CN106504237A - Determine method and the image acquiring method of matching double points - Google Patents
Determine method and the image acquiring method of matching double points Download PDFInfo
- Publication number
- CN106504237A CN106504237A CN201610916842.0A CN201610916842A CN106504237A CN 106504237 A CN106504237 A CN 106504237A CN 201610916842 A CN201610916842 A CN 201610916842A CN 106504237 A CN106504237 A CN 106504237A
- Authority
- CN
- China
- Prior art keywords
- image
- point
- double points
- pixel
- matching
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4038—Scaling the whole image or part thereof for image mosaicing, i.e. plane images composed of plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30008—Bone
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
- G06V2201/033—Recognition of patterns in medical or anatomical images of skeletal patterns
Abstract
The method for determining matching double points.Methods described includes:Pyramid decomposition, the first difference pyramid of construction and the second difference pyramid is carried out to the first image and the second image;The each layer of the first difference pyramid is traveled through, the extreme point that searches in every layer of difference pyramid, each layer of the second difference pyramid of traversal, the extreme point that searches in every layer of difference pyramid;Remove in the pyramidal extreme point of the first difference the characteristic point in the first image is obtained for the pixel in the first image strong edge;Remove in the pyramidal extreme point of the second difference the characteristic point in the second image is obtained for the pixel in the second image strong edge;The characteristic vector of characteristic point in characteristic vector and the second image based on characteristic point in the first image, carries out bi-directional matching and generates initial matching point pair to characteristic point in the first image and the second image;The matching double points for removing initial matching point centering mistake obtain matching double points.This programme complexity is low, the simple and quick matching double points for determining that accuracy is high of amount of calculation I.
Description
Technical field
The present invention relates to technical field of image processing, the detection method of characteristic point and device in more particularly to a kind of image,
Determine method and device, image acquiring method and the device and X-ray shooting system of matching double points.
Background technology
Digital X-ray photography (DR, Digital Radiography) equipment is computer digital image treatment technology and X
A kind of advanced armarium that ray irradiation technology combines and formed.Digital X-ray photographic equipment because its radiation dose is little,
The quality of image is high, the accuracy of the recall rate of disease and diagnosis is higher and be widely used.
In orthopaedic diseases such as diagnosis vertebra prolapse, lateral bending, lower limb malformations, or carry out the reduction of the fracture, Joint shift, osteotomy
When art and pedicle screw are inserted, the auxiliary of medical imaging device is needed, and X-ray equipment is due to its low cost, dosage
Little and the advantages of noinvasive is visualized can be realized, become the preferred unit of above-mentioned diagnosis at present.But as X-ray sets
Standby restriction in terms of panel size, its areas imaging are difficult to cover complete vertebra or limbic areas.Now, generally by right
Image sequence is accurately spliced using image processing algorithm, is provided for doctor by vertebra or long bone of limbs collection image sequence
Accurate image information, to complete the diagnosis to above-mentioned orthopaedic disease.
When gathering image by X-ray equipment, the adjacent human dissection position for shooting twice is different, therefore adopt
Dosage is also different, and therefore between adjacent image, the gray difference of overlapping region is also larger.And during shooting image, for keeping away
Exempt from patient by excessive radiation dose, it will usually unnecessary primary X ray is screened off to by beam-defining clipper, which can be by roentgenogram
Launched field is limited in required minimum zone, but the region that beam-defining clipper is covered in image can be interfered to accurate splicing.In addition,
In order to protect person under inspection, therefore the reduction radiation dose that can also try one's best in shooting process shoots noise in the radioscopic image of acquisition
Disturb larger, poor in image definition.And person under inspection is also difficult to hold one's breath always in multiple image process are obtained, have unavoidably
Slight breathing is slight mobile, above-mentioned refer to the precision of stitching algorithm will have been affected.Additionally, being to avoid requirement from receiving
Inspection person's long-time is remained stationary as and is held one's breath, and reduces the waiting time of doctor, improves diagnosis efficiency, also requires that image mosaic is calculated
Method has execution speed quickly.
Accordingly, it is capable to the image of quick acquisition splicing high precision is the technological difficulties in image mosaic field.Existing to X ray
The splicing of image is generally using feature based, based on gray scale, the joining method based on transform domain.For the image of feature based is spelled
For connecing method, which carries out pretreatment first to image subject to registration, extracts specific feature set;Then according to similarity measurement letter
Number, mates to the description of feature set, and then realizes the location matches of image.The image split-joint method of feature based is to image
Grey scale change there is stronger robustness, and when carrying out image mosaic using the image split-joint method of feature based, be extracted
The marked feature of image to be spliced, have compressed the quantity of information of image to a great extent, thus using the joining method amount of calculation compared with
Little, execute speed.For the image split-joint method based on gray scale, which mainly uses the half-tone information of image, choosing
A cost function that can suitably measure similarity degree between image is selected, then certain search strategy is adopted, taking makes the phase
Like the parameter value that property cost function value obtains the correspondent transform model being most worth.But the image split-joint method based on gray scale is searched for most
Than larger, algorithm execution time is long for the process amount of calculation of excellent parameter, and the sensitivity to picture noise is strong, to gradation of image dependency
Stronger, and the size of the gray difference between two image overlapping regions to be spliced will affect the degree of accuracy of splicing.And base
It is then that two width images are transformed from a spatial domain to by frequency domain by fast Fourier transform in the image split-joint method of transform domain, leads to
Cross the crosspower spectrum of two width images phase place directly calculate two width images translation, rotation and scaling corresponding parameter, with reality
The splicing of existing image.But the image split-joint method based on transform domain is affected by image border significant difference and noise etc., having can
Cannot can correctly splice parameter, therefore the image split-joint method more application for being typically based on transform domain is registering when splicing initial
The determination of parameter.Therefore, the more at present joining method for adopting feature based is realizing the splicing to radioscopic image.But adopt
Characteristic point when being spliced to radioscopic image with the image split-joint method of existing feature based, in stitching image is treated
When being detected, the complexity of detection algorithm is high and detection speed is slow.In addition using the image mosaic of existing distinguished point based
When method realizes image mosaic, the matching double points of mistake are still suffered from matching double points, and the accuracy of matching double points is low, and then causes
When carrying out image mosaic with the image split-joint method of feature based, the splicing precision for splicing the image for obtaining is not high, and clinic is examined
Break and have a certain impact.
Therefore, how with the characteristic point in the quick detection image of relatively low complexity, simple and accurately determine
Matching double points, obtain the image of splicing high precision, become one of current problem demanding prompt solution.
Content of the invention
The problem to be solved in the present invention is to provide a kind of method for determining matching double points, with relatively low complexity and less
The simple and quick accurate matching double points for determining that accuracy is high of amount of calculation, and a kind of image acquisition side of splicing high precision is provided
Method.
For solving the above problems, technical solution of the present invention provides a kind of method for determining matching double points, including:
Pyramid decomposition is carried out to the first image and the second image, based on decomposition after the first image and the second image difference
Construct the first difference pyramid diagram picture corresponding with described first image and the second difference corresponding with second image gold
Word tower image;Described first image and second image are adjacent image;
Each layer of the first difference pyramid diagram picture is traveled through, the extreme point that searches in every layer of difference pyramid diagram picture,
Each layer of the second difference pyramid diagram picture is traveled through, the extreme point that searches in every layer of difference pyramid diagram picture;The extreme value
Point is associated with the absolute value of the gray value of the pixel in default neighborhood;
Remove in the extreme point of the first difference pyramid diagram picture for the pixel in described first image strong edge with
Obtain the characteristic point in described first image;It is second image in the extreme point for removing the second difference pyramid diagram picture
Pixel in strong edge is obtaining the characteristic point in second image;
The feature of the characteristic point in the characteristic vector of the characteristic point in based on described first image and second image to
Amount, carries out bi-directional matching to generate initial matching point pair to the characteristic point in described first image and second image;
Remove the matching double points of the initial matching point centering mistake to obtain matching double points.
Optionally, the extreme point of the first difference pyramid diagram picture refers to every layer of difference pyramid diagram pixel as in
The absolute value of gray value is the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of the pixel;Described
The extreme point of two difference pyramid diagram pictures refers to that the absolute value of the gray value of pixel in every layer of difference pyramid diagram picture is described
The pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of pixel.
Optionally, the extreme point of the first difference pyramid diagram picture is referred in the first difference pyramid diagram picture and is revised
Pixel P afterwards, revised pixel P are obtained after the position to pixel P is modified, and the pixel P is referred to
In every tomographic image of the first difference pyramid diagram picture, the absolute value of the gray value of pixel is the default neighborhood of the pixel
The pixel of the maximum absolute value of the gray value of middle pixel;The extreme point of the second difference pyramid diagram picture refers to described
Revised pixel P in two difference pyramid diagram pictures, the revised pixel P is that the position of pixel P is repaiied
Just obtaining afterwards, pixel P refers to the absolute of the gray value of pixel in every tomographic image of the second difference pyramid diagram picture
It is worth the pixel for the maximum absolute value of the gray value of pixel in the default neighborhood of the pixel.
Optionally, the extreme point of the first difference pyramid diagram picture refer to the absolute value of the gray value to pixel P by
According to the pixel P for being located at top N after order sequence from large to small, the pixel P refers to the first difference pyramid diagram
In every tomographic image of picture, the absolute value of the gray value of pixel is the exhausted of the gray value of pixel in the default neighborhood of the pixel
To the pixel that value is maximum;The extreme point of the second difference pyramid diagram picture refers to the absolute value of the gray value to pixel P
According to the pixel P for being located at top N after order sequence from large to small, the pixel P refers to the second difference pyramid
In every tomographic image of image, the absolute value of the gray value of pixel is the gray value of pixel in the default neighborhood of the pixel
The pixel of maximum absolute value.
Optionally, described based on described first image in characteristic point characteristic vector and second image in feature
The characteristic vector of point, carries out bi-directional matching to the characteristic point in described first image and second image to generate initial matching
Point is to including:
Spy with the characteristic point in the characteristic vector of each characteristic point in described first image and second image
It is the first ratio with the ratio of secondary little Euclidean distance to levy the minimum euclidean distance between vector;
First ratio be less than first threshold when, with described first image corresponding with the minimum euclidean distance in
Characteristic point and second image in characteristic point be the 3rd matching double points, generate the with the 3rd matching double points as element
One set;
Spy with the characteristic point in the characteristic vector of each characteristic point in second image and described first image
It is the second ratio with the ratio of secondary little Euclidean distance to levy the minimum euclidean distance between vector;
Second ratio be less than Second Threshold when, with second image corresponding with the minimum euclidean distance in
Characteristic point and described first image in characteristic point be the 4th matching double points, generate the with the 4th matching double points as element
Two set;
Take the common factor of the first set and the second set to obtain described first image and second image
Initial matching point pair.
Optionally, the characteristic point in the characteristic vector of each characteristic point in described first image and second image
Characteristic vector between minimum euclidean distance refer to the ratio of secondary little Euclidean distance:Each feature in described first image
The characteristic vector of point is preset less than first apart from its poor absolute value with the abscissa with this feature point in second image
Minimum euclidean distance between the characteristic vector of distance and the characteristic point for belonging to same difference pyramid image layer with this feature point
Ratio with secondary little Euclidean distance;
The feature of the characteristic point in the characteristic vector of each characteristic point in second image and described first image
Minimum euclidean distance between vector is referred to the ratio of secondary little Euclidean distance:The spy of each characteristic point in second image
Levy vector with described first image with the abscissa of this feature point apart from its poor absolute value less than the second predeterminable range and
Minimum euclidean distance between the characteristic vector of the characteristic point for belonging to same difference pyramid image layer with this feature point and time little
The ratio of Euclidean distance.
Optionally, the characteristic point in the characteristic vector of each characteristic point in described first image and second image
Characteristic vector between minimum euclidean distance refer to the ratio of secondary little Euclidean distance:Each feature in described first image
The characteristic vector of point is preset less than the 3rd apart from its poor absolute value with the vertical coordinate with this feature point in second image
Minimum euclidean distance between the characteristic vector of distance and the characteristic point for belonging to same difference pyramid image layer with this feature point
Ratio with secondary little Euclidean distance;
The feature of the characteristic point in the characteristic vector of each characteristic point in second image and described first image
Minimum euclidean distance between vector is referred to the ratio of secondary little Euclidean distance:The spy of each characteristic point in second image
Levy vector with described first image with the vertical coordinate of this feature point apart from its poor absolute value less than the 4th predeterminable range and
Minimum euclidean distance between the characteristic vector of the characteristic point for belonging to same difference pyramid image layer with this feature point and time little
The ratio of Euclidean distance.
Optionally, the matching double points for removing the initial matching point centering mistake are included with obtaining matching double points:
Slope of the initial matching point of described first image and second image to place line is calculated, with described initial
The slope of matching double points place line is abscissa, and initial matching point corresponding with the slope is that vertical coordinate is generated tiltedly to number
Rate rectangular histogram;
Determine initial matching point in the slope histogram to the initial matching point of number sum place cluster when maximum to for
First matching double points;
As abscissa, corresponding with the difference of the first coordinate first mates difference with the first coordinate of first matching double points
Point is that vertical coordinate generates the first rectangular histogram to number;
When determining that the first matching double points number sum is maximum in first rectangular histogram, the first matching double points of place cluster are
Second matching double points;
As abscissa, corresponding with the difference of the second coordinate second mates difference with the second coordinate of second matching double points
Point is that vertical coordinate generates the second rectangular histogram to number;
When determining that the second matching double points number sum is maximum in second rectangular histogram, the second matching double points of place cluster are
Matching double points.
Optionally, the initial matching point for calculating described first image and second image in the following way connects to being located
The slope of line:
Wherein:KiFor slope, (x of the i-th pair initial matching point to place line1i,y1i)、(x2i,y2i) it is that i-th pair is initial
Matching double points, (x1i,y1i) for i-th initial matching point in described first image position, (x2i,y2i) it is second figure
The position of i-th initial matching point as in, W1Width for described first image.
For solving the above problems, technical solution of the present invention also provides a kind of image acquiring method, including:
The matching double points that first image and the second image are determined using the method for above-mentioned determination matching double points;
Determined based on the position relationship between the matching double points inclined between described first image and second image
Move;
Described first image and second figure are determined according to the skew between described first image and second image
The overlapping region of picture;
Described first image and second image are spliced according to the overlapping region.
Compared with prior art, technical solution of the present invention has advantages below:
Pyramid decomposition is carried out to described image first, based on decomposition after image configuration difference pyramid diagram picture;Time then
Each layer of the difference pyramid diagram picture is gone through, the extreme point that searches in every layer of difference pyramid diagram picture, the extreme point association
The absolute value of the gray value of the pixel in default neighborhood;Finally remove in the extreme point as in described image strong edge
Pixel is obtaining the characteristic point in described image.Per layer is searched for due to having traveled through each layer of the difference pyramid diagram picture
Extreme point in difference pyramid diagram picture, determines the side of extreme point relative to existing with difference pyramid diagram picture place space
For method, the complexity of detection is low, and then improves the speed of detection characteristic point to a certain extent, due to eliminating extreme point
In for the pixel in described image strong edge, therefore stable characteristic point can be obtained.
Description of the drawings
Fig. 1 be the embodiment of the present invention one image in characteristic point detection method schematic flow sheet;
Fig. 2 is the schematic flow sheet of the method for the determination matching double points of the embodiment of the present invention two;
Fig. 3 is the sub- schematic diagram of feature point description of the embodiment of the present invention two;
Fig. 4 is the schematic flow sheet of the method for the determination matching double points of the embodiment of the present invention three;
Fig. 5 be the embodiment of the present invention three the first image of acquisition and the second image in initial matching point to the oblique of place line
The schematic diagram of rate;
Fig. 6 is the schematic flow sheet of the image acquiring method of the embodiment of the present invention four;
Fig. 7-a to Fig. 7-d are the position relationship schematic diagrams between the first image and the second image.
Specific embodiment
Understandable for enabling the above objects, features and advantages of the present invention to become apparent from, below in conjunction with the accompanying drawings to the present invention
Specific embodiment be described in detail.Elaborate detail in order to fully understanding the present invention in the following description.But
It is that the present invention can be implemented different from alternate manner described here with multiple, those skilled in the art can be without prejudice to this
Similar popularization is done in the case of invention intension.Therefore the present invention is not limited by following public specific embodiment.
