CN110111372A - Medical figure registration and fusion method based on SIFT+RANSAC algorithm - Google Patents
Medical figure registration and fusion method based on SIFT+RANSAC algorithm Download PDFInfo
- Publication number
- CN110111372A CN110111372A CN201910303778.2A CN201910303778A CN110111372A CN 110111372 A CN110111372 A CN 110111372A CN 201910303778 A CN201910303778 A CN 201910303778A CN 110111372 A CN110111372 A CN 110111372A
- Authority
- CN
- China
- Prior art keywords
- image
- point
- registration
- sift
- medical
- 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
- 238000007500 overflow downdraw method Methods 0.000 title claims abstract description 19
- 239000011159 matrix material Substances 0.000 claims abstract description 33
- 239000013598 vector Substances 0.000 claims abstract description 16
- 230000000694 effects Effects 0.000 claims abstract description 14
- 238000001514 detection method Methods 0.000 claims abstract description 10
- 238000005070 sampling Methods 0.000 claims abstract 2
- 238000012417 linear regression Methods 0.000 claims description 6
- 238000012986 modification Methods 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 230000035772 mutation Effects 0.000 claims description 3
- 238000012887 quadratic function Methods 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 238000005259 measurement Methods 0.000 claims 1
- 230000004927 fusion Effects 0.000 abstract description 8
- 238000012545 processing Methods 0.000 abstract description 2
- 238000011160 research Methods 0.000 abstract description 2
- 230000009466 transformation Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 15
- 238000002474 experimental method Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000000034 method Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 4
- 230000004256 retinal image Effects 0.000 description 4
- 239000003814 drug Substances 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 2
- 210000005252 bulbus oculi Anatomy 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- 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
- G06T2207/10004—Still image; Photographic 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
- G06T2207/20212—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
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to medical figure registrations and fusion method based on SIFT+RANSAC algorithm, belong to technical field of medical image processing.The invention includes the following steps: building Gaussian scale-space；The detection of Gaussian scale-space characteristic point and accurate positioning；Remove unstable point；Point including low contrast and the point positioned at image border；Calculate characteristic point direction；Feature point description for constructing image, forms feature vector；Image registration thus obtains the effect picture of a width clearly image registration；RANSAC algorithm finds out homography matrix, does linear weighted function according to homography matrix, obtains the connection between two images by linear weighting function, is finally piece image by two image co-registrations；Modify the image merged.The present invention can effectively realize the registration of image, then calculate projective transformation matrix by RANSAC random sampling unification algorism to realize the fusion of image, lay the groundwork for subsequent some research work.
Description
Technical field
The present invention relates to medical figure registrations and fusion method based on SIFT+RANSAC algorithm, belong at medical image
Manage technical field.
Background technique
An important technology of the image registration as image information field, computer vision, Medical Image Processing and
The fields such as the mechanics of materials have a wide range of applications.In recent years, the method for registering images based on feature extraction achieves quick hair
Exhibition.Relative to the image registration algorithm based on region, the method for registering images based on feature has more powerful separating capacity, energy
It is enough to realize registration in image aspects variation, there are near real-time in the case that noise even centainly blocks.Image registration is image
A typical problem and technological difficulties in treatment research field, its object is to compare or merge for same target in difference
Under the conditions of the image that obtains, such as image can be derived from different time, different shooting visual angles etc. from different acquisition equipment
Deng being also required to use the image registration problem for different objects sometimes.Specifically, for two in one group of image data set
Piece image is mapped to another piece image by finding a kind of spatial alternation by width image, so that it is same to correspond to space in two figures
The point of one position corresponds, to achieve the purpose that information merges.
Image co-registration is by the imaging sensor from different type or same type by a mathematical model not
The source images generated with time or same time combine, the process of the comprehensive image for forming a width application-specific demand.
Image fusion technology combines various images, shows respective information on the same image, provides for clinical medicine diagnosis
For majority according to the image of multi information, this becomes the technology of great application value, and the image registration of precise and high efficiency be then it is crucial with
Difficult point.Thus image registration and integration technology are no matter in terms of computer vision, or all have in clinical medicine diagnosis and extremely weigh
The meaning wanted.
