve arithmetic speed, and then improve the speed of graphical analysis;The image characteristic point locating module pair arranged
Extreme point carries out low contrast point and the removal of mobile rim point, it is ensured that the effectiveness of characteristic point, the wherein gray value to image
Strengthen, it is possible to be greatly increased the stability of image, the most accurate low contrast point is removed, and then improve image
The accuracy analyzed;Principal direction is set and determines module, it is proposed that the judgement formula of optimum line segment, with characteristic point gradient direction Nogata
The direction of the optimum line segment in the line segment that two peak value lines of the arbitrary neighborhood in figure are formed is as the principal direction of characteristic point, line segment phase
More stable for point so that the descriptor of image characteristic of correspondence point has repeatability, improves the accurate of feature descriptor
Property, and then can more fast and accurately image be identified detection, there is the highest robustness;The present embodiment takes threshold value
T1=0.05, T2=14, T3=0.02, the precision generating content improves 1.8%, and speed improves 1.5%.
Last it should be noted that, above example is only in order to illustrate technical scheme, rather than to scope
Restriction, although having made to explain to the present invention with reference to preferred embodiment, it will be understood by those within the art that,
Technical scheme can be modified or equivalent, without deviating from the spirit and scope of technical solution of the present invention.
Claims (3)
1., for generating an automatic image capture device for content, it is characterized in that, including:
(1) image pre-processing module, it includes the image transform subblock for coloured image is converted into gray level image and is used for
The image filtering submodule that described gray level image is filtered, the gradation of image conversion formula of described image transform subblock is:
Wherein, (x, y), (x, y), (x, (x, y) the intensity red green blue value at place, (x y) represents I B G R y) to represent pixel respectively
Pixel (x, y) gray value at place;
(2) image extreme point detection module, it by being carried out the Gauss of the image that convolution is created as by difference of Gaussian and image
Difference scale space detects the position of each extreme point, when sampled point relative to it with 8 consecutive points of yardstick and neighbouring chi
When the value of 18 points that degree is corresponding is the biggest, described sampled point is maximum point, when sampled point relative to it with 8 phases of yardstick
When the value of 18 points that adjoint point is corresponding with neighbouring yardstick is the least, described sampled point is minimum point, described Gaussian difference scale
The reduced mechanical model in space is:
D (x, y, σ)=(G (x, k σ)-G (x, σ)) * I'(x, y)+(G (y, k σ)-G (y, σ)) * I'(x, y)
Herein
Wherein, D (x, y, σ) represents Gaussian difference scale space function, I'(x, is y) by the image letter of image transformant module output
Number, * represents that convolution algorithm, σ represent the Gaussian function that the metric space factor, G (x, σ), G (y, σ) they are the changeable scale defined,
K is constant multiplication factor;
(3) image characteristic point locating module, it is by rejecting in described each extreme point the low contrast point of noise-sensitive and not
Stable marginal point determines the extreme point as characteristic point, including be sequentially connected with for pinpoint first locator of extreme point
Module, for removing the second locator module of low contrast point and for removing the 3rd locator module of mobile rim point, its
In:
A, described first locator module are by carrying out secondary Taylor expansion to described Gaussian difference scale space function and derivation obtains
The exact position of extreme point, the metric space function of extreme point is:
Wherein,Represent the metric space function of extreme point, D (x, y, σ)TFor the side-play amount relative to extreme point,Represent
The exact position of extreme point;
B, described second locator module carry out grey level enhancement, normalization successively to the image exported soon by image conversion submodule
Rejecting described low contrast point after reason, enhanced gray value is:
Herein
Described low contrast point judge formula as:
Wherein, I " (x, y) represents the enhanced image function of gray value,For comprising the correction coefficient of local message, M is
The maximum gradation value of pixel, described maximum gradation value M=255, mHFor all pixels equal higher than 128 of the gray value in image
Value, mLBeing the gray value average that is less than all pixels of 128, (x y) is the image after being processed by image filtering submodule, T to ψ1
For the threshold value set;
C, described 3rd locator module obtain this extreme value by the Hessian matrix H that Location Scale is 2 × 2 calculating extreme point
The principal curvatures of point, and by rejecting principal curvatures ratio more than threshold value T set2Extreme point reject described mobile rim point,
Wherein threshold value T2Span be [10,15], described principal curvatures ratio is come really by the ratio between the eigenvalue of comparator matrix H
Fixed.