As described in prior art, when the characteristic point in stitching image is treated is detected, existing detection
Algorithm complex is high, and detection speed is slow, additionally, when the characteristic point to detecting is mated, the accuracy of matching double points
Not high, therefore to be spliced based on the matching double points, the splicing precision of the image obtained after splicing is low, and clinical diagnosises are brought
Affect.
Therefore, inventor propose existing feature point detecting method is improved, and based on improvement after characteristic point inspection
Survey method further improves distinguished point based come the method for determining matching double points, improves the image split-joint method of feature based,
To realize characteristic point in image is quickly detected with relatively low complexity, simple and quick accurately determine adjacent two width image
Between matching double points, and then realize obtaining the splicing higher image of precision with speed faster.
Technical scheme is described in detail below by way of specific embodiment.
Embodiment one
Refer to Fig. 1, Fig. 1 be the embodiment of the present invention one image in characteristic point detection method schematic flow sheet;Such as
Shown in Fig. 1, in the image of the present embodiment, the detection method of characteristic point includes:
S101:Pyramid decomposition is carried out to described image, based on decomposition after image configuration difference pyramid diagram picture;
S102:Each layer of the difference pyramid diagram picture is traveled through, the extreme point that searches in every layer of difference pyramid diagram picture,
The extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
S103:Remove is the pixel in described image strong edge in the extreme point obtaining the feature in described image
Point.
S101 is executed, pyramid decomposition is carried out to described image, be that Gauss gold word is carried out to described image in the present embodiment
Tower decomposes, and sets up the gaussian pyramid of described image in other words.And based on decomposition after image configuration difference of Gaussian pyramid diagram
Picture.In the present embodiment, carrying out gaussian pyramid decomposition to described image can be carried out with described image by Gaussian function first
Convolution, obtains the Gaussian smoothing image of different scale, then the Gaussian smoothing image of different scale is carried out down-sampled with composition
Gaussian pyramid image.Specifically, in the present embodiment, gaussian pyramid image is obtained by equation below:
L (x, y, σ)=G (x, y, σ) * I (x, y)
Wherein, I (x, y) is input picture, and G (x, y, σ) is Gaussian function, and L (x, y, σ) is gaussian pyramid image, and σ is
The yardstick of gaussian kernel, the coordinate of (x, y) for pixel.
For for Gaussian function G (x, y, σ),
(xi,yi) for gaussian kernel center.In the present embodiment, with the gradually increase of σ, the Gaussian smoothing figure of different scale can be generated
Picture, during σ gradually increases, carries out down-sampled Gaussian smoothing figure after down-sampled to the Gaussian smoothing image
As constituting gaussian pyramid image.Can be controlled by the size of the value to σ in the present embodiment, to control the spy for detecting
Quantity a little is levied, and then reduces the time of the detection characteristic point.Next, the Gaussian difference of the construction gaussian pyramid image
Divide pyramid image, difference of Gaussian function D (x, y, σ)=[G (x, y, k σ)-G (x, y, σ)] * I (x, y)=G (x, y, k σ) * I
(x,y)-G(x,y,σ)*I(x,y).
From the foregoing, G (x, y, σ) * I (x, y) is the image that described image is carried out to obtain after gaussian pyramid decomposition,
That is L (x, y, σ).Therefore, the difference of Gaussian pyramid diagram picture of described image is the difference of the gaussian pyramid image of adjacent yardstick,
I.e.:D (x, y, σ)=L (x, y, k σ)-L (x, y, σ).
The difference of Gaussian pyramid diagram picture that described image is obtained by S101, next executes S102, determines the height
Extreme point in this difference pyramid diagram picture, specifically, exactly travels through each layer of the difference of Gaussian pyramid diagram picture, search
Extreme point present in each layer of difference of Gaussian pyramid diagram picture, in the present embodiment, the extreme point is associated with default neighborhood
In pixel gray value absolute value, in the present embodiment, the extreme point can be picture in every layer of difference pyramid diagram picture
The absolute value of the gray value of vegetarian refreshments is the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of the pixel.
The big I of the default neighborhood according to depending on actual demand (such as:The final intensity and characteristic point of characteristic point to be determined
The size in shared region in described image), in the present embodiment, the default neighborhood can be eight centered on the pixel
Neighborhood, namely the pixel array of 3 × 3 centered on the pixel, can also be 24 centered on the pixel
Neighborhood, namely the pixel array of 5 × 5 centered on the pixel.For example, it is exactly the Gaussian difference parting to each layer
For word tower image, by the absolute value of the gray value of each pixel in this layer and the pixel of predetermined neighborhood around which
The absolute value of gray value is compared, when the pixel gray value absolute value than which around all pixels point gray value
Absolute value all big when, then the pixel is the extreme point in this layer of difference of Gaussian pyramid diagram picture.
In the present embodiment, the accuracy in order to improve the extreme point for searching further is improved in the image for eventually detecting
The accuracy of characteristic point, the extreme point can also be revised pixel P, and the revised pixel P is to pixel
The position of point P is obtained after being modified, and the pixel P refers to the gray value of pixel in every layer of difference pyramid diagram picture
Absolute value is the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of the pixel.Pixel P herein
Acquisition be also by will be around the absolute value of the gray value of each pixel in every layer of difference pyramid diagram picture and its in advance
The absolute value for determining the gray value of the pixel of neighborhood is compared, when the pixel gray value absolute value than which around institute
When having the absolute value of the gray value of pixel all big, the pixel is pixel P.After the pixel P is obtained, to its position
It is modified to obtain revised pixel P, in the following way the position of the pixel P is repaiied in the present embodiment
Just:
First, side-play amount is first obtained, and the offset delta is obtained in the following way:
Wherein:D be difference of Gaussian function, X=(Δ x, Δ y, Δ σ)T, Δ x, Δ y, Δ σ are respectively the x of extreme point and sit
The x of mark, y-coordinate and the σ and pixel PpCoordinate, ypCoordinate and σpDifference.
After above-mentioned formula solves δ, by the x of the pixel PpCoordinate, ypCoordinate and σpPlus right with which in δ
The vector that answers, then can obtain the position of revised pixel P, namely the position of extreme point.
In the present embodiment, the time of algorithm performs during in order to control to detect characteristic point, characteristic point in detection image is improved
Speed, the extreme point can also be that the absolute value of the gray value to pixel P is located at after sorting according to order from large to small
The pixel P of top N, the pixel P refer to that the absolute value of the gray value of pixel in every layer of difference pyramid diagram picture is institute
State the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of pixel.Similarly, pixel P herein is obtained
Must be also to pass through to make a reservation for neighbour around the absolute value of the gray value of each pixel in every layer of difference pyramid diagram picture and its
The absolute value of the gray value of the pixel in domain is compared, when the pixel gray value absolute value than which around all pictures
When the absolute value of the gray value of vegetarian refreshments is all big, the pixel is pixel P.In the present embodiment, the value of N is empirical value, passes through
The time of algorithm performs during the accuracy rate of the characteristic point detected in multiple image and detection characteristic point is traded off with
Final acquisition N values, in the present embodiment, the span of N can be 30~50, and such as N can take 40, the ash of the pixel P to obtaining
The absolute value of angle value is ranked up according to descending order, and the absolute value of the gray value of capture vegetarian refreshments P is located at first 40
Pixel P removes the weaker pixel of other features in pixel P as extreme point.
Additionally, in the present embodiment when extreme point is determined, or first to the absolute value of the gray value of pixel P according to
Order from large to small is ranked up, and the absolute value for first taking out the gray value of pixel P is located at the pixel P of top N, then right
The position for being located at the pixel P of top N is modified, and the absolute value of the gray value of pixel P is located at top N and to this finally
The position of N number of pixel P be modified after pixel P as extreme point, extreme point is determined using which, final is obtained
The accuracy of extreme point is high, and can improve the speed of characteristic point in subsequent detection image.
The extreme point obtained by S102, if as characteristic point, can wherein include unstable characteristic point, therefore need
Unstable characteristic point is gone the characteristic point stable divided by acquisition.Therefore, it is the figure to execute S013 and remove in the extreme point
As the pixel in strong edge is obtaining stable characteristic point in described image.Due to the pixel in image strong edge across
There is larger principal curvatures in the direction at edge, and the direction principal curvatures along vertical edge is then less, therefore can by principal curvatures come
Judge that those pixels in extreme point are the pixels in described image strong edge.Further, since the size of principal curvatures and
The eigenvalue of Hessian matrixes is proportional, so in the present embodiment, can pass through the eigenvalue for calculating Hessian matrixes
To determine principal curvatures, and then determine the pixel being located in described image strong edge.
In the present embodiment, specifically, by calculating maximum principal curvatures and minimum master of the extreme point in its predetermined field
Curvature, and determine that according to the ratio of the two whether the extreme point is the pixel in described image strong edge.From the foregoing,
The eigenvalue of principal curvatures and Hessian matrixes is directly proportional, and therefore the ratio of maximum principal curvatures and minimum principal curvatures can pass through meter
The ratio of eigenvalue of maximum and minimal eigenvalue of Hessian matrixes is calculated obtaining.In the present embodiment, Hessian matrixes are as follows
Shown:
Wherein:D is difference of Gaussian function
Dxx=D (y, x+1, σ)+D (y, x-1, σ) -2 × D (x, y, σ)
Dyy=D (y+1, x, σ)+D (y-1, x, σ) -2 × D (x, y, σ)
Dxy=0.25 × [D (y+1, x+1, σ)+D (y-1, x-1, σ)-D (y+1, x-1, σ)-D (y-1, x+1, σ)]
(x, y) is the coordinate of the extreme point, calculates the eigenvalue of maximum and minimal eigenvalue of above-mentioned Hessian matrixes
Between ratio, then the ratio is compared with default threshold value, the default threshold value rule of thumb depending on, its close
The order of accuarcy for rejecting pixel in image strong edge is coupled to, and remaining pixel quantity executes the shadow of time to subsequent algorithm
Ring.Default threshold value described in the present embodiment may range from 1~30, such as:The default threshold value can be 20.By above-mentioned
Formula is could be aware that when the extreme point in the maximum principal curvatures of its 2 × 2 neighborhood and the ratio of minimum principal curvatures more than described
During default threshold value, the extreme point is the pixel in described image strong edge.By to the extreme point in its 2 × 2 neighborhood
Maximum principal curvatures and minimum principal curvatures ratio judgement, can be by the figure present in the extreme point searched in S102
As the pixel in strong edge is removed, the more stable characteristic point of final acquisition.
So far, the detection to characteristic point in image is achieved by above-mentioned S01~S103, due to having traveled through the difference
Each layer of pyramid diagram picture searching for the extreme point in every layer of difference pyramid diagram picture, relative to existing with difference pyramid
Image place space is come, for determining the method for extreme point, the complexity of detection is low, and then improves detection to a certain extent
The speed of characteristic point, due to eliminating in extreme point as the pixel in described image strong edge, therefore can obtain stable spy
Levy a little.
Furthermore, it is necessary to illustrate, in actual applications to image in characteristic point when detecting, should according to different
With scene, feature point detection can be carried out to entire image, it is also possible to feature point detection is carried out just for parts of images, for example:
If being to be applied to subsequently spell the first image and the second image after carrying out feature point detection to the first image and the second image
Connect, then when feature point detection is carried out to the first image and the second image, can be just for the first image and the second image substantially
The region that the region (initial overlapping region) of overlap is located in the first image carries out feature point detection, the first image and the second figure
The region that the initial overlapping region of picture is located in the second image carries out feature point detection.
Corresponding to the detection method of characteristic point in above-mentioned image, the present embodiment also provides a kind of inspection of characteristic point in image
Device is surveyed, the detection means of characteristic point includes in described image:
Resolving cell, for carrying out pyramid decomposition to described image, based on decomposition after image configuration difference pyramid
Image;
Search unit, for traveling through each layer of the difference pyramid diagram picture, in every layer of difference pyramid diagram picture of search
Extreme point, the extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
First removal unit, described to obtain for the pixel in described image strong edge in the extreme point for removing
Characteristic point in image.
The enforcement of the detection method for being embodied as referring to the characteristic point of the detection means of characteristic point in described image,
Will not be described here.
The present embodiment also provides a kind of X-ray shooting system, including the detection means of characteristic point in above-mentioned image.
Embodiment two
The present embodiment provides a kind of method for determining matching double points, refers to Fig. 2, Fig. 2 be the embodiment of the present invention two really
Determine the schematic flow sheet of the method for matching double points, as shown in Fig. 2 the method for determining matching double points includes:
S101′:Pyramid decomposition is carried out to the first image and the second image, based on decomposition after the first image and the second figure
As construct respectively corresponding with described first image the first difference pyramid diagram picture and with second image corresponding second
Difference pyramid diagram picture;Described first image and second image are adjacent image;
S102′:Each layer of the first difference pyramid diagram picture is traveled through, the pole that searches in every layer of difference pyramid diagram picture
Value point, travels through each layer of the second difference pyramid diagram picture, and the extreme point that searches in every layer of difference pyramid diagram picture is described
Extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
S103′:Remove in the extreme point of the first difference pyramid diagram picture as the picture in described first image strong edge
Vegetarian refreshments is obtaining the characteristic point in described first image;It is described in the extreme point for removing the second difference pyramid diagram picture
Pixel in two image strong edges is obtaining the characteristic point in second image;
S104:Characteristic point in described first image and second image is mated to generate initial matching point
Right;
S105:Difference with the first coordinate of the initial matching point pair is as abscissa, corresponding with the difference of the first coordinate first
Beginning matching double points number is that vertical coordinate generates the first rectangular histogram, determines that initial matching point is to number sum in first rectangular histogram
When maximum, the initial matching point of place cluster is to for the first matching double points;
S106:Difference with the second coordinate of first matching double points as abscissa, corresponding with the difference of the second coordinate
One matching double points number is that vertical coordinate generates the second rectangular histogram, determines the first matching double points number sum in second rectangular histogram
When maximum, the first matching double points of place cluster are matching double points.
In the present embodiment, S101 '~S103 ' detects the characteristic point in the second image of the characteristic point in the first image and detection
Similar with the S101~S103 in embodiment one, here is omitted, is detecting described first image and second figure
After characteristic point as in, S104 is executed, the characteristic point in the two is mated to generate initial matching point pair, specifically, this
In the following way the characteristic point in described first image and the second image is mated in embodiment:
First, the characteristic vector of described first image and the characteristic point in second image is obtained, for characteristic point
For characteristic vector, the information typically by the pixel near this feature point is retouching to the characteristic vector of this feature point
State, be the gradient of the pixel of predetermined neighborhood with the characteristic point in the first image, gradient magnitude, gradient direction in the present embodiment
To be described to the characteristic point in the first image, namely the characteristic vector of the characteristic point in described first image is predetermined by which
The gradient of the pixel of neighborhood, gradient magnitude and gradient vector are characterizing.
Hereinafter, by taking described first image as an example, the acquisition of the characteristic vector of the characteristic point in the first image is carried out accordingly
Explanation.With the characteristic point as origin in the present embodiment, formed centered on the characteristic point for radius by predetermined length
Predetermined neighborhood, in the present embodiment, the radius radius can be obtained by equation below:
Wherein:σoctThe yardstick of the gaussian kernel of the difference pyramid diagram picture being located for the characteristic point, d is variable, its value
Be associated with order of accuarcy and algorithm execution time that the characteristic vector of generation is described to feature vertex neighborhood, d may range from 2~
8, d=4 in the present embodiment.After the predetermined neighborhood centered on the characteristic point is determined, a certain in difference pyramid diagram picture
Gradient gradI (x, y) of pixel of the characteristic point of layer in the predetermined neighborhood, gradient magnitude m (x, y), gradient direction θ
(x, y) is obtained by equation below respectively:
Wherein, I (x, y) be input picture, gradI (x, y) be gradient, m (x, y) be gradient magnitude, θ (x, y) be gradient
Direction, the coordinate that (x, y) is pixel.