Pass through the retrieval discovery to existing patent and the relevant technologies, the relevant technologies of registration and fusion for medical image
It is not mature enough, still remain many problems: the overlong time of feature extracting and matching is easy to appear match point redundancy, image
Fusion is there are gap or generates phenomena such as fuzzy.
Summary of the invention
The present invention provides medical figure registrations and fusion method based on SIFT+RANSAC algorithm, for solving doctor
The following technical problem learning image registration and merging: the 1) extraction of characteristic point and matched overlong time；2) matched school
Direct problem；3) fusion of image is easy to produce gap or fuzzy.
The technical scheme is that medical figure registration and fusion method based on SIFT+RANSAC algorithm, the base
In the medical figure registration and fusion method of SIFT+RANSAC algorithm, specific step is as follows:
Step1, building Gaussian scale-space；
Step2, the detection of Gaussian scale-space characteristic point and accurate positioning；
The unstable point of Step3, removal；Point including low contrast and the point positioned at image border；
Step4, characteristic point direction is calculated；
Step5, feature point description for constructing image, form feature vector；
Step6, image registration, i.e., connect the match point of two images, thus obtains a width clearly image
The effect picture of registration；
Step7, RANSAC algorithm find out homography matrix, do linear weighted function according to homography matrix, pass through linear weighting function
The connection between two images is obtained, is finally piece image by two image co-registrations；
The image that Step8, modification were merged.
Further, in the Step1, the Gaussian scale-space of image is defined as function L (x, y, σ), by changeable ruler
Gaussian convolution G (x, y, σ) is spent to generate；Variable dimension Gaussian convolution isWherein (x, y)
It is space coordinate, σ is scale coordinate, and the size of σ determines the smoothness of image, and the general picture feature of large scale correspondence image is small
The minutia of scale correspondence image, output image are I (x, y), i.e. L (x, y, σ)=G (x, y, σ) * I (x, y).
Further, in the Step2, the detection of Gaussian scale-space characteristic point and accurate positioning: a characteristic point is high
Local Extremum in this scale space；All scales and picture position are searched in Gaussian scale-space, pass through difference of Gaussian letter
There are the potential characteristic points of scale and direction invariance for number identification；
Since the first floor and last layer all lack an adjacent layer, searches for since every group of the second layer, be with the second layer
Current layer, first layer and third layer are done using third layer as current layer after the completion of search again respectively as the upper and lower level of cube
Same search；So every layer of point search is twice；Pass through difference of Gaussian function 26 pictures that current pixel is adjacent with other
If it is maximum value or minimum value that element, which is compared it, that is, select the point as there are the potential spies of scale and direction invariance
Sign point；Then by being fitted three-dimensional quadratic function accurately to determine key point position and scale.
Further, in the Step3, unstable point is removed；Point including low contrast and positioned at image border
Point: both unstable points are removed by setting contrast threshold and Hessian matrix.
Further, it in the Step4, calculates characteristic point direction: utilizing the gradient direction distribution of key point neighborhood territory pixel
Characteristic is each characteristic point assigned direction parameter, so that operator is had rotational invariance, to realize image rotation invariance；In reality
It is sampled when border calculates, in the neighborhood window centered on characteristic point, and with the gradient direction of statistics with histogram neighborhood territory pixel；Gradient
The range of histogram is 0-360 °, histogram is divided into 8 directions, i.e., each characteristic point has 8 Gradient direction informations；Histogram
The peak value of figure represents the principal direction of neighborhood gradient at this feature point, the i.e. direction as this feature point；It is got over away from central point
Remote field its to the contribution of histogram also responsive to reduction；Carry out smooth, reduction mutation to histogram using Gaussian function simultaneously
Influence, in gradient orientation histogram, when there are the peak value that another is equivalent to 80% energy of main peak value, then by this
Direction is considered the auxiliary direction of this feature point；One characteristic point may be designated with multiple directions, a principal direction, and one
The above auxiliary direction, for enhancing matched robustness.