A kind of automatic image capture device for generating content the most according to claim 1, is characterized in that, also include:
(1) principal direction determines module, including the connection sub module being sequentially connected with, merges submodule and processes submodule, described company
Line is used for two peak value lines of the arbitrary neighborhood in the gradient orientation histogram about described characteristic point in module to form many height
Line segment, described merging submodule is for merging formation one in the longitudinal direction by having close slope and adjacent sub-line section
Line segment, described process submodule for using the direction of the optimum line segment in a plurality of line segment as the principal direction of characteristic point, described optimum
Line segment judge formula as:
Wherein, LYRepresent optimum line segment,For average gradient value it isLine segment,For nth bar in described a plurality of line segment
The average gradient value of line segment, gkFor the kth strip line segment in described nth bar line segment, LυFor described a plurality of line segment middle conductor length
Line segment aggregate more than average line segment length;
(2) characteristic extracting module, it is according to described main formula always hyperspin feature neighborhood of a point, and according to postrotational neighborhood to institute
State characteristic point to be described, thus generate the descriptor of described characteristic point;
(3) content generating module, the feature of extraction, through processing, completes content and generates.
A kind of automatic image capture device for generating content the most according to claim 1, is characterized in that, described in have close
The sub-line section of slope is that slope differences is less than predetermined threshold value T3Sub-line section, described threshold value T3Span be (0,0.1].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610231049.7A CN105930853A (en) | 2016-04-14 | 2016-04-14 | Automatic image capturing device for content generation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610231049.7A CN105930853A (en) | 2016-04-14 | 2016-04-14 | Automatic image capturing device for content generation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105930853A true CN105930853A (en) | 2016-09-07 |
Family
ID=56838180
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610231049.7A Pending CN105930853A (en) | 2016-04-14 | 2016-04-14 | Automatic image capturing device for content generation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105930853A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190637A (en) * | 2018-07-31 | 2019-01-11 | 北京交通大学 | A kind of image characteristic extracting method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101470896A (en) * | 2007-12-24 | 2009-07-01 | 南京理工大学 | Automotive target flight mode prediction technique based on video analysis |
CN103020945A (en) * | 2011-09-21 | 2013-04-03 | 中国科学院电子学研究所 | Remote sensing image registration method of multi-source sensor |
CN104978709A (en) * | 2015-06-24 | 2015-10-14 | 北京邮电大学 | Descriptor generation method and apparatus |
CN105046681A (en) * | 2015-05-14 | 2015-11-11 | 江南大学 | Image salient region detecting method based on SoC |
-
2016
- 2016-04-14 CN CN201610231049.7A patent/CN105930853A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101470896A (en) * | 2007-12-24 | 2009-07-01 | 南京理工大学 | Automotive target flight mode prediction technique based on video analysis |
CN103020945A (en) * | 2011-09-21 | 2013-04-03 | 中国科学院电子学研究所 | Remote sensing image registration method of multi-source sensor |
CN105046681A (en) * | 2015-05-14 | 2015-11-11 | 江南大学 | Image salient region detecting method based on SoC |
CN104978709A (en) * | 2015-06-24 | 2015-10-14 | 北京邮电大学 | Descriptor generation method and apparatus |
Non-Patent Citations (2)
Title |
---|
吴京辉: "视频监控目标的跟踪与识别研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
张建兴: "基于注意力的目标识别算法及在移动机器人的应用研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190637A (en) * | 2018-07-31 | 2019-01-11 | 北京交通大学 | A kind of image characteristic extracting method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11580647B1 (en) | Global and local binary pattern image crack segmentation method based on robot vision | |
CN105844337A (en) | Intelligent garbage classification device | |
US9953229B2 (en) | Traffic light detection | |
CN104751142B (en) | A kind of natural scene Method for text detection based on stroke feature | |
CN107563377A (en) | It is a kind of to detect localization method using the certificate key area of edge and character area | |
CN108182383B (en) | Vehicle window detection method and device | |
CN105913093A (en) | Template matching method for character recognizing and processing | |
CN106446952A (en) | Method and apparatus for recognizing score image | |
CN112052782B (en) | Method, device, equipment and storage medium for recognizing parking space based on looking around | |
CN106709518A (en) | Android platform-based blind way recognition system | |
CN108830133A (en) | Recognition methods, electronic device and the readable storage medium storing program for executing of contract image picture | |
CN113034452A (en) | Weldment contour detection method | |
US8249387B2 (en) | Image processing method and apparatus for detecting lines of images and start and end points of lines | |
CN105928099A (en) | Intelligent air purifier | |
CN112991374A (en) | Canny algorithm-based edge enhancement method, device, equipment and storage medium | |
CN104281850B (en) | character area identification method and device | |
CN105844260A (en) | Multifunctional smart cleaning robot apparatus | |
CN116152261A (en) | Visual inspection system for quality of printed product | |
CN105844651A (en) | Image analyzing apparatus | |
CN104715250A (en) | Cross laser detection method and device | |
CN105933698A (en) | Intelligent satellite digital TV program play quality detection system | |
CN109146863A (en) | A kind of pavement marker line defect detection device | |
CN105930853A (en) | Automatic image capturing device for content generation | |
CN104102911A (en) | Image processing for AOI (automated optical inspection)-based bullet appearance defect detection system | |
CN105930779A (en) | Image scene mode generation device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160907 |
|
RJ01 | Rejection of invention patent application after publication |