After calculating gradient, gradient magnitude and the gradient direction of the pixel of the predetermined neighborhood of each characteristic point, to pre-
The gradient direction and gradient magnitude for determining the pixel of neighborhood carries out statistics with histogram, to obtain the characteristic vector of this feature point.This
In embodiment, (each unit represents a pixel, and characteristic point is 8 × 8 for unit with the predetermined neighborhood of characteristic point as 8 × 8
Unit center), namely predetermined neighborhood is to illustrate as a example by 64 pixels to obtain with this feature corresponding characteristic vector of point
?.Refer to Fig. 3, Fig. 3 is the sub- schematic diagram of feature point description of the embodiment of the present invention two, and left side show 64 pixels in Fig. 3
Point (represented with 64 cells), the gradient direction and gradient magnitude of each pixel by Fig. 3 shown in left side with side
Represent to the small arrow with size, in order to carry out statistics with histogram to the gradient direction and gradient magnitude of 64 pixels, first by 8
× 8 cell is divided into the cell of 44 × 4, the gradient direction and gradient width to the pixel in 44 × 4 cells
Value carries out statistics with histogram, in the present embodiment for each 4 × 4 cell, draw respectively from all directions to Gradient distribution
Rectangular histogram, that is to say, that for 16 pixels in each 4 × 4 cell, according to certain rule come to this 16
The gradient direction and gradient magnitude of pixel carries out statistics with histogram, such as:By the gradient direction of pixel finally statistics 0 °,
On 45 °, 90 °, 135 °, 180 °, 225 °, 270 °, 315 ° and 360 ° this 8 directions, and why the gradient direction of pixel is when being worth
The gradient direction is classified as one of above-mentioned 8 directions, then can be according to predetermined rule depending on, such as:Gradient side when pixel
To during more than 0 ° less than 45 °, the gradient direction of the pixel is classified as 45 °, when the gradient direction of pixel is less than more than 45 °
When 90 °, the gradient direction of the pixel is classified as 90 ° etc., the gradient direction of pixel is being classified as one of above-mentioned 8 directions
Afterwards, the length of arrow has then corresponded to the superposition of different pixels point gradient magnitude in this direction in this direction, by this 44 × 4
Cell as seed point, shown in Figure 3, in Fig. 3 right side unit center of a lattice (round dot of black) be characterized a little, be
The distance between embodiment different pixels point and characteristic point, enclose Gauss weight to the gradient magnitude of each pixel, and then one
Individual characteristic point defines 8 × 4=32 dimensional vectors, so as to obtain 32 dimension description of characteristic point:W=(w1, w2... w32), in addition
For mitigating the impact that different characteristic point is caused to peripheral region because of illumination difference, it is standardized to describing son, in the present embodiment
32 above-mentioned dimension description are standardized by equation below:
And then obtain the characteristic vector of the later characteristic point of standardization:L=(l1, l2... l32).So far pre- by characteristic point
Gradient, gradient vector and the gradient magnitude for determining the pixel of neighborhood is described to the characteristic point, obtains the feature
The characteristic vector of point.
For similarly, for the characteristic point in second image, it is also adopted by above-mentioned method and obtains second figure
The characteristic vector of the characteristic point as in.Next based on described first image in characteristic point characteristic vector and second figure
The characteristic vector of the characteristic point as in, carries out bi-directional matching with life to the characteristic point in described first image and second image
Into initial matching point pair.
Specifically, in the following way the characteristic point in described first image and the second image is carried out in the present embodiment double
To coupling, first:Calculate the characteristic point in characteristic vector and second image of each characteristic point in described first image
Characteristic vector between European geometric distance, for example, the set that the characteristic vector of the characteristic point in the first image is located
For { a1、a2、a3…an(in set, each characteristic vector is 32 dimensions), the characteristic vector of the characteristic point in the second image is located
Set { b1、b2、b3…bn(in set, each characteristic vector is 32 dimensions), then calculate a respectively1And b1、a1And b2、…a1With
bnBetween European geometric distance, a2And b1、a2And b2、…a2And bnBetween European geometric distance, anAnd b1、anAnd b2、…an
And bnBetween European geometric distance.For a1For itself and b1、b2、b3…bnBetween European geometric distance in, have one
Minimum European geometric distance and one little European geometric distance, calculate the first ratio f therebetween1, similarly for a2And
Say itself and b1、b2、b3…bnBetween European geometric distance in, there is also the European geometric distance of minimum and one little Ou
Formula geometric distance, calculates the first ratio f therebetween2, the rest may be inferred obtains the spy of each characteristic point in the first image
Levy the minimum European geometric distance and time little European geometric distance between the characteristic vector of the characteristic point in vector and the second image
The first ratio, i.e. f1、f2、f3、f4…fn.If f1、f2、f3、f4…fnMiddle exist less than first threshold the first ratio, then with
The characteristic point in characteristic point and second image in described first image corresponding with the minimum European geometric distance is
3rd matching double points, for example, if f4The first threshold is less than, with f4Corresponding minimum European geometric distance be characterized to
Amount a4With characteristic vector b12Between European geometric distance, then in the characteristic point 4 in described first image and second image
Characteristic point 12 be the 3rd matching double points.For all the first ratios less than the first threshold, corresponding characteristic point is found out
To obtain the 3rd matching double points, all of 3rd matching double points constitute first set.In the present embodiment, the first threshold
For empirical value, the accuracy of the 3rd matching double points when being associated with coupling, first threshold described in the present embodiment may range from
0.1~10, if the first threshold can be 6.
Then, similarly calculate in second image in the characteristic vector of each characteristic point and described first image
Euclidean distance between the characteristic vector of characteristic point, i.e., calculate b respectively1And a1、b1And a2、…b1And anBetween European geometry away from
From b2And a1、b2And a2、…b2And anBetween European geometric distance, bnAnd a1、bnAnd a2、…bnAnd anBetween European geometry
Distance.For b1For itself and a1、a2、a3…anBetween European geometric distance in, there is also the European geometric distance of minimum
With one little European geometric distance, the second ratio s therebetween is calculated1, the rest may be inferred obtain each in the second image
Minimum European geometric distance between the characteristic vector of the characteristic point in the characteristic vector of individual characteristic point and the first image and time little
Second ratio of European geometric distance, i.e. s1、s2、s3、s4…sn.If s1、s2、s3、s4…snMiddle exist less than Second Threshold the
Two ratios, then with the characteristic point and described first image in second image corresponding with the minimum European geometric distance in
Characteristic point be the 4th matching double points, for example, if s4The Second Threshold is less than, with s4Corresponding minimum European geometry away from
From being characterized vectorial b4With characteristic vector a12Between European geometric distance, then with the characteristic point 4 in second image and institute
It is the 4th matching double points to state the characteristic point 12 in the first image.For all the second ratios less than the Second Threshold, find out
To obtain the 4th matching double points, all of 4th matching double points constitute second set to corresponding feature point pairs.In the present embodiment,
The Second Threshold be empirical value, the accuracy of the 4th matching double points when being associated with coupling, Second Threshold described in the present embodiment
May range from 0.1~10, as the Second Threshold can be 5.
Common factor is taken to above-mentioned first set and second set, the matching double points in common factor are described first image and described
The initial matching point pair of the second image.By way of bi-directional matching, the matching degree between matching double points can be improved, therefore
The matching degree of the final initial matching point pair for obtaining is preferable.
In the present embodiment, when carrying out bi-directional matching, for the characteristic vector of each characteristic point in the first image
During European geometric distance between the characteristic vector of the characteristic point in which is calculated with the second image, traveled through in the second image
The characteristic vector of all of characteristic point being calculated, similarly, for the feature of each characteristic point in the second image
During European geometric distance between the characteristic vector of the characteristic point for vector in which is calculated with the first image, it is to have traveled through the
In one image, the characteristic vector of all of characteristic point is being calculated.Consider actual application scenarios, described first image
And second image between may in the horizontal direction overall offset larger, skew in vertical direction is very little or does not deposit
?;Or may be larger in vertical direction overall offset between described first image and second image, in the horizontal direction partially
Move very little or do not exist.Such as:For the first image and the second image that X-ray shooting system is collected, person under inspection
It is parallel between place plane and detector place plane, the amplification difference between the first image and the second image can be neglected
Slightly disregard, therefore, between the first image and the second image, only exist the translation relation on position, i.e. the first image and the second image
Only exist in the horizontal direction with vertical direction on skew, when gathering image using X-ray shooting system, detector and bulb can
To move (also commonly referred to as detector and bulb are vertically moved) along column place direction, such as:Vertical position shoots, and also may be used
To move (also commonly referred to as detector and bulb are moved in the horizontal direction) along level bed X-axis place direction, such as:Clinostatism shoots.
If detector and bulb are moved along column place direction or level bed X-axis (major axis) place direction, it is located when detector is moved and leads
The precision of rail is sufficiently high, detector during moving along column place direction perpendicular direction movement less or
Movement is there is no, detector is less in perpendicular direction movement during moving along level bed X-axis place direction
Or there is no movement, then the first image and the second image the direction vertical with column place direction or with level bed
The skew in the vertical direction in X-axis place direction is smaller or is substantially absent from, namely between the first image and the second image
Skew only in a direction is larger.
It should be noted that stand position and shoot for detector and bulb are moved along column place direction in the present embodiment,
With column place direction as vertical direction (vertical coordinate), the direction vertical with column place direction is entered for horizontal direction (abscissa)
Row explanation, the abscissa and vertical coordinate of the image that corresponding vertical position is collected when shooting are corresponding with the direction of foregoing description.Right
Moved in detector and bulb in the horizontal direction, i.e., clinostatism shoots, with level bed X-axis place direction as horizontal direction (abscissa),
The direction vertical with level bed X-axis place direction be vertical direction (vertical coordinate), the image collected when being shot with clinostatism accordingly
Abscissa and vertical coordinate corresponding with the direction which describes.
In the present embodiment, if between described first image and second image may in the horizontal direction (abscissa) overall
Skew is larger (for example:Clinostatism shoots), the skew in vertical direction is very little or does not exist;Now first figure is being calculated
Minimum European between the characteristic vector of the characteristic point in the characteristic vector and second image of each characteristic point as in
During the ratio of distance and secondary little Euclidean distance, only can calculate the characteristic vector of each characteristic point in described first image with
It is less than the spy of the characteristic point of the first predeterminable range in second image with the vertical coordinate of this feature point apart from its poor absolute value
Levy the ratio of the minimum euclidean distance and time little Euclidean distance between vector;First predeterminable range be associated with matching accuracy rate and
The practical situation of the compromise and different system of algorithm execution time, first predeterminable range may range from 1~500
Pixel, if first predeterminable range can be 300 pixels.Each characteristic point in second image is calculated
Characteristic vector and described first image in characteristic point characteristic vector between minimum euclidean distance and secondary little Euclidean distance
Ratio when, only can calculate the characteristic vector of each characteristic point in second image with described first image with should
The vertical coordinate of characteristic point apart from its difference absolute value less than the second predeterminable range characteristic point characteristic vector between minimum Europe
Formula distance and the ratio of time little Euclidean distance;Second predeterminable range is associated with the compromise of matching accuracy rate and algorithm execution time
And the practical situation of different system, second predeterminable range may range from 1~500 pixel, and such as described second
Predeterminable range can be 300 pixels.For example, if the coordinate of the characteristic point 4 in the first image is (x4, y4), the second figure
The coordinate of the characteristic point 12 as in is (x12, y12), the coordinate of characteristic point 21 is (x21, y21), | y12-y4|、|y21-y4| respectively less than
First predeterminable range, then only calculate in the first image in the characteristic vector of characteristic point 4 and the second image characteristic point 12 and
The ratio of minimum European geometric distance and time little European geometric distance between the characteristic vector of characteristic point 21.
If may be in the larger (example of vertical direction (vertical coordinate) overall offset between described first image and second image
Such as:Vertical position shoots), offset very little in the horizontal direction or do not exist;Now calculate described first image in each
Minimum euclidean distance between the characteristic vector of the characteristic point in the characteristic vector of characteristic point and second image and secondary little Ou
During the ratio of formula distance, only can calculate the characteristic vector of each characteristic point in described first image with second image
In with the abscissa of this feature point apart from its poor absolute value less than between the characteristic vector of the characteristic point of the 3rd predeterminable range
Minimum euclidean distance and the ratio of time little Euclidean distance;3rd predeterminable range is associated with matching accuracy rate and algorithm execution time
Compromise and different system practical situation, the 3rd predeterminable range may range from 1~500 pixel, this reality
It can be 100 pixels to apply the 3rd predeterminable range described in example.Each characteristic point in second image is calculated
Minimum euclidean distance between the characteristic vector of the characteristic point in characteristic vector and described first image and secondary little Euclidean distance
Than when, only can calculate the characteristic vector of each characteristic point in second image with described first image with the spy
Levy minimum European between characteristic vector of the absolute value of abscissa a little apart from its difference less than the characteristic point of the 4th predeterminable range
Distance and the ratio of time little Euclidean distance;4th predeterminable range be associated with matching accuracy rate and algorithm execution time compromise with
And the practical situation of different system, the 4th predeterminable range may range from 1~500 pixel, institute in the present embodiment
It can be 400 pixels to state the 4th predeterminable range.For example, if the coordinate of the characteristic point 4 in the first image is (x4, y4),
The coordinate of the characteristic point 12 in the second image is (x12, y12), the coordinate of characteristic point 21 is (x21, y21), | x12-x4|、|x21-x4|
Respectively less than described 3rd predeterminable range, then only calculate in the first image characteristic point in the characteristic vector of characteristic point 4 and the second image
12 and characteristic point 21 characteristic vector between minimum European geometric distance and time little European geometric distance ratio.
Additionally, for above-mentioned X-ray shooting system, due to putting down between person under inspection place plane and detector place plane
OK, the amplification difference between the first image and the second image is negligible, now in described first image is calculated
Minimum euclidean distance between the characteristic vector of the characteristic point in the characteristic vector of each characteristic point and second image with
During the ratio of secondary little Euclidean distance, only can calculate the characteristic vector of each characteristic point in described first image with described the
Between the characteristic vector of the characteristic point for belonging to same difference pyramid image layer in two images with this feature point minimum European away from
From the ratio with secondary little Euclidean distance.For example, the characteristic point 4 in described first image the in difference pyramid diagram picture the 12nd
Layer, and therefore the characteristic point 1,3,6,12,21 in second image only can be counted also at the 12nd layer of difference pyramid diagram picture
Calculate the feature of the characteristic point 1,3,6,12,21 in characteristic vector and second image of the characteristic point 4 in described first image
The ratio of minimum European geometric distance and time little European geometric distance between vector.And it is each in second image is calculated
Minimum euclidean distance between the characteristic vector of the characteristic point in the characteristic vector and described first image of individual characteristic point is little with secondary
During the ratio of Euclidean distance, only can calculate the characteristic vector of each characteristic point in second image with first figure
As in belong to same difference pyramid image layer with this feature point characteristic point characteristic vector between minimum euclidean distance with
The ratio of secondary little Euclidean distance.
In addition, from the above for the image of X-ray shooting system collection, may between the first image and the second image
There is the skew of horizontal direction or vertical direction and amplification difference therebetween is negligible, therefore to described first
When characteristic point in image and the second image carries out bi-directional matching, between described first image and second image in water
Square larger to overall offset, skew in vertical direction is very little or does not exist (for example:Clinostatism shoots), the present embodiment exists
Calculate the characteristic vector of the characteristic point in characteristic vector and second image of each characteristic point in described first image
Between minimum euclidean distance and secondary little Euclidean distance ratio when, can also only calculate each feature in described first image
The characteristic vector of point is preset less than first apart from its poor absolute value with the vertical coordinate with this feature point in second image
Minimum euclidean distance between the characteristic vector of distance and the characteristic point for belonging to same difference pyramid image layer with this feature point
Ratio with secondary little Euclidean distance;The characteristic vector and described first image of each characteristic point in second image is calculated
In characteristic point characteristic vector between minimum euclidean distance and secondary little Euclidean distance ratio when, can also only calculate described
The characteristic vector of each characteristic point in two images with poor apart from it with the vertical coordinate of this feature point in described first image
Absolute value less than the second predeterminable range and with this feature point belong to the feature of the characteristic point of same difference pyramid image layer to
The ratio of minimum euclidean distance and time little Euclidean distance between amount.
Similarly, when the characteristic point in described first image and the second image carries out bi-directional matching, for vertical position is clapped
When taking the photograph, larger in vertical direction overall offset between described first image and second image, offset in the horizontal direction very
Little or do not exist;The characteristic vector of each characteristic point of the present embodiment in described first image is calculated and second figure
During the ratio of minimum euclidean distance and secondary little Euclidean distance between the characteristic vector of the characteristic point as in, only can also calculate described
The characteristic vector of each characteristic point in the first image with second image with the abscissa of this feature point apart from it
Poor absolute value is less than the 3rd predeterminable range and is belonged to the feature of the characteristic point of same difference pyramid image layer with this feature point
The ratio of minimum euclidean distance and time little Euclidean distance between vector;Each characteristic point in second image is calculated
Minimum euclidean distance between the characteristic vector of the characteristic point in characteristic vector and described first image and secondary little Euclidean distance
Than when, only can also calculate the characteristic vector of each characteristic point in second image with described first image with should
The abscissa of characteristic point is less than the 4th predeterminable range apart from the absolute value of its difference and belongs to same difference pyramid with this feature point
The ratio of minimum euclidean distance and time little Euclidean distance between the characteristic vector of the characteristic point of image layer.