Further, in the Step5, description of image characteristic point is constructed, forms feature vector: for any one
Key point, i.e. characteristic point, scale space where it take 16 pixels centered on key point × 16 pixel sizes neighborhood, then
This neighborhood region is uniformly divided into 4 × 4 sub-regions, each key point there are 8 gradient directions therefore to share 4 × 4 × 8=
128 data, i.e., 128 dimension SIFT feature vectors.
Further, in the Step6, after the SIFT feature vector of two images generates, two width image registration: are calculated
The Euclidean distance of characteristic point feature vector is measured as similarity determination on image, will be apart from the smallest characteristic point as initial
With point, and being less than some proportion threshold value T according to the ratio between Euclidean distance of arest neighbors and time neighbour is 0.8, is determined as a pair of of matching
Point, and remove error matching points；The match point in reference picture and image subject to registration is connected with line again realizes image
Registration.Selection about proportion threshold value: if reducing this threshold value, SIFT match point number can be reduced, but more stable；If
This threshold value is improved, SIFT match point number will increase, and corresponding mispairing point number will increase.By testing eyeball figure to this
It is relatively stable finally to choose a match point number for the matching test of picture, and accuracy is relatively high as a result, its is corresponding
Threshold value T=0.8 is determined as required proportion threshold value.
Further, in the Step7, RANSAC algorithm finds out homography matrix: from image subject to registration to reference picture, appointing
It takes 4 pairs of points to obtain homography matrix H, the characteristic point in image subject to registration is then projected into reference picture, detection projection with matrix H
The distance between accurate characteristic point in obtained point and reference picture: a threshold value: image subject to registration and reference is set first
Characteristic point in image is to meet a linear regression relation, the match point of mistake to linear regression line by homography matrix H
Sum of the distance minimum value is fixed threshold value；It is judged as interior point if being less than the threshold value that this sets；Put in statistics
Number, if more than the threshold value that this sets, is then judged as available homography matrix, then recalculates list using all interior points
Answer matrix；Linear weighted function is done according to homography matrix, the connection between two images is obtained by linear weighting function, finally by two
A image co-registration is piece image.
Further, in the Step8, the image merged is modified: in order to seem the syncretizing effect of two images
More natural, with the color of corresponding interior point surrounding or luminance information carries out brightness to wherein piece image or color adjusts, after adjustment
Image be final result.
The beneficial effects of the present invention are:
The present invention can quickly and effectively realize a whole set of stream of the feature extraction of medical image, matching, fusion, correction
Journey, characteristics of image not only has scale invariability, illumination invariant shape, rotational invariance, but also the time consumed is seldom, emulation
The efficiency of experiment is very high.
Detailed description of the invention
Fig. 1 is the effect picture that reference picture and the match point in image subject to registration are connected in the present invention；
Fig. 2 is the effect picture after image co-registration of the present invention；
Fig. 3 is the final result after present invention modification blending image；
Fig. 4 is the flow chart in the present invention.
Specific embodiment
Embodiment 1: as shown in Figs 1-4, medical figure registration and fusion method based on SIFT+RANSAC algorithm are described
Specific step is as follows for medical figure registration and fusion method based on SIFT+RANSAC algorithm:
Step1, building Gaussian scale-space；
Step2, the detection of Gaussian scale-space characteristic point and accurate positioning；
The unstable point of Step3, removal；Point including low contrast and the point positioned at image border；
Step4, characteristic point direction is calculated；
Step5, feature point description for constructing image, form feature vector；
Step6, image registration, i.e., connect the match point of two images, thus obtains a width clearly image
The effect picture of registration；
Step7, RANSAC algorithm find out homography matrix, do linear weighted function according to homography matrix, pass through linear weighting function
The connection between two images is obtained, is finally piece image by two image co-registrations；
The image that Step8, modification were merged.