In the present embodiment, when the characteristic point in described first image and second image carries out bi-directional matching, root
The feature in the characteristic vector of characteristic point in described first image and second image is calculated according to different application scenarios
Minimum European geometric distance and time little European geometric distance between the characteristic vector of point, can improve the speed of Feature Points Matching
Degree, and then the speed for determining matching double points is also improved, while the accurate of the initial matching point pair that coupling is obtained can also be improved
Degree.
So far above-mentioned steps are passed through, it is achieved that the described first image detected by S101 '~S103 ' and second figure
The coupling of the characteristic point as in, obtains initial matching point pair.But for the initial matching point of coupling acquisition is for, wherein
Can there are the matching double points of mistake, accordingly, it would be desirable to the matching double points of the initial matching point centering mistake are removed, in the present embodiment
The matching double points of the initial matching point centering mistake are removed especially by S105~S106.
In the present embodiment, the angle of the position relationship between the first image and the second image is accounted for, and removes institute
State the initial matching double points with point centering mistake that may be present.Consider between the first image and the second image in the horizontal direction
May there is skew, the first image for such as collecting and the second image using above-mentioned X-ray shooting system with vertical direction,
Therebetween only exist the translation relation on position, i.e. the first image and the second image is only existed in the horizontal direction and vertical direction
On skew, therefore the initial matching point pair can be removed according to actual position relationship between the first image and the second image
The middle wrong matching double points matching double points high to obtain accuracy.
From the foregoing, for different shooting positions, between the first image and the second image in the horizontal direction and vertically
The skew in direction is different, and for vertical position shoots, detector and bulb are vertically moved, and guide rail can only ensure the
One image and the second image are not in larger skew on level (abscissa) direction, and in vertical (vertical coordinate) direction
On, even if on the premise of feature point detection and characteristic vector coupling are relatively accurate, calculating all initial matching points pair for obtaining
The difference of vertical coordinate can be bigger than the difference of the abscissa of all initial matching points pair, therefore, the real coupling of initial matching point centering
Difference and the matching double points of mistake of the point to the vertical coordinate of (the final matching double points for determining or referred to as correct matching double points)
The difference of the difference of vertical coordinate also can be larger.And when detector and bulb are moved in the horizontal direction, guide rail can only ensure to treat first
Image and the second image are not in larger skew on vertical (vertical coordinate) direction, and on level (abscissa) direction,
Even if on the premise of feature point detection and characteristic vector coupling are relatively accurate, calculating the horizontal stroke of all initial matching points pair for obtaining
The difference of coordinate can be bigger than the difference of the vertical coordinate of all initial matching points pair, therefore, the real matching double points of initial matching point centering
The horizontal seat of the matching double points of the difference and mistake of the abscissa of (the final matching double points for determining or referred to as correct matching double points)
The difference of the difference of mark also can be larger.
Therefore, for detector and bulb are vertically moving, to the feature in the first image and the second image
Point is mated, and the difference of the vertical coordinate of the initial matching point pair of generation can bigger than the difference of abscissa (skew of vertical direction be more than
The skew of horizontal direction).And for detector and bulb are moved in the horizontal direction, to the first image and the second image in
Characteristic point is mated, and the difference of the abscissa of the initial matching point pair of generation can be than (the skew of horizontal direction greatly of the difference of vertical coordinate
The skew of vertical direction can be more than).Therefore, in the initial matching point pair for obtaining described first image and the second image in the present embodiment
Afterwards, can be by way of statistics with histogram respectively from the corresponding initial matching point of the difference that counts more close vertical coordinate to individual
Number, the corresponding initial matching point of the difference of more close abscissa to number two in terms of carry out, can to remove initial matching point centering
The matching double points of the mistake that can also exist.
Execute S105:Difference with the first coordinate of the initial matching point pair is as abscissa, corresponding with the difference of the first coordinate
Initial matching point to number be vertical coordinate generate the first rectangular histogram, determine that initial matching point is to number in first rectangular histogram
When sum is maximum, the initial matching point of place cluster is to for the first matching double points;In the present embodiment, if vertical position shoots, then the first figure
Less, the skew in vertical direction is larger for skew between picture and the second image in the horizontal direction, therefore, initial matching point centering
The difference of the difference of the vertical coordinate of the matching double points of the difference and mistake of the vertical coordinate of correct matching double points also can be larger, so, right
For Yu Liwei shoots, first can carry out statistics with histogram to the difference of the vertical coordinate of the initial matching point pair to obtain first
Right with point, be exactly specifically with the difference of the vertical coordinate of the initial matching point pair as abscissa, corresponding with the difference of vertical coordinate at the beginning of
Beginning matching double points number is that vertical coordinate generates the first rectangular histogram, determines that initial matching point is to number sum in first rectangular histogram
When maximum, the initial matching point of place cluster is to for the first matching double points.If clinostatism shoots, then the first image and the second image it
Between vertical direction skew less, offset in the horizontal direction larger, therefore, the correct matching double points of initial matching point centering
The difference of the difference of the abscissa of the matching double points of the difference of abscissa and mistake also can be larger, so for clinostatism shoots, can
First to carry out statistics with histogram to obtain the first matching double points to the difference of the abscissa of the initial matching point pair, it is exactly specifically
As abscissa, initial matching point corresponding with the difference of abscissa is vertical to number to difference with the abscissa of the initial matching point pair
The first rectangular histogram of Coordinate generation, determine initial matching point in first rectangular histogram to number sum when maximum place cluster initial
Matching double points are the first matching double points.
So that vertical position shoots as an example, the first image and the second image are respectively I1And I2, to I1And I2Characteristic point inspection is carried out
Survey and coupling is to obtain initial matching point to rear, if belonging to I in initial matching point pair1Pixel coordinate be (x1i,y1i)、
Belong to I2Pixel coordinate be (x2i,y2i), (number of i ∈ [1, N], N for initial matching point pair), with y2i-y1iFor horizontal seat
Mark, and to y2i-y1iThe number of corresponding initial matching point pair is counted, such as:Difference when the vertical coordinate of initial matching double points
For 6 pixel units when, the number of corresponding initial matching point pair is 20, according to the statistical information, generates with initial
With point to the difference of vertical coordinate be abscissa, the number of initial matching point pair corresponding with the difference of the vertical coordinate for vertical coordinate the
One rectangular histogram, by first rectangular histogram to initial matching point to corresponding initial matching point under the difference of different vertical coordinates
Statistics to number, can remove the initial matching point pair of mistake, this is because for the first figure for mating acquisition by S104
For, correct matching double points still account for major part to the initial matching point of picture and the second image, therefore, straight described first
When the initial matching point included in a cluster in square figure is most to number, the initial matching point pair included by its cluster being located
Probability for correct matching double points is maximum, therefore, retains in first rectangular histogram, and initial matching point to number sum is
Included initial matching point pair in its place cluster when maximum, using included initial matching point in other clusters to as mistake
Match point is removed, and using the initial matching point for retaining to as the first matching double points.
For shooting similarly, for clinostatism, by initial of first image and the second image of S104 coupling acquisitions
With a centering, correct matching double points still account for major part, therefore, with the difference of the abscissa of the initial matching point pair be
Abscissa, initial matching point corresponding with the difference of abscissa generate institute in a cluster in the first rectangular histogram to number for vertical coordinate
Comprising initial matching point most to number when, the initial matching point that included of cluster which is located is to for correct matching double points
Probability is maximum, therefore retains in first rectangular histogram, and initial matching point is to number sum by wrapping in its place cluster during maximum
The initial matching point pair for including, included initial matching point in other clusters is removed to the match point as mistake, and will be retained
Initial matching point to as the first matching double points.
In the present embodiment, the cluster can based on experience value depending on can also be obtained by experiment, namely described generating
Before first rectangular histogram, need based on experience value or test the value for obtaining and the width of cluster is defined.In the present embodiment, described
The width of cluster can be between 4~6 pixels.
The vertical seat of the most initial matching point pair that the difference for eliminating vertical coordinate by execution S105 is not belonging to calculate
The part of the difference of mark, or the difference of the abscissa difference of the abscissa of most initial matching point pair that is not belonging to calculate
Part, namely eliminate the matching double points of the initial matching point centering mistake to obtain the first matching double points.In the present embodiment
Either stand position to shoot or clinostatism shooting, which has skew with vertical direction in the horizontal direction, is eliminating described first
Image and second image after the matching double points of mistake that may be present, execute S106 in one direction, remove therebetween
Along the matching double points of another direction mistake that may be present, namely with the difference of the second coordinate of first matching double points as horizontal stroke
Coordinate, the first matching double points number corresponding with the difference of the second coordinate be vertical coordinate generate the second rectangular histogram, similarly, for
The initial matching point centering eliminates the coupling of partial error for the first matching double points of rear acquisition, at described first
Remain with the point correct matching double points of centering account for most, therefore, the first matching double points in second rectangular histogram
When number sum is maximum, the first matching double points that its place cluster is included are correct matching double points (the final matching double points for determining)
Probability be also maximum, therefore, retain in second rectangular histogram, the first matching double points number sum for maximum when its institute
The first included matching double points in cluster, included the first matching double points in other clusters are gone as the matching double points of mistake
Remove, and using the first matching double points for retaining as matching double points.In the present embodiment, if vertical position shoots then second coordinate being
Abscissa, that is, generate with the difference of abscissa as abscissa, and the first matching double points number corresponding with the difference of abscissa is vertical coordinate
The second rectangular histogram.If it is vertical coordinate that clinostatism shoots then second coordinate, that is, generate with the difference of vertical coordinate as abscissa, with
Second rectangular histogram of the corresponding first matching double points number of the difference of vertical coordinate for vertical coordinate.Finally in second rectangular histogram,
When determining that the first matching double points number sum is maximum, the first matching double points of place cluster are matching double points, and the width of the cluster can be with
Between 4~6 pixels.By executing S106, the difference for eliminating abscissa is not belonging to most first for calculating
With point to abscissa difference part, or the difference of vertical coordinate most first matching double points that are not belonging to calculate
The part of the difference of vertical coordinate, namely eliminate wrong matching double points in first matching double points to obtain matching double points.
It should be noted that during the actual matching double points for removing the mistake in the initial matching point pair, can
Can occur that the initial matching point in the first rectangular histogram included in different clusters, to number identical situation, is now needed to first
Rectangular histogram carries out mean filter, retains initial matching point to number sum in the first rectangular histogram after to have passed through mean filter
For maximum when its place cluster included in initial matching point to for the first matching double points, then with the of first matching double points
The difference of two coordinates is abscissa, and the first matching double points number corresponding with the difference of the second coordinate is that vertical coordinate generates the second Nogata
Figure, if there is also the first matching double points number identical situation included in different clusters in second rectangular histogram, now
Mean filter need to be carried out to second rectangular histogram, determine the first matching double points in the second rectangular histogram that have passed through mean filter
The first matching double points when number sum is maximum included in its place cluster are matching double points.
In the present embodiment, for the shooting of different positions, the angle of the position relationship between the first image and the second image
Degree considered, first carries out statistics with histogram to the difference of the first coordinate of the initial matching point pair wherein wrong to remove
Matching double points carry out statistics with histogram obtaining the first matching double points, then to the difference of the second coordinate of the first matching double points for obtaining
With the further matching double points for removing the mistake that may possibly still be present, accurate matching double points are filtered out, is therefore improve most
The accuracy of the matching double points for obtaining eventually, and match point other side is determined by the way of statistics with histogram in initial matching point pair
Method simple computation amount is little.
Additionally, in the present embodiment, described first image and the second image are whole except existing with vertical direction in the horizontal direction
Outside solid offsetting, it is also possible to there is the first image and the second image and there is larger skew in vertical direction and do not exist in the horizontal direction partially
Move or skew is less, or the first image and the second image have larger skew in the horizontal direction and do not exist partially in vertical direction
Move or offset less.For example:For above-mentioned X-ray shooting system, when detector and bulb vertically move (vertical position
Shoot), or when moving (clinostatism shooting) in the horizontal direction, if the precision of place guide rail is sufficiently high when detector is moved, detection
Device during vertically moving is moved in the horizontal direction less or there is no movement, and detector is along level
Direction is less in vertical direction movement during moving or there is no movement, then between the first image and the second image
The skew being likely to occur only in a direction is larger, and the skew in another direction is less or is substantially absent from, now
Initial matching point, now can be from the more close vertical seat of statistics to being close to zero along the skew of a coordinate in other words in a direction
The corresponding initial matching point of difference of mark to number, or from the corresponding initial matching point of the difference that counts more close abscissa to individual
Count to remove wrong matching double points present in initial matching point pair.
Specifically, the difference in the present embodiment with the coordinate of the initial matching point pair is as abscissa, corresponding with the difference of coordinate
Initial matching point be that vertical coordinate generates rectangular histogram to number, determine in the rectangular histogram that initial matching point is maximum to number sum
When place cluster initial matching point to for matching double points.Similarly, for the initial matching detected under this application scenarios
Point centering, account for most remain correct matching double points, therefore, included in a cluster in the rectangular histogram
When initial matching point is most to number, the initial matching point included by its cluster being located is to the probability for correct matching double points
Maximum, therefore initial matching point is to included initial matching point in its place cluster when number sum is maximum in reservation rectangular histogram
Right, included initial matching point in other clusters is removed to the matching double points as mistake.In the present embodiment, for the first figure
There is larger skew in vertical direction and there is no skew or skew less (vertical position shoots) in the horizontal direction in picture and the second image,
The coordinate is vertical coordinate;Do not exist in vertical direction for the first image and the second image have larger skew in the horizontal direction
Skew offsets less (clinostatism shooting), and the coordinate is abscissa;The width of the cluster before the rectangular histogram is generated,
Need based on experience value or test the value for obtaining to be defined.In the present embodiment, the width of the cluster can be in 4~6 pixels
Between point.
So far, achieved to the matching double points in the first image and the second image really by above-mentioned S101'~S106
Fixed, during determining matching double points, the detection complexity low velocity of characteristic point is fast, therefore also reduces to a certain extent
Determine the complexity of matching double points, improve the speed for determining matching double points.Characteristic point due to detecting is more stable, therefore
The accuracy of the initial matching point pair that coupling is obtained is also improved, in addition wrong coupling in the initial matching point pair is removed
During point pair, from the point of view of the position relationship between the first image and the second image, and then by way of statistics with histogram
Determine matching double points, method is simple, and amount of calculation is little, the accuracy of the matching double points of acquisition is high.Therefore, the present embodiment determines coupling
Point to method, reduce to a certain extent determine matching double points complexity, improve to a certain extent determination coupling
Point to speed, and the accuracy of the final matching double points for obtaining is high.
Corresponding to the method for above-mentioned determination matching double points, the present embodiment also provides a kind of device for determining matching double points,
The device for determining matching double points includes:
Resolving cell, for carrying out pyramid decomposition to the first image and the second image, based on decomposition after the first image
With the second image construct respectively the first difference pyramid diagram picture corresponding with described first image and with second image pair
The the second difference pyramid diagram picture that answers;Described first image and second image are adjacent image;
Search unit, for traveling through each layer of the first difference pyramid diagram picture, searches for every layer of difference pyramid diagram
Extreme point as in, travels through each layer of the second difference pyramid diagram picture, the pole that searches in every layer of difference pyramid diagram picture
Value point;The extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
First removal unit is strong for described first image in the extreme point of the first difference pyramid diagram picture for removing
Pixel on edge is obtaining the characteristic point in described first image;Remove the extreme point of the second difference pyramid diagram picture
In be pixel in the second image strong edge obtaining the characteristic point in second image;
Matching unit, initial to generate for being mated to the characteristic point in described first image and second image
Matching double points;
First histogram production unit, for the difference of the first coordinate of the initial matching point pair as abscissa, with
The corresponding initial matching point of the difference of one coordinate is that vertical coordinate generates the first rectangular histogram to number;
First matching double points determining unit, for determining that initial matching point is maximum to number sum in first rectangular histogram
When place cluster initial matching point to for the first matching double points;
Second histogram production unit, for the difference of the second coordinate of first matching double points as abscissa, with
The corresponding first matching double points number of the difference of two coordinates is that vertical coordinate generates the second rectangular histogram;
Matching double points determining unit, for determining the first matching double points number sum maximum when institute in second rectangular histogram
It is matching double points in the first matching double points of cluster.
The device of the determination matching double points of the present embodiment, removes the matching double points of the initial matching point centering mistake to obtain
The second removal unit for obtaining matching double points includes:First histogram production unit, the first matching double points determining unit, the second Nogata
Figure signal generating unit and matching double points determining unit.Described determine matching double points device be embodied as refer to described determine
With point to method enforcement, will not be described here.
The present embodiment also provides a kind of X-ray shooting system, including the device of above-mentioned determination matching double points.
Embodiment three
The present embodiment provides a kind of method for determining matching double points, from unlike embodiment two, removes in the present embodiment
The method of the matching double points of the initial matching point centering mistake is different from embodiment two, refers to Fig. 4, and Fig. 4 is the present invention
The schematic flow sheet of the method for the determination matching double points of embodiment three;In the present embodiment, the method bag for determining matching double points
Include:
S101′:Pyramid decomposition is carried out to the first image and the second image, based on decomposition after the first image and the second figure
As construct respectively corresponding with described first image the first difference pyramid diagram picture and with second image corresponding second
Difference pyramid diagram picture;Described first image and second image are adjacent image;
S102′:Each layer of the first difference pyramid diagram picture is traveled through, the pole that searches in every layer of difference pyramid diagram picture
Value point, travels through each layer of the second difference pyramid diagram picture, and the extreme point that searches in every layer of difference pyramid diagram picture is described
Extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
S103′:Remove in the extreme point of the first difference pyramid diagram picture as the picture in described first image strong edge
Vegetarian refreshments is obtaining the characteristic point in described first image;It is described in the extreme point for removing the second difference pyramid diagram picture
Pixel in two image strong edges is obtaining the characteristic point in second image;
S104:Characteristic point in described first image and second image is mated to generate initial matching point
Right;
S105':Slope of the initial matching point of described first image and the second image to place line is calculated, with described first
The slope of beginning matching double points place line is abscissa, and initial matching point corresponding with the slope is vertical coordinate generation to number
Slope histogram;
S106':Determine initial matching of the initial matching point to number sum place cluster when maximum in the slope histogram
Point is to for the first matching double points;
S107:Difference with the first coordinate of first matching double points as abscissa, corresponding with the difference of the first coordinate
One matching double points number is that vertical coordinate generates the first rectangular histogram, determines the first matching double points number sum in first rectangular histogram
When maximum, the first matching double points of place cluster are the second matching double points;
S108:Difference with the second coordinate of second matching double points as abscissa, corresponding with the difference of the second coordinate
Two matching double points numbers are that vertical coordinate generates the second rectangular histogram, determine the second matching double points number sum in second rectangular histogram
When maximum, the second matching double points of place cluster are matching double points.
In the present embodiment, S101 '~S104 generates initial matching point pair and the reality of described first image and second image
Apply similar in example two, here is omitted, in the present embodiment, removing the initial of described first image and second image
In matching double points during the matching double points of mistake, it is contemplated that the spy for the first image and the second image, with identical characteristics
Line between levying a little should be parallel, such as:For the first image and the second image that are collected by X-ray shooting system, the two
Line between the characteristic point in same anatomical should be parallel, correct between the first image and the second image in other words
Line has between matching double points parallel nature (namely the slope of correct matching double points place line is identical or very
It is close to), the line between characteristic point in non-equal anatomical structure can assume rambling state (the first image and the second figure
Line as between the matching double points of mistake is in disorderly and unsystematic state, namely the slope of the matching double points place line of mistake
It is in different value), therefore, in the present embodiment, statistics with histogram is carried out to the slope between initial matching point pair first, to pass through slope
Rectangular histogram is removed wherein wrong matching double points come the initial matching point to generating to screening, then again from the first image
And second the position relationship angle between image account for, carry out the initial matching point centering to the first image and the second image wrong
Matching double points are further removed by mistake, are collected with the first image and the second image as X-ray shooting system below
As a example by image, illustrate to removing the process of matching double points of initial matching point centering mistake, but technical scheme
This is not limited.
Execute S105':Calculate described first image I1With the second image I2Middle initial matching point is to the oblique of place line
Rate, referring to Fig. 5, Fig. 5 be the embodiment of the present invention three the first image of acquisition and the second image in initial matching point to place line
Slope schematic diagram, as shown in Figure 5:First image I described in the present embodiment1With the second image I2Close to and along level
Direction is placed side by side, I1The upper left corner be zero, I1Horizontal boundary (length of horizontal boundary be I1Wide W1) and X-axis
Overlap, (length on vertical border is the I on vertical border1And I2Height vertically) overlap with Y-axis, I2Horizontal sides
Boundary is also overlapped with X-axis.Pixel P1And P2For initial matching point pair, wherein pixel P1It is located at I1In, its coordinate is (x11,y11),
Pixel P2It is located at I2In, due to I1And I2Place side by side, therefore, in coordinate system as shown in Figure 5, pixel P2Coordinate be
(x21+W1,y22), m1For pixel P1With pixel P2Between line, then m1SlopeCan be obtained by equation below
?:
For I1And I2In all of initial matching point for, the slope of line can then pass through as follows therebetween
Formula is obtained:
Wherein:KiFor slope, (x of the i-th pair initial matching point to place line1i,y1i)、(x2i,y2i) it is that i-th pair is initial
Matching double points, (x1i,y1i) it is described first image I1In the position of i-th initial matching point, (x2i,y2i) it is described second
Image I2In the position of i-th initial matching point, W1For described first image I1Width.
Described first image I can be obtained by above-mentioned formula1With the second image I2In initial matching point to be located
The slope of line, to calculate the slope of acquisition as abscissa, the number of initial matching point pair corresponding with the slope is vertical coordinate
Generate slope histogram.For example, initial matching point is to P1And P2Between the slope of line beTo withCorresponding
The number of initial matching point pair is counted, with continued reference to Fig. 5, with m in Fig. 51Parallel m2、m3、m4Slope withPhase
With, therefore, in Fig. 5 withThe number of corresponding initial matching point pair is 4.According to the statistical information, you can set up slope
And the mapping relations between initial matching point pair, obtain slope histogram.By in the slope histogram to institute under Different Slope
Statistics of the corresponding initial matching point to number, can remove the matching double points of partial error, this is because mating by S104
The initial matching point centering of acquisition, correct matching double points still account for major part, therefore, in the slope histogram one
When initial matching point included in individual cluster is most to number, the initial matching point included by its cluster being located is to for correct
With point to probability also maximum.
Therefore, S106' is executed, determines that initial matching point is to place cluster during number sum maximum in the slope histogram
Initial matching point is to for the first matching double points.Namely retain in slope histogram, initial matching point is to when number sum is maximum
Included initial matching point pair in its place cluster, using included initial matching point in other clusters to the match point as mistake
To removing.In the present embodiment, the cluster can based on experience value depending on can also be obtained by experiment, namely generate described tiltedly
Before rate rectangular histogram, need based on experience value or test the value for obtaining and the width of cluster is defined.In the present embodiment, the cluster
Width can be between 4~6 pixels.For Fig. 5, exactly remain in Fig. 5 with m1、m2、m3、m4Corresponding first
Beginning matching double points, remove and m5、m6Corresponding initial matching point pair.
It should be noted that in the present embodiment, being to set up coordinate system in the way shown in fig. 5 to obtain described first image
With slope of the initial matching point to place line in the second image, in other embodiments, described first image and the second image
Vertically can also place up and down, now, initial of described first image and second image under the coordinate system
The expression way of the slope of place line is also slightly different with above-mentioned with, but for described first image and described second
For image, no matter setting up coordinate system in which way, in the slope histogram for ultimately generating, the initial matching point of reservation is to individual
When number sum is maximum, the initial matching point of place cluster is to all should be identical.Therefore, the first image described in the present embodiment and described
The initial matching point of two images should not be used as the restriction to technical solution of the present invention to the calculation of the slope of place line.
From embodiment two, described first image and second image may be present in horizontally and vertically direction partially
Move, still by taking X-ray shooting system as an example, when detector and vertically moving bulb, to the first image and the second image in
Characteristic point mated, the difference of the vertical coordinate of the initial matching point pair of generation can bigger than the difference of abscissa (vertical direction inclined
Move the skew more than horizontal direction).And detector and bulb are when moving in the horizontal direction, to the first image and the second image in
Characteristic point is mated, and the difference of the abscissa of the initial matching point pair of generation can be than (the skew of horizontal direction greatly of the difference of vertical coordinate
The skew of vertical direction can be more than).
Therefore, the error matching points in initial matching point pair is eliminated by slope histogram are to obtaining the first matching double points
Afterwards, can be by way of statistics with histogram respectively from corresponding first matching double points of the difference that counts more close vertical coordinate
Carry out in terms of number, the corresponding first matching double points number two of the difference of more close abscissa, so that remove can in the first matching double points
The matching double points of the mistake that can also exist.Execute S107:Difference with the first coordinate of first matching double points as abscissa, with
The corresponding first matching double points number of the difference of the first coordinate is that vertical coordinate generates the first rectangular histogram, determines in first rectangular histogram
When first matching double points number sum is maximum, the first matching double points of place cluster are the second matching double points;In the present embodiment, if the
Less, the skew in vertical direction is larger, such as skew between one image and the second image in the horizontal direction:Vertical position shoots, then
The difference of the difference of the vertical coordinate of the matching double points of the difference of the vertical coordinate of correct matching double points and mistake in first matching double points
Can be larger, so, first can carry out statistics with histogram to the difference of the vertical coordinate of first matching double points to obtain the second coupling
Point is right, is exactly specifically with the difference of the vertical coordinate of first matching double points as abscissa corresponding with the difference of vertical coordinate first
Matching double points number is that vertical coordinate generates the first rectangular histogram, determines that the first matching double points number sum is most in first rectangular histogram
When big, the first matching double points of place cluster are the second matching double points.If in the inclined of vertical direction between the first image and the second image
Move less, skew in the horizontal direction is larger, such as:Clinostatism shoot, then in the first matching double points correct matching double points horizontal seat
The difference of the difference of the abscissa of the matching double points of the difference of mark and mistake also can be larger, so, can first to first match point
To the difference of abscissa carry out statistics with histogram to obtain the second matching double points, specific with first matching double points
The difference of abscissa is abscissa, and the first matching double points number corresponding with the difference of abscissa is that vertical coordinate generates the first rectangular histogram,
When determining that the first matching double points number sum is maximum in first rectangular histogram, the first matching double points of place cluster match somebody with somebody point for second
Right.
So that vertical position shoots as an example, the first image and the second image are respectively I1And I2, to I1And I2Carry out based on slope
After the mode of statistics with histogram eliminates matching double points first matching double points of acquisition of partial error, if in the first matching double points
Belong to I1Pixel coordinate be (x1i,y1i), belong to I2Pixel coordinate be (x2i,y2i), (i ∈ [1, N], N are
The number of one matching double points), with y2i-y1iFor abscissa, and to y2i-y1iThe number of corresponding first matching double points is united
Meter, such as:When the difference of the vertical coordinate of the first matching double points is 4 pixel units, the number of corresponding the first matching double points
For 120, according to the statistical information, the difference pair with the difference of the vertical coordinate of the first matching double points as abscissa, with the vertical coordinate is generated
First rectangular histogram of the number of the first matching double points that answers for vertical coordinate, by existing to the first matching double points in first rectangular histogram
The statistics of the first corresponding matching double points number under the difference of different vertical coordinates, can remove the first matching double points of mistake, this
It is because still account for major part by correct matching double points in the first matching double points of determination in S106', therefore, described
When the first matching double points number included in a cluster in first rectangular histogram is most, included first of cluster which is located
Maximum to the probability for correct matching double points with point, therefore, retain in first rectangular histogram, the first matching double points number
Included the first matching double points in its place cluster when sum is maximum, using included the first matching double points in other clusters as
The match point of mistake is removed, and using the first matching double points for retaining as the second matching double points.
For shooting similarly, for clinostatism, to the first image I1With the second image I2Carry out based on slope histogram
After the mode of statistics eliminates matching double points first matching double points of acquisition of partial error, correct matching double points still account for greatly
Part, therefore, with the difference of the abscissa of first matching double points as abscissa, corresponding with the difference of abscissa first mates
When the first matching double points number included in a cluster in the first rectangular histogram that point is generated for vertical coordinate to number is most, its
The first matching double points that the cluster at place is included are that the probability of correct matching double points is maximum, therefore retain first rectangular histogram
In, when the first matching double points number sum is maximum, included the first matching double points in its place cluster, will be wrapped in other clusters
The first matching double points for including are removed as the match point of mistake, and using the first matching double points for retaining as the second matching double points.
In the present embodiment, the cluster can based on experience value depending on can also be obtained by experiment, namely described generating
Before first rectangular histogram, need based on experience value or test the value for obtaining and the width of cluster is defined.In the present embodiment, described
The width of cluster can be between 4~6 pixels.
The vertical seat of most first matching double points that the difference for eliminating vertical coordinate by execution S107 is not belonging to calculate
The part of the difference of mark, or the difference of the abscissa difference of the abscissa of most first matching double points that is not belonging to calculate
Part, namely eliminate wrong matching double points in first matching double points to obtain the second matching double points.In the present embodiment,
As described first image and second image have skew with vertical direction in the horizontal direction, therefore first is being eliminated
Image I1With the second image I2In one direction after the matching double points of mistake that may be present, S108 is executed, edge therebetween is removed
The matching double points of another direction mistake that may be present, namely with the difference of the second coordinate of second matching double points as horizontal seat
Mark, the second matching double points number corresponding with the difference of the second coordinate are that vertical coordinate generates the second rectangular histogram, determine that described second is straight
When in square figure, the second matching double points number sum is maximum, the second matching double points of place cluster are matching double points.In the present embodiment, if
Offset less between described first image and the second image in the horizontal direction, the skew in vertical direction is larger, such as:Clap vertical position
Take the photograph, then second coordinate is abscissa, that is, generate with the difference of abscissa as abscissa, corresponding with the difference of abscissa second
Match somebody with somebody the second rectangular histogram that puts to number is vertical coordinate.If between described first image and the second image vertical direction skew not
Greatly, in the horizontal direction skew is larger, such as:Clinostatism shoots, then second coordinate is vertical coordinate, that is, generate the difference with vertical coordinate
For abscissa, second rectangular histogram of the second matching double points number corresponding with the difference of vertical coordinate for vertical coordinate.Final described the
In two rectangular histograms, when determining that the second matching double points number sum is maximum, the second matching double points of place cluster are matching double points, described
The width of cluster can be between 4~6 pixels.By executing S108, eliminate that the difference of abscissa is not belonging to calculate is big
The part of the difference of the abscissa of the second partial matching double points, or the difference of vertical coordinate be not belonging to calculate most
The part of the difference of the vertical coordinate of two matching double points, namely eliminate wrong matching double points in second matching double points to obtain
Matching double points.
It should be noted that during the actual matching double points for removing the mistake in the initial matching point pair, can
Can occur that the initial matching point in slope histogram included in different clusters, to number identical situation, is now needed to slope
Rectangular histogram carries out mean filter, retains initial matching point to number sum in the slope histogram after to have passed through mean filter
For maximum when its place cluster included in initial matching point to for the first matching double points, then with the of first matching double points
The difference of one coordinate is abscissa, and the first matching double points number corresponding with the difference of the first coordinate is that vertical coordinate generates the first Nogata
Figure, if there is also the first matching double points number identical situation included in different clusters in first rectangular histogram, now
Mean filter need to be carried out to first rectangular histogram, retain first in the first rectangular histogram after to have passed through mean filter and mate
Point to the first matching double points included in its place cluster when number sum is maximum be the second matching double points, then with this second
The difference of the second coordinate of matching double points is abscissa, and the second matching double points number corresponding with the difference of the second coordinate is that vertical coordinate is given birth to
Into the second rectangular histogram, if still suffering from the second matching double points number identical feelings included in different clusters in second rectangular histogram
Condition, then carry out mean filter to second rectangular histogram, determines the second match point in the second rectangular histogram that have passed through mean filter
The second matching double points when being maximum to number sum included in its place cluster are matching double points.