Further, in the Step1, the Gaussian scale-space of image is defined as function L (x, y, σ), by changeable ruler
Gaussian convolution G (x, y, σ) is spent to generate；Variable dimension Gaussian convolution isWherein (x, y)
It is space coordinate, σ is scale coordinate, and the size of σ determines the smoothness of image, and the general picture feature of large scale correspondence image is small
The minutia of scale correspondence image, output image are I (x, y), i.e. L (x, y, σ)=G (x, y, σ) * I (x, y).
Further, in the Step2, the detection of Gaussian scale-space characteristic point and accurate positioning: a characteristic point is high
Local Extremum in this scale space；All scales and picture position are searched in Gaussian scale-space, pass through difference of Gaussian letter
There are the potential characteristic points of scale and direction invariance for number identification；
Since the first floor and last layer all lack an adjacent layer, searches for since every group of the second layer, be with the second layer
Current layer, first layer and third layer are done using third layer as current layer after the completion of search again respectively as the upper and lower level of cube
Same search；So every layer of point search is twice；Pass through difference of Gaussian function 26 pictures that current pixel is adjacent with other
If it is maximum value or minimum value that element, which is compared it, that is, select the point as there are the potential spies of scale and direction invariance
Sign point；Then by being fitted three-dimensional quadratic function accurately to determine key point position and scale.
Further, in the Step3, unstable point is removed；Point including low contrast and positioned at image border
Point: both unstable points are removed by setting contrast threshold and Hessian matrix.
Further, it in the Step4, calculates characteristic point direction: utilizing the gradient direction distribution of key point neighborhood territory pixel
Characteristic is each characteristic point assigned direction parameter, so that operator is had rotational invariance, to realize image rotation invariance；In reality
It is sampled when border calculates, in the neighborhood window centered on characteristic point, and with the gradient direction of statistics with histogram neighborhood territory pixel；Gradient
The range of histogram is 0-360 °, histogram is divided into 8 directions, i.e., each characteristic point has 8 Gradient direction informations；Histogram
The peak value of figure represents the principal direction of neighborhood gradient at this feature point, the i.e. direction as this feature point；It is got over away from central point
Remote field its to the contribution of histogram also responsive to reduction；Carry out smooth, reduction mutation to histogram using Gaussian function simultaneously
Influence, in gradient orientation histogram, when there are the peak value that another is equivalent to 80% energy of main peak value, then by this
Direction is considered the auxiliary direction of this feature point；One characteristic point may be designated with multiple directions, a principal direction, and one
The above auxiliary direction, for enhancing matched robustness.
Further, in the Step5, description of image characteristic point is constructed, forms feature vector: for any one
Key point, i.e. characteristic point, scale space where it take 16 pixels centered on key point × 16 pixel sizes neighborhood, then
This neighborhood region is uniformly divided into 4 × 4 sub-regions, each key point there are 8 gradient directions therefore to share 4 × 4 × 8=
128 data, i.e., 128 dimension SIFT feature vectors.
Further, in the Step6, after the SIFT feature vector of two images generates, two width image registration: are calculated
The Euclidean distance of characteristic point feature vector is measured as similarity determination on image, will be apart from the smallest characteristic point as initial
With point, and being less than some proportion threshold value T according to the ratio between Euclidean distance of arest neighbors and time neighbour is 0.8, is determined as a pair of of matching
Point, and remove error matching points；The match point in reference picture and image subject to registration is connected with line again realizes image
Registration, effect picture such as Fig. 1.Selection about proportion threshold value: if reducing this threshold value, SIFT match point number can be reduced, but
It is more stable；If improving this threshold value, SIFT match point number be will increase, and corresponding mispairing point number will increase.By right
It is relatively stable finally to choose a match point number for the matching test of this experiment eyeball image, and accuracy is relatively high
As a result, its corresponding threshold value T=0.8, is determined as required proportion threshold value.