In the present embodiment, for being offset between described first image and the second image in the horizontal direction less, in vertically side
To skew larger, such as:Vertical position shoots, and first eliminates initial matching point centering portion mistake based on the mode of slope histogram
Matching double points obtain the first matching double points, and then the difference with the vertical coordinate of first matching double points is as abscissa, with vertical coordinate
The corresponding first matching double points number of difference be that vertical coordinate generates the first rectangular histogram to determine the second matching double points, finally with described
The difference of the abscissa of the second matching double points is abscissa, and the second matching double points number corresponding with the difference of abscissa is that vertical coordinate is given birth to
Into the second rectangular histogram determining matching double points.For between described first image and the second image vertical direction skew not
Greatly, in the horizontal direction skew is larger, such as:Clinostatism shoots, and first eliminates initial matching point pair based on the mode of slope histogram
The matching double points of middle partial error obtain the first matching double points, and then the difference with the abscissa of first matching double points is as horizontal seat
Mark, the first matching double points number corresponding with the difference of abscissa are that vertical coordinate generates the first rectangular histogram to determine the second match point
Right, finally the difference with the vertical coordinate of second matching double points is as abscissa, the second matching double points corresponding with the difference of vertical coordinate
Number is that vertical coordinate generates the second rectangular histogram to determine matching double points.For the shooting of different positions, the first image had both been considered
And the second anatomical structure characteristic between image, and the angle of the position relationship between the first image and the second image carries out
Consider to remove the match point of the initial matching point centering mistake of the first image and the second image as far as possible, therefore in very great Cheng
On improve final acquisition matching double points accuracy, and determined in initial matching point pair by the way of statistics with histogram
Matching double points method simple computation amount is little.
The method of the determination matching double points of the present embodiment, when feature point detection is carried out, detects that the complexity of characteristic point is low,
Speed is fast, therefore reduces the complexity for determining matching double points to a certain extent, improves the speed for determining matching double points.By
More stable in the characteristic point for detecting, the accuracy of the initial matching point pair that coupling is obtained therefore is also improved, described removing
During the matching double points of initial matching point centering mistake, by the way of based on slope histogram statistics and statistics with histogram, remove mistake
Matching double points, complexity is low, amount of calculation is little, and the accuracy of the matching double points of acquisition is high by mistake.Therefore, the determination coupling of the present embodiment
Point to method, reduce to a great extent determine matching double points complexity, improve to a great extent determination coupling
Point to speed, and the accuracy of the final matching double points for obtaining is high.
From embodiment two, described first image and second image are except existing with vertical direction in the horizontal direction
Outside overall offset, it is also possible to there is described first image and second image and there is larger skew in level side in vertical direction
To there is no skew or offseting less, or there is larger skew in described first image and second image in the horizontal direction
There is no skew in vertical direction or offset less.For example:For above-mentioned X-ray shooting system, when detector and bulb
(vertical position shoot) is vertically moved, or when moving (clinostatism shooting) in the horizontal direction, if be located when detector is moved leading
The precision of rail is sufficiently high, and detector during vertically moving is moved in the horizontal direction less or be there is no
Mobile, detector is less in vertical direction movement during moving in the horizontal direction or there is no movement, initially
Skew of the matching double points in a direction in other words along a coordinate is close to zero.
Therefore, in another embodiment, statistics with histogram can be carried out to the slope between initial matching point pair first, is passed through
Slope histogram is removed wherein wrong matching double points come the initial matching point to generating to screening, then again from statistics
The corresponding initial matching point of the difference of more close vertical coordinate is to number or corresponding just from the difference for counting more close abscissa
Beginning matching double points number is removing the matching double points of the mistake that initial matching point centering is still suffered from.
Specifically, it is exactly first to calculate the initial matching point of described first image and second image to the oblique of place line
Rate, with the initial matching point to the slope of place line as abscissa, initial matching point corresponding with the slope is to number
Slope histogram is generated for vertical coordinate;Determine that initial matching point is to place cluster during number sum maximum in the slope histogram
Initial matching point is to for the first matching double points;Difference pair with the difference of the coordinate of first matching double points as abscissa, with coordinate
The the first matching double points number that answers is that vertical coordinate generates rectangular histogram;Determine that the first matching double points number sum is most in the rectangular histogram
When big, the first matching double points of place cluster are matching double points.If described first image and second image are present in vertical direction
There is no skew in larger skew or skew less in the horizontal direction, and such as vertical position shoots, then with the vertical of first matching double points
The difference of coordinate is abscissa, and the first matching double points number corresponding with the difference of vertical coordinate is that vertical coordinate generates rectangular histogram.If described
First image and second image exist in the horizontal direction larger skew vertical direction do not exist skew or offset less,
As clinostatism shoots, then the difference with the abscissa of first matching double points is as abscissa, corresponding with the difference of abscissa first
It is that vertical coordinate generates rectangular histogram to number with point.
In another embodiment, it is also possible to by way of slope histogram, only remove the initial matching point centering mistake
Matching double points.
So far, the initial matching point centering of described first image and second image is eliminated by S101 '~S108
The matching double points of the mistake of presence, in the present embodiment, first eliminate initial matching point to middle part based on the mode of slope histogram
Misclassification matching double points by mistake obtain the first matching double points, and then the difference with the first coordinate of first matching double points is as horizontal seat
Mark, the first matching double points number corresponding with the difference of the first coordinate are that vertical coordinate generates the first rectangular histogram to determine the second match point
Right, finally with the second of second matching double points the difference that marks as abscissa, the second match point corresponding with the difference of the second coordinate
It is that vertical coordinate generates the second rectangular histogram to determine matching double points to number.Both solution first image and second image between had been considered
Architectural characteristic is cutd open, and the angle of the position relationship between the first image and the second image has carried out considering to remove as far as possible
The match point of the initial matching point centering mistake of the first image and the second image, therefore improves final acquisition on very great Cheng
The accuracy of matching double points, reduces the complexity for obtaining matching double points to a great extent, and using slope histogram and straight
The mode of side's figure statistics determines that in initial matching point pair matching double points method simple computation amount is little.
The method of corresponding above-mentioned determination matching double points, the present embodiment also provide a kind of device for determining matching double points, institute
State and determine that the device of matching double points includes:
Resolving cell:For carrying out pyramid decomposition to the first image and the second image, based on decomposition after the first image
With the second image construct respectively the first difference pyramid diagram picture corresponding with described first image and with second image pair
The the second difference pyramid diagram picture that answers;Described first image and second image are adjacent image;
Search unit:For traveling through each layer of the first difference pyramid diagram picture, every layer of difference pyramid diagram is searched for
Extreme point as in, travels through each layer of the second difference pyramid diagram picture, the pole that searches in every layer of difference pyramid diagram picture
Value point, the extreme point are associated with the absolute value of the gray value of the pixel in default neighborhood;
First removal unit:Strong for described first image in the extreme point of the first difference pyramid diagram picture for removing
Pixel on edge is obtaining the characteristic point in described first image;Remove the extreme point of the second difference pyramid diagram picture
In be pixel in the second image strong edge obtaining the characteristic point in second image;
Matching unit:Initial to generate for being mated to the characteristic point in described first image and second image
Matching double points;
Slope histogram signal generating unit:Connect to being located for calculating the initial matching point of described first image and the second image
The slope of line, with the initial matching point to the slope of place line as abscissa, initial matching point corresponding with the slope
It is that vertical coordinate generates slope histogram to number;
First matching double points determining unit:For determining that initial matching point is maximum to number sum in the slope histogram
When place cluster initial matching point to for the first matching double points;
First histogram production unit:For with the difference of the first coordinate of first matching double points as abscissa, with
The corresponding first matching double points number of the difference of one coordinate is that vertical coordinate generates the first rectangular histogram;
Second matching double points determining unit:For determining that the first matching double points number sum is maximum in first rectangular histogram
When place cluster the first matching double points be the second matching double points;
Second histogram production unit:For with the difference of the second coordinate of second matching double points as abscissa, with
The corresponding second matching double points number of the difference of two coordinates is that vertical coordinate generates the second rectangular histogram;
Matching double points determining unit, for determining the second matching double points number sum maximum when institute in second rectangular histogram
It is matching double points in the second matching double points of cluster.
The device of the determination matching double points of the present embodiment, removes the matching double points of the initial matching point centering mistake to obtain
The second removal unit for obtaining matching double points includes:Slope histogram signal generating unit, the first matching double points determining unit, the first Nogata
Figure signal generating unit, the second matching double points determining unit, the second histogram production unit and matching double points determining unit.The determination
The enforcement for being embodied as referring to the method for determining matching double points of the device of matching double points, will not be described here.
The present embodiment also provides a kind of X-ray shooting system, including the device of above-mentioned determination matching double points.
Example IV
The present embodiment provides a kind of image acquiring method, refers to Fig. 6, and Fig. 6 is that the image of the embodiment of the present invention four is obtained
The schematic flow sheet of method, as shown in Figure 6:Described image acquisition methods include:
S201:Pyramid decomposition is carried out to the first image and the second image, based on decomposition after the first image and the second figure
As construct respectively corresponding with described first image the first difference pyramid diagram picture and with second image corresponding second
Difference pyramid diagram picture;Described first image and second image are adjacent image;
S202:Each layer of the first difference pyramid diagram picture is traveled through, the pole that searches in every layer of difference pyramid diagram picture
Value point, travels through each layer of the second difference pyramid diagram picture, and the extreme point that searches in every layer of difference pyramid diagram picture is described
Extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
S203:Remove in the extreme point of the first difference pyramid diagram picture as the pixel in described first image strong edge
Put to obtain the characteristic point in described first image;It is described second to remove in the extreme point of the second difference pyramid diagram picture
Pixel in image strong edge is obtaining the characteristic point in second image;
S204:Characteristic point in described first image and second image is mated to generate initial matching point
Right;
S205:Remove the matching double points of the initial matching point centering mistake to obtain matching double points;
S206:Determined between described first image and second image based on the position relationship between the matching double points
Skew;
S207:Described first image and described are determined according to the skew between described first image and second image
The overlapping region of two images;
S208:Described first image and second image are spliced according to the overlapping region.
The image obtained in the present embodiment is spliced image, and described first image and second image can be logical
Cross the image that X-ray shooting system is collected, due to described first image and second image are carried out feature point detection and
Coupling, is mainly used for splicing the first image and the second image, therefore, in the present embodiment before S201 is executed, can
First to carry out pretreatment to described first image and second image, specifically, can be estimated according to actual experience first
Described first image and the initial overlapping region of second image, according to the initial of described first image and second image
The height of overlapping region determines first area corresponding with the initial overlapping region in described first image, claims in the present embodiment
Be the 3rd image, determine second area corresponding with the initial overlapping region in second image, title in the present embodiment
Be the 4th image.Pretreatment is carried out to the 3rd image and the 4th image, is exactly specifically to the 3rd image
Tonal range mapping is carried out with the 4th image, the gray value of the 3rd image and the 4th image is mapped to identical
Scope.For example, if the gray value of the 3rd image belongs to 0~255, the gray value of the 4th image belongs to 0~
4096, then the 0~4096 of the 4th image can be mapped to by linear for the 0~255 of the 3rd image gray value
In intensity value ranges.
Furthermore, it is contemplated that when splicing to described first image and second image, region (real weight to be spliced
Folded region) belong to the initial overlapping region, so in detection described first image and second image in the present embodiment
In characteristic point when, that is, execute S201~S203 during, be the part area to described first image and second image
Domain is that above-mentioned initial overlapping region is detected, namely the of the first area to described first image and second image
One region is detected, or perhaps the characteristic point to the 3rd image and the 4th image is detected.
Therefore, when executing S201 during stitching image is obtained, it is the parts of images to described first image and institute
The parts of images for stating the second image carries out pyramid decomposition, that is to say, that be to the 3rd image and the 4th figure in S201
As carrying out pyramid decomposition, and based on decomposition after the 3rd image and the 4th image construct respectively corresponding with the 3rd image
3rd difference pyramid diagram picture and the 4th difference pyramid diagram picture corresponding with the 4th image.And it is concrete how to the 3rd
Image and the 4th image carry out pyramid decomposition, and construct the 3rd difference pyramid diagram picture and the 4th difference pyramid diagram picture, can
So that referring to S101, here is omitted.
Similarly, when executing S202 during stitching image is obtained, it is traversal the 3rd difference pyramid diagram picture
Each layer, the extreme point that searches in every layer of difference pyramid diagram picture travels through each layer of the 4th difference pyramid diagram picture,
Extreme point in every layer of difference pyramid diagram picture of search, the extreme point are associated with the gray value of the pixel in default neighborhood
Absolute value;When executing S203, it in the extreme point for remove the 3rd difference pyramid diagram picture is the 3rd image strong edge to be
On pixel obtaining the characteristic point in the 3rd image;In the extreme point for removing the 4th difference pyramid diagram picture it is
Pixel in the 4th image strong edge is to obtain the characteristic point in the 4th image, and the spy in the 3rd image
The characteristic point a little and in described first image is levied, the characteristic point in the 4th image is also the feature in second image
Point.
Additionally, from the foregoing, determine difference of Gaussian pyramid diagram picture in extreme point when, the extreme point can be
The pixel P of top N is located at after sorting according to order from large to small to the absolute value of the gray value of pixel P, is spelled obtaining
During map interlinking picture, when determining the extreme point in the 3rd difference pyramid diagram picture, the extreme point for being taken can be then right
According to the pixel P for being located at front M positions after order sequence from large to small, (pixel P is referred to the absolute value of the gray value of pixel P
In per layer of the 3rd difference pyramid diagram picture, the absolute value of the gray value of pixel is picture in the default neighborhood of the pixel
The pixel of the maximum absolute value of the gray value of vegetarian refreshments), M=N × initial overlapping region height.Similarly, determine the described 4th
During extreme point in difference pyramid diagram picture, the extreme point for being taken can also be the gray value to pixel P absolute value according to
After order sequence from large to small, positioned at the pixel P of front M positions, (pixel P refers to the every of the 4th difference pyramid diagram picture
In layer, the absolute value of the gray value of pixel is the maximum absolute value of the gray value of pixel in the default neighborhood of the pixel
Pixel).
The present embodiment when stitching image is obtained, due to considering the detection to characteristic point and coupling is mainly used for the
One image and the second image are spliced, and therefore, when characteristic point is detected, the first image of view picture and the second image are not carried out
Detection, but it is pointed to the initial overlapping region in the first image i.e. the 3rd image and the initial overlay region in the second image
Domain is detection and the coupling that the 4th image has carried out characteristic point, therefore can improve the detection speed of characteristic point, further
The speed of the matching double points for determining described first image and second image can be improved.
After the characteristic point in execution S201~S203 acquisition described first images and second image, S204 is executed
~S205 is mated to generate initial matching point pair to the characteristic point in described first image and second image, removes institute
State the matching double points of initial matching point centering mistake to obtain matching double points.To described first image and described in the present embodiment
How characteristic point in two images carries out mating the description that may refer in embodiment two, removes the initial matching point centering wrong
Matching double points by mistake are described initial to obtain the removal that matching double points are then may refer to described in embodiment two and embodiment three
The method of the matching double points with a centering mistake, here is omitted.
After matching double points being obtained by S201~S205, execute S206 based on the position relationship between the matching double points
Determine the skew between described first image and second image.Specifically, it is with the horizontal seat of matching double points in the present embodiment
The average of the difference of mark is used as the skew between described first image and second image in the horizontal direction, indulging with matching double points
The average of the difference of coordinate is used as the skew between described first image and second image in vertical direction.For example:If
The final matching double points for obtaining are K, wherein belong to the first image I1Pixel coordinate be (x1i,y1i), belong to the second figure
As I2Pixel coordinate be (x2i,y2i), then the skew between described first image and second image in the horizontal directionBetween described first image and second image vertical direction skew
In the present embodiment, if the skew between described first image and second image in the horizontal direction is very little (such as:
Vertical position shoots or longitudinal spliced), therefore Δ x goes to zero, if in vertical direction between described first image and second image
Skew very little (such as:Clinostatism shoots or horizontally-spliced), then Δ y goes to zero.
S207 is executed, between the described first image obtained according to S206 and second image in the horizontal direction and vertically
The skew in direction determines overlapping region therebetween.From the foregoing, being to estimate institute according to actual experience in the present embodiment
The initial overlapping region of the first image and second image is stated, to described first image and described only in initial overlapping region
The characteristic point of the second image is detected, can so improve the detection speed of characteristic point, and then improve determination matching double points
Speed and the speed of image mosaic.Therefore in this step, still with described first image I1With the second image I2In, respectively
It is extracted initial overlapping region I1' (the 3rd image) and initial overlapping region I2' (the 4th image) (I1' and I2' picture size
Identical) to initial overlapping region I1' and I2' carry out feature point detection and coupling after obtain I1' and I2' between in the horizontal direction
Shifted by delta x, as a example by shifted by delta y of vertical direction, illustrate how obtain I1' and I2' between overlapping region.
Specifically, I is determined as follows in the present embodiment1' and I2' between overlapping region.Set up in the present embodiment
Level is X-axis positive direction to the right, is the rectangular coordinate system of Y-axis positive direction straight down, due to I1' and I2' picture size phase
With, therefore I1' and I2' height H (along Y direction) and I1' and I2' width W (along X-direction) all same.With I2' top
Boundary and I1' coboundary between relative position relation as I1' and I2' vertical direction shifted by delta y, with I2' left margin
With I1' left margin relative position relation as I1' and I2' shifted by delta x in the horizontal direction.May in actual splicing
Occur that Δ x, Δ y are all higher than zero, Δ x is less than zero more than zero, Δ y, and Δ x is more than zero less than zero, Δ y, and Δ x, Δ y are respectively less than zero
Situation, below in conjunction with the position relationship schematic diagram between first second image of image of Fig. 7-a to Fig. 7-d, to these four feelings
Under condition, the determination of overlapping region carries out simple illustration.
Furthermore, it is necessary to illustrate, Fig. 7-a to Fig. 7-d are only that the position relationship between the first image and the second image shows
It is intended to, it illustrates the situation that there is skew between the first image and the second image in the horizontal direction with vertical direction, not
Illustrate between the first image that may be present and the second image in the horizontal direction or vertical direction the feelings that go to zero of skew
Condition, but technical scheme is not limited to the situation shown in Fig. 7-a to Fig. 7-d.
In the present embodiment, if Δ x, Δ y are all higher than zero, as shown in Fig. 7-a, I2' coboundary and I1' coboundary equal
It is located at Y-axis positive, I2' coboundary be located at I1' coboundary lower section, I2' coboundary relative to I1' coboundary along Y-axis
Positive skew is I1' and I2' vertical direction shifted by delta y, now I1' and I2' vertical direction overlap height be H- Δs
y;I2' left margin and I1' left margin positive, the I that is respectively positioned on X-axis2' left margin be located at I1' left margin right, I2'
Left margin is relative to I1' left margin be I along the positive skew of X-axis1' and I2' shifted by delta x in the horizontal direction, I1' and I2'
The width that horizontal direction is overlapped is W- Δ x.Determining I1' and I2' in the horizontal direction with vertical direction overlap width after, then
Can determine I1' and I2' overlapping region, and I1' and I2' between overlapping region, namely I1And I2Between overlapping region.
If Δ x is less than zero, as shown in Fig. 7-b, I more than zero, Δ y2' coboundary be located at Y-axis negative sense, I1' coboundary
It is located at X-axis, I2' coboundary be located at I1' coboundary top, I2' coboundary relative to I1' coboundary skew along Y
Axle negative sense, now I1' and I2' vertical direction overlap height be H+ Δ y;I2' left margin and I1' left margin be respectively positioned on X
Axle is positive, I2' left margin be located at I1' left margin right, I2' left margin relative to I1' left margin positive along X-axis
Skew be I1' and I2' shifted by delta x in the horizontal direction, I1' and I2' the width that overlaps in the horizontal direction is W- Δ x.Determining
I1' and I2' in the horizontal direction with vertical direction overlap width after, then can determine I1' and I2' overlapping region, and I1' and
I2' between overlapping region, namely I1And I2Between overlapping region.
If Δ x is more than zero, as shown in Fig. 7-c, I less than zero, Δ y2' coboundary and I1' coboundary be respectively positioned on Y-axis
Forward direction, I2' coboundary be located at I1' coboundary lower section, now I1' and I2' vertical direction overlap height be H- Δ y;
I2' left margin be located at Y-axis, I1' left margin positive, the I that is located at X-axis2' left margin be located at I1' left margin left, I2′
Left margin relative to I1' left margin skew along X-axis negative sense, I1' and I2' the width that overlaps in the horizontal direction is W+ Δ x.
Determining I1' and I2' in the horizontal direction with vertical direction overlap width after, then can determine I1' and I2' overlapping region,
And I1' and I2' between overlapping region, namely I1And I2Between overlapping region.
If Δ x is less than zero, as shown in Fig. 7-d, I less than zero, Δ y2' coboundary be located at Y-axis negative sense, I1' coboundary
It is located at X-axis, I2' coboundary be located at I1' coboundary top, I2' coboundary relative to I1' coboundary skew along Y
Axle negative sense, now I1' and I2' vertical direction overlap height be H+ Δ y;I2' left margin be located at Y-axis, I1' left margin
It is located at X-axis positive, I2' left margin be located at I1' left margin left, I2' left margin relative to I1' left margin inclined
Move along X-axis negative sense, I1' and I2' the width that overlaps in the horizontal direction is W+ Δ x.Determining I1' and I2' in the horizontal direction and perpendicular
Nogata then can determine I to after the width for overlapping1' and I2' overlapping region, and I1' and I2' between overlapping region, namely I1
And I2Between overlapping region.
It should be noted that being with I in the present embodiment2' coboundary and I1' coboundary between relative position relation make
For I1' and I2' vertical direction shifted by delta y, with I2' left margin and I1' left margin relative position relation as I1' and
I2' shifted by delta x in the horizontal direction.In other embodiments, I1' and I2' can also be with I in shifted by delta y of vertical direction2'
Lower boundary and I1' lower boundary between relative position relation depending on, I1' and I2' shifted by delta x in the horizontal direction can also be with
I2' right margin and I1' right margin relative position relation depending on, this is not limited in the present embodiment.
S208 is executed, according to the overlapping region of first image and the second image of S207 acquisitions, to spelling therebetween
Connect, still with above-mentioned I1And I2As a example by, from the foregoing, being actually to the first image I in S2071With the second image I2In
Initial overlapping region I1' and I2' overlapping region determined, and I1' and I2' overlapping region namely I1And I2Overlay region
Domain.In the present embodiment, using the center of the overlapping region as splice point, first determine splice point in I1' and I2' in position coordinateses,
Due to I1' and I2' and original image I1And I2Between relative position relation be known, therefore could be aware that splice point in I1Position
Put and splice point is in I2Position, the position for being then based on splice point is come to I1And I2Spliced, and to spliced image
Merged.Specifically, still set up in the present embodiment with level to the right as X-axis forward direction, be that the positive right angle of Y-axis is sat straight down
Mark system, I1And I2The upper left corner be respectively positioned on the origin of the coordinate system, with I1For the piece image for collecting, I2Collect
As a example by second sub-picture, then in I1The middle vertical coordinate for retaining pixel is less than I1In splice point vertical coordinate pixel be located
Region, in I2Region of the middle vertical coordinate for retaining pixel more than the pixel place of the vertical coordinate of splice point, by this two
Spliced so that being located at I subregion1And I2In splice point overlap.Due to I1And I2Between there is gray difference, therefore exist
To I1And I2After being spliced, can there is obvious gray difference in the image above and below stitching portion, thus need to be to splicing after
Image merged.Specifically, splice point is crossed in spliced image and do the straight line parallel with X-axis, be called I1And I2It
Between piece, in the top of the piece (for I1) and lower section (for I2) N number of pixel is respectively taken, each picture in 2N pixel
The gray value g of vegetarian refreshments is obtained by equation below:
G=a1×I1(x1,y1)+a2×I2(x2,y2)
Wherein, a1+a2=1, I1(x1,y1) for the pixel in I1In gray value, I2(x2,y2) for the pixel in I2
In gray value.a1And a2For weight, when every string pixel position from being located at I1To piece is moved closer to, it is then within
Piece, then from away from piece to be located at I2In, weight a in above-mentioned formula10 is gradually become by 1, weight a21 is gradually become by zero,
Cause the transition in the region near piece smoother by average weighted mode so that spliced image is more accorded with
Close actual clinical demand.
In practical application, to I1And I2After completing splicing, (part does not have to produce therebetween the relative part for offseting
Image) can be filled using black, or corresponding using carrying out to the gray value of the background area that the Offset portion is close to
Filling.
So far, achieved by above-mentioned step and the characteristic point in described first image and second image is examined
Survey, and which is mated, generate matching double points, described first image and second figure is finally determined according to matching double points
The overlapping region of picture, and then complete the splicing to described first image and second image.Complexity is employed in the present embodiment
Degree is low, and amount of calculation is little and fireballing mode determines the high matching double points of accuracy, and then is spelled based on the matching double points
The splicing high precision of the image obtained after connecing.
Corresponding above-mentioned image acquiring method, the present embodiment also provide a kind of image acquiring device, and described image obtains dress
Put including:
Resolving cell, for carrying out pyramid decomposition to the first image and the second image, based on decomposition after the first image
With the second image construct respectively the first difference pyramid diagram picture corresponding with described first image and with second image pair
The the second difference pyramid diagram picture that answers;Described first image and second image are adjacent image;
Search unit, for traveling through each layer of the first difference pyramid diagram picture, searches for every layer of difference pyramid diagram
Extreme point as in, travels through each layer of the second difference pyramid diagram picture, the pole that searches in every layer of difference pyramid diagram picture
Value point;The extreme point is associated with the absolute value of the gray value of the pixel in default neighborhood;
First removal unit is strong for described first image in the extreme point of the first difference pyramid diagram picture for removing
Pixel on edge is obtaining the characteristic point in described first image;Remove the extreme point of the second difference pyramid diagram picture
In be pixel in the second image strong edge obtaining the characteristic point in second image;
Matching unit, initial to generate for being mated to the characteristic point in described first image and second image
Matching double points;
Second removal unit, for removing the matching double points of the initial matching point centering mistake to obtain matching double points;
Offset-determining unit, for determining described first image and described based on the position relationship between the matching double points
Skew between second image;
Overlapping region determining unit, described in determining according to the skew between described first image and second image
First image and the overlapping region of second image;
Concatenation unit, for splicing to described first image and second image according to the overlapping region.
The enforcement for being embodied as referring to described image acquisition methods of described image acquisition device, will not be described here.
The present embodiment also provides a kind of X-ray shooting system, including above-mentioned image acquiring device.
In sum, the detection method of characteristic point in the image that embodiment of the present invention is provided, at least has beneficial as follows
Effect:
Pyramid decomposition is carried out to described image first, based on decomposition after image configuration difference pyramid diagram picture;Time then
Each layer of the difference pyramid diagram picture is gone through, the extreme point that searches in every layer of difference pyramid diagram picture, the extreme point association
The absolute value of the gray value of the pixel in default neighborhood;Finally remove in the extreme point as in described image strong edge
Pixel is obtaining the characteristic point in described image.Per layer is searched for due to having traveled through each layer of the difference pyramid diagram picture
Extreme point in difference pyramid diagram picture, determines the side of extreme point relative to existing with difference pyramid diagram picture place space
For method, the complexity of detection is low, and then improves the speed of detection characteristic point to a certain extent, due to eliminating extreme point
In for the pixel in described image strong edge, therefore stable characteristic point can be obtained.
Although the present invention is disclosed as above with preferred embodiment, which is not any this area for limiting the present invention
Technical staff without departing from the spirit and scope of the present invention, may be by the methods and techniques content of the disclosure above to this
Bright technical scheme makes possible variation and modification, and therefore, every content without departing from technical solution of the present invention, according to the present invention
Technical spirit any simple modification, equivalent variations and modification that above example is made, belong to technical solution of the present invention
Protection domain.
Claims (10)
1. a kind of determine matching double points method, it is characterised in that include:
Pyramid decomposition is carried out to the first image and the second image, based on decomposition after the first image and the second image construct respectively
The first difference pyramid diagram picture corresponding with described first image and the second difference pyramid corresponding with second image
Image;Described first image and second image are adjacent image;
Each layer of the first difference pyramid diagram picture is traveled through, the extreme point that searches in every layer of difference pyramid diagram picture, traversal
Each layer of the second difference pyramid diagram picture, the extreme point that searches in every layer of difference pyramid diagram picture;The extreme point is closed
The absolute value of the gray value of the pixel being coupled in default neighborhood;
Remove is the pixel in described first image strong edge in the extreme point of the first difference pyramid diagram picture obtaining
Characteristic point in described first image;It is the strong side of second image in the extreme point for removing the second difference pyramid diagram picture
Pixel on edge is obtaining the characteristic point in second image;
The characteristic vector of the characteristic point in the characteristic vector of the characteristic point in based on described first image and second image is right
Characteristic point in described first image and second image carries out bi-directional matching to generate initial matching point pair;
Remove the matching double points of the initial matching point centering mistake to obtain matching double points.
2. as claimed in claim 1 determine matching double points method, it is characterised in that the first difference pyramid diagram as
Extreme point refers to that the absolute value of the gray value of pixel in every layer of difference pyramid diagram picture is picture in the default neighborhood of the pixel
The pixel of the maximum absolute value of the gray value of vegetarian refreshments;The extreme point of the second difference pyramid diagram picture refers to every layer of difference gold
In word tower image, the absolute value of the gray value of pixel is the absolute value of the gray value of pixel in the default neighborhood of the pixel
Maximum pixel.
3. as claimed in claim 1 determine matching double points method, it is characterised in that the first difference pyramid diagram as
Extreme point refers to that revised pixel P in the first difference pyramid diagram picture, revised pixel P are to pixel P
Position be modified after obtain, the pixel P refers to pixel in every tomographic image of the first difference pyramid diagram picture
The absolute value of the gray value of point is the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of the pixel;Institute
The extreme point for stating the second difference pyramid diagram picture refers to revised pixel P in the second difference pyramid diagram picture, described
Revised pixel P is obtained after the position to pixel P is modified, and pixel P refers to the second difference gold word
In every tomographic image of tower image, the absolute value of the gray value of pixel is the gray value of pixel in the default neighborhood of the pixel
Maximum absolute value pixel.
4. as claimed in claim 1 determine matching double points method, it is characterised in that the first difference pyramid diagram as
Extreme point refers to the absolute value of the gray value to pixel P according to the pixel for being located at top N after order sequence from large to small
P, the pixel P refer to that the absolute value of the gray value of pixel in every tomographic image of the first difference pyramid diagram picture is institute
State the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of pixel;The second difference pyramid diagram as
Extreme point refers to the absolute value of the gray value to pixel P according to the pixel for being located at top N after order sequence from large to small
P, the pixel P refer to that the absolute value of the gray value of pixel in every tomographic image of the second difference pyramid diagram picture is institute
State the pixel of the maximum absolute value of the gray value of pixel in the default neighborhood of pixel.
5. as claimed in claim 1 determine matching double points method, it is characterised in that described based on described first image in
The characteristic vector of the characteristic point in the characteristic vector of characteristic point and second image, to described first image and second figure
Characteristic point as in carries out bi-directional matching to generate initial matching point to including:
With the feature of the characteristic point in the characteristic vector of each characteristic point in described first image and second image to
Minimum euclidean distance between amount is the first ratio with the ratio of secondary little Euclidean distance;
When first ratio is less than first threshold, with the spy in described first image corresponding with the minimum euclidean distance
It is the 3rd matching double points to levy the characteristic point a little and in second image, generates the first collection with the 3rd matching double points as element
Close;
With the feature of the characteristic point in the characteristic vector of each characteristic point in second image and described first image to
Minimum euclidean distance between amount is the second ratio with the ratio of secondary little Euclidean distance;
When second ratio is less than Second Threshold, with the spy in second image corresponding with the minimum euclidean distance
It is the 4th matching double points to levy the characteristic point a little and in described first image, generates the second collection with the 4th matching double points as element
Close;
Take the common factor of the first set and the second set to obtain the initial of described first image and second image
Matching double points.
6. as claimed in claim 5 determine matching double points method, it is characterised in that
The characteristic vector of the characteristic point in the characteristic vector of each characteristic point in described first image and second image
Between minimum euclidean distance refer to the ratio of secondary little Euclidean distance:The feature of each characteristic point in described first image to
Amount with second image with the abscissa of this feature point apart from its poor absolute value less than the first predeterminable range and with this
The minimum euclidean distance that characteristic point belongs between the characteristic vector of the characteristic point of same difference pyramid image layer is little European with secondary
The ratio of distance;
The characteristic vector of the characteristic point in the characteristic vector of each characteristic point in second image and described first image
Between minimum euclidean distance refer to the ratio of secondary little Euclidean distance:The feature of each characteristic point in second image to
Amount with described first image with the abscissa of this feature point apart from its poor absolute value less than the second predeterminable range and with this
The minimum euclidean distance that characteristic point belongs between the characteristic vector of the characteristic point of same difference pyramid image layer is little European with secondary
The ratio of distance.
7. as claimed in claim 5 determine matching double points method, it is characterised in that
The characteristic vector of the characteristic point in the characteristic vector of each characteristic point in described first image and second image
Between minimum euclidean distance refer to the ratio of secondary little Euclidean distance:The feature of each characteristic point in described first image to
Amount with second image with the vertical coordinate of this feature point apart from its poor absolute value less than the 3rd predeterminable range and with this
The minimum euclidean distance that characteristic point belongs between the characteristic vector of the characteristic point of same difference pyramid image layer is little European with secondary
The ratio of distance;
The characteristic vector of the characteristic point in the characteristic vector of each characteristic point in second image and described first image
Between minimum euclidean distance refer to the ratio of secondary little Euclidean distance:The feature of each characteristic point in second image to
Amount with described first image with the vertical coordinate of this feature point apart from its poor absolute value less than the 4th predeterminable range and with this
The minimum euclidean distance that characteristic point belongs between the characteristic vector of the characteristic point of same difference pyramid image layer is little European with secondary
The ratio of distance.
8. the method for determining matching double points as claimed in claim 1, it is characterised in that the removal initial matching point pair
The matching double points of middle mistake are included with obtaining matching double points:
Slope of the initial matching point of described first image and second image to place line is calculated, with the initial matching
Point is abscissa to the slope of place line, and initial matching point corresponding with the slope is that vertical coordinate generation slope is straight to number
Fang Tu;
Determine initial matching point in the slope histogram to the initial matching point of number sum place cluster when maximum to for first
Matching double points;
Difference with the first coordinate of first matching double points as abscissa, the first matching double points corresponding with the difference of the first coordinate
Number is that vertical coordinate generates the first rectangular histogram;
When determining that the first matching double points number sum is maximum in first rectangular histogram, the first matching double points of place cluster are second
Matching double points;
Difference with the second coordinate of second matching double points as abscissa, the second matching double points corresponding with the difference of the second coordinate
Number is that vertical coordinate generates the second rectangular histogram;
When determining that the second matching double points number sum is maximum in second rectangular histogram, the second matching double points of place cluster are coupling
Point is right.
9. as claimed in claim 8 determine matching double points method, it is characterised in that calculate described first in the following way
Slope of the initial matching point of image and second image to place line:
Wherein:KiFor slope, (x of the i-th pair initial matching point to place line1i,y1i)、(x2i,y2i) it is i-th pair initial matching
Point is to, (x1i,y1i) for i-th initial matching point in described first image position, (x2i,y2i) in second image
The position of i-th initial matching point, W1Width for described first image.
10. a kind of image acquiring method, it is characterised in that include:
Using the method for the determination matching double points described in any one of claim 1~9 determine the first image and the second image
Right with;
Skew between described first image and second image is determined based on the position relationship between the matching double points;
Described first image and second image are determined according to the skew between described first image and second image
Overlapping region;
Described first image and second image are spliced according to the overlapping region.
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610916842.0A CN106504237A (en) | 2016-09-30 | 2016-09-30 | Determine method and the image acquiring method of matching double points |
US15/649,819 US10580135B2 (en) | 2016-07-14 | 2017-07-14 | System and method for splicing images |
US16/806,207 US11416993B2 (en) | 2016-07-14 | 2020-03-02 | System and method for splicing images |
US17/819,925 US11893738B2 (en) | 2016-07-14 | 2022-08-15 | System and method for splicing images |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610916842.0A CN106504237A (en) | 2016-09-30 | 2016-09-30 | Determine method and the image acquiring method of matching double points |
CN201610877441.9A CN106504229B (en) | 2016-09-30 | 2016-09-30 | The detection method of characteristic point in image |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610877441.9A Division CN106504229B (en) | 2016-07-14 | 2016-09-30 | The detection method of characteristic point in image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106504237A true CN106504237A (en) | 2017-03-15 |
Family
ID=58294764
Family Applications (3)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610916844.XA Pending CN106548492A (en) | 2016-07-14 | 2016-09-30 | Determine method and device, the image acquiring method of matching double points |
CN201610877441.9A Active CN106504229B (en) | 2016-07-14 | 2016-09-30 | The detection method of characteristic point in image |
CN201610916842.0A Pending CN106504237A (en) | 2016-07-14 | 2016-09-30 | Determine method and the image acquiring method of matching double points |
Family Applications Before (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610916844.XA Pending CN106548492A (en) | 2016-07-14 | 2016-09-30 | Determine method and device, the image acquiring method of matching double points |
CN201610877441.9A Active CN106504229B (en) | 2016-07-14 | 2016-09-30 | The detection method of characteristic point in image |
Country Status (1)
Country | Link |
---|---|
CN (3) | CN106548492A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109242769A (en) * | 2018-12-13 | 2019-01-18 | 腾讯科技(深圳)有限公司 | A kind of image processing method and device |
CN111898396A (en) * | 2019-05-06 | 2020-11-06 | 北京四维图新科技股份有限公司 | Obstacle detection method and device |
CN112837225A (en) * | 2021-04-15 | 2021-05-25 | 浙江卡易智慧医疗科技有限公司 | Method and device for automatically and seamlessly splicing vertical full-spine images |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019041265A1 (en) * | 2017-08-31 | 2019-03-07 | 深圳市大疆创新科技有限公司 | Feature extraction circuit and integrated image processing circuit |
CN109508571B (en) * | 2017-09-14 | 2021-08-27 | 杭州海康威视数字技术股份有限公司 | Strip-space positioning method and device, electronic equipment and storage medium |
CN107609562B (en) * | 2017-09-14 | 2020-09-04 | 电子科技大学 | Scale space feature detection method based on SIFT algorithm |
CN108573269B (en) * | 2017-10-24 | 2021-02-05 | 北京金山云网络技术有限公司 | Image feature point matching method, matching device, electronic device and storage medium |
CN107918938A (en) * | 2017-11-16 | 2018-04-17 | 凌云光技术集团有限责任公司 | A kind of matching process and device of point set and point set |
CN108960251A (en) * | 2018-05-22 | 2018-12-07 | 东南大学 | A kind of images match description generates the hardware circuit implementation method of scale space |
CN109431511B (en) * | 2018-11-14 | 2021-09-24 | 南京航空航天大学 | Human back scoliosis spine contour characteristic curve fitting method based on digital image processing |
CN110335297B (en) * | 2019-06-21 | 2021-10-08 | 华中科技大学 | Point cloud registration method based on feature extraction |
CN111311626A (en) * | 2020-05-11 | 2020-06-19 | 南京安科医疗科技有限公司 | Skull fracture automatic detection method based on CT image and electronic medium |
CN113421249B (en) * | 2021-06-30 | 2024-02-06 | 上海申瑞继保电气有限公司 | Variable-scale image numerical value processing method for substation equipment |
CN114913316B (en) * | 2022-04-02 | 2023-04-07 | 淮沪电力有限公司田集第二发电厂 | Image classification method and device for meter recognition of industrial equipment, electronic equipment and storage medium |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510257A (en) * | 2009-03-31 | 2009-08-19 | 华为技术有限公司 | Human face similarity degree matching method and device |
CN101770568A (en) * | 2008-12-31 | 2010-07-07 | 南京理工大学 | Target automatically recognizing and tracking method based on affine invariant point and optical flow calculation |
CN103593832A (en) * | 2013-09-25 | 2014-02-19 | 重庆邮电大学 | Method for image mosaic based on feature detection operator of second order difference of Gaussian |
CN103729654A (en) * | 2014-01-22 | 2014-04-16 | 青岛新比特电子科技有限公司 | Image matching retrieval system on account of improving Scale Invariant Feature Transform (SIFT) algorithm |
CN103886611A (en) * | 2014-04-08 | 2014-06-25 | 西安煤航信息产业有限公司 | Image matching method suitable for automatically detecting flight quality of aerial photography |
CN104376552A (en) * | 2014-09-19 | 2015-02-25 | 四川大学 | Virtual-real registering algorithm of 3D model and two-dimensional image |
CN104515502A (en) * | 2013-09-28 | 2015-04-15 | 沈阳新松机器人自动化股份有限公司 | Robot hand-eye stereo vision measurement method |
CN104616297A (en) * | 2015-01-26 | 2015-05-13 | 山东省计算中心(国家超级计算济南中心) | Improved SIFI algorithm for image tampering forensics |
CN104717456A (en) * | 2013-12-11 | 2015-06-17 | 杭州海康威视数字技术股份有限公司 | Method and device for processing mobile detection |
CN104717465A (en) * | 2014-12-23 | 2015-06-17 | 北京亿羽舜海科技有限公司 | Wide-angle video monitoring device and method for multiple IP cameras |
CN105447864A (en) * | 2015-11-20 | 2016-03-30 | 小米科技有限责任公司 | Image processing method, device and terminal |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101206213B1 (en) * | 2010-04-19 | 2012-11-28 | 인하대학교 산학협력단 | High speed slam system and method based on graphic processing unit |
CN103761721B (en) * | 2013-12-30 | 2016-10-05 | 西北工业大学 | One is applicable to space rope system machine human stereo vision fast image splicing method |
CN104850851A (en) * | 2015-04-22 | 2015-08-19 | 福州大学 | ORB feature point matching method with scale invariance |
CN104820965A (en) * | 2015-04-30 | 2015-08-05 | 武汉大学 | Geocoding-free rapid image splicing method of low-altitude unmanned plane |
CN104881671B (en) * | 2015-05-21 | 2018-01-19 | 电子科技大学 | A kind of high score remote sensing image Local Feature Extraction based on 2D Gabor |
CN104881877A (en) * | 2015-06-12 | 2015-09-02 | 哈尔滨工业大学 | Method for detecting image key point based on convolution and time sequence optimization of FPGA |
-
2016
- 2016-09-30 CN CN201610916844.XA patent/CN106548492A/en active Pending
- 2016-09-30 CN CN201610877441.9A patent/CN106504229B/en active Active
- 2016-09-30 CN CN201610916842.0A patent/CN106504237A/en active Pending
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101770568A (en) * | 2008-12-31 | 2010-07-07 | 南京理工大学 | Target automatically recognizing and tracking method based on affine invariant point and optical flow calculation |
CN101510257A (en) * | 2009-03-31 | 2009-08-19 | 华为技术有限公司 | Human face similarity degree matching method and device |
CN103593832A (en) * | 2013-09-25 | 2014-02-19 | 重庆邮电大学 | Method for image mosaic based on feature detection operator of second order difference of Gaussian |
CN104515502A (en) * | 2013-09-28 | 2015-04-15 | 沈阳新松机器人自动化股份有限公司 | Robot hand-eye stereo vision measurement method |
CN104717456A (en) * | 2013-12-11 | 2015-06-17 | 杭州海康威视数字技术股份有限公司 | Method and device for processing mobile detection |
CN103729654A (en) * | 2014-01-22 | 2014-04-16 | 青岛新比特电子科技有限公司 | Image matching retrieval system on account of improving Scale Invariant Feature Transform (SIFT) algorithm |
CN103886611A (en) * | 2014-04-08 | 2014-06-25 | 西安煤航信息产业有限公司 | Image matching method suitable for automatically detecting flight quality of aerial photography |
CN104376552A (en) * | 2014-09-19 | 2015-02-25 | 四川大学 | Virtual-real registering algorithm of 3D model and two-dimensional image |
CN104717465A (en) * | 2014-12-23 | 2015-06-17 | 北京亿羽舜海科技有限公司 | Wide-angle video monitoring device and method for multiple IP cameras |
CN104616297A (en) * | 2015-01-26 | 2015-05-13 | 山东省计算中心(国家超级计算济南中心) | Improved SIFI algorithm for image tampering forensics |
CN105447864A (en) * | 2015-11-20 | 2016-03-30 | 小米科技有限责任公司 | Image processing method, device and terminal |
Non-Patent Citations (4)
Title |
---|
仵建宁等: "一种基于兴趣点匹配的图像拼接方法", 《计算机应用》 * |
曹新华: "基于特征点驱动的CT图像配准与拼接方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
肖南峰: "《服务机器人》", 31 January 2013, 清华大学出版社 * |
韩晓微等: "《数字图像融合》", 31 December 2010, 东北大学出版社 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109242769A (en) * | 2018-12-13 | 2019-01-18 | 腾讯科技(深圳)有限公司 | A kind of image processing method and device |
CN109242769B (en) * | 2018-12-13 | 2019-03-19 | 腾讯科技(深圳)有限公司 | A kind of image processing method and device |
CN111898396A (en) * | 2019-05-06 | 2020-11-06 | 北京四维图新科技股份有限公司 | Obstacle detection method and device |
CN112837225A (en) * | 2021-04-15 | 2021-05-25 | 浙江卡易智慧医疗科技有限公司 | Method and device for automatically and seamlessly splicing vertical full-spine images |
CN112837225B (en) * | 2021-04-15 | 2024-01-23 | 浙江卡易智慧医疗科技有限公司 | Automatic seamless splicing method and device for standing full-spine images |
Also Published As
Publication number | Publication date |
---|---|
CN106504229B (en) | 2019-11-12 |
CN106504229A (en) | 2017-03-15 |
CN106548492A (en) | 2017-03-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106504237A (en) | Determine method and the image acquiring method of matching double points | |
CN109754387B (en) | Intelligent detection and positioning method for whole-body bone imaging radioactive concentration focus | |
CN102737395B (en) | Image processing method and device in a kind of medical X-ray system | |
CN106023200B (en) | A kind of x-ray chest radiograph image rib cage suppressing method based on Poisson model | |
CN110111313A (en) | Medical image detection method and relevant device based on deep learning | |
CN109938764A (en) | A kind of adaptive multiple location scan imaging method and its system based on deep learning | |
CN103871036B (en) | Rapid registering and joining method for three dimensional digital subtraction angiography image | |
CN106408609A (en) | Parallel mechanism end motion pose detection method based on binocular vision | |
CN102999917B (en) | Cervical cancer image automatic segmentation method based on T2-MRI and DW-MRI | |
CN101258524A (en) | Adaptive radiation therapy method with target detection | |
WO2021073120A1 (en) | Method and device for marking lung area shadows in medical image, server, and storage medium | |
CN113139962B (en) | System and method for scoliosis probability assessment | |
CN106447664A (en) | Matching pair determination method and image capturing method | |
CN113034389A (en) | Image processing method, image processing device, computer equipment and storage medium | |
CN104933719B (en) | One kind integration segment spacing is from detection image edge method | |
CN105405101B (en) | Radioscopic image treating method and apparatus | |
US20160210740A1 (en) | Method and system for spine position detection | |
CN105193436B (en) | Method for evaluating exposure dose of digital mammary X-ray imaging system | |
CN108416735B (en) | Method and device for splicing digital X-ray images based on geometric features | |
CN104715478B (en) | The method and system of exposure area in a kind of detection image picture | |
CN103366348A (en) | Processing method and processing device for restraining bone image in X-ray image | |
Yao et al. | Employing topographical height map in colonic polyp measurement and false positive reduction | |
CN111402231A (en) | Automatic evaluation system and method for lung CT image quality | |
CN106097306A (en) | Obtain the method for Image Edge-Detection operator, method for detecting image edge and device | |
CN106600575A (en) | Spliced image verifying method and device, and X-ray photographing system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information |
Address after: 201807 Shanghai city Jiading District Industrial Zone Jiading Road No. 2258 Applicant after: Shanghai Lianying Medical Technology Co., Ltd Address before: 201807 Shanghai city Jiading District Industrial Zone Jiading Road No. 2258 Applicant before: SHANGHAI UNITED IMAGING HEALTHCARE Co.,Ltd. |
|
CB02 | Change of applicant information | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170315 |
|
RJ01 | Rejection of invention patent application after publication |