Further, in the Step7, RANSAC algorithm finds out homography matrix: from image subject to registration to reference picture, appointing
It takes 4 pairs of points to obtain homography matrix H, the characteristic point in image subject to registration is then projected into reference picture, detection projection with matrix H
The distance between accurate characteristic point in obtained point and reference picture: a threshold value: image subject to registration and reference is set first
Characteristic point in image is to meet a linear regression relation, the match point of mistake to linear regression line by homography matrix H
Sum of the distance minimum value is fixed threshold value；It is judged as interior point if being less than the threshold value that this sets；Put in statistics
Number, if more than the threshold value that this sets, is then judged as available homography matrix, then recalculates list using all interior points
Answer matrix；Linear weighted function is done according to homography matrix, the connection between two images is obtained by linear weighting function, finally by two
A image co-registration is piece image, effect picture such as Fig. 2.
Further, in the Step8, the image merged is modified: in order to seem the syncretizing effect of two images
More natural, with the color of corresponding interior point surrounding or luminance information carries out brightness to wherein piece image or color adjusts, after adjustment
Image be final as a result, effect picture such as Fig. 3.
In the present embodiment, main parameter are as follows: one shares 129 retinal images compositions in data set, forms 134 figures
As right.These images are divided into 3 different classifications to according to its feature.It is acquired and is schemed using Nidek AFC-210 fundus camera
Picture, the image resolution ratio of acquisition are 512 × 512 pixels, and field range is 45 °, and image comes from Aristotle university
39 patients of Papageorgiou hospital.Retinal images arrange in pairs, are respectively as follows: colored ROI mask and function ROI is covered
Code.The two is all binary picture.Experiment of the invention is to carry out emulation experiment for certain a pair of of retinal images at random, one
Found in the emulation experiment of series: the treatment effect of every a pair of retinal images almost without difference, the registration of image with merge
Speed and quality it is very good.
The simulation experiment result shows that method of the invention is highly suitable for the registration and fusion experiment of medical image, experiment
Effect is very clear, and can save the time, and average whole flow process probably needs 3-4 seconds, leads in medicine, computer vision etc.
Domain can promote and apply.
The present invention can quickly and effectively realize a whole set of stream of the feature extraction of medical image, matching, fusion, correction
Journey, characteristics of image not only has scale invariability, illumination invariant shape, rotational invariance, but also the time consumed is seldom, emulation
The efficiency of experiment is very high.
Above in conjunction with attached drawing, the embodiment of the present invention is explained in detail, but the present invention is not limited to above-mentioned
Embodiment within the knowledge of a person skilled in the art can also be before not departing from present inventive concept
Put that various changes can be made.
Claims (9)
1. medical figure registration and fusion method based on SIFT+RANSAC algorithm, it is characterised in that:
Specific step is as follows for the medical figure registration and fusion method based on SIFT+RANSAC algorithm:
Step1, building Gaussian scale-space；
Step2, the detection of Gaussian scale-space characteristic point and accurate positioning；
The unstable point of Step3, removal；Point including low contrast and the point positioned at image border；
Step4, characteristic point direction is calculated；
Step5, feature point description for constructing image, form feature vector；
Step6, image registration, i.e., connect the match point of two images, thus obtains a width clearly image registration
Effect picture；
Step7, RANSAC algorithm find out homography matrix, do linear weighted function according to homography matrix, are obtained by linear weighting function
Two image co-registrations are finally piece image by the connection between two images；
The image that Step8, modification were merged.
2. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step1, the Gaussian scale-space of image is defined as function L (x, y, σ), by variable dimension Gaussian convolution G (x,
Y, σ) it generates；Variable dimension Gaussian convolution isWherein (x, y) is space coordinate, and σ is
Scale coordinate, the smoothness of the size decision image of σ, the general picture feature of large scale correspondence image, small scale correspondence image
Minutia, output image are I (x, y), i.e. L (x, y, σ)=G (x, y, σ) * I (x, y).
3. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step2, the detection of Gaussian scale-space characteristic point and accurate positioning: a characteristic point is in Gaussian scale-space
Local Extremum；All scales and picture position are searched in Gaussian scale-space, there are scales by the identification of difference of Gaussian function
With the potential characteristic point of direction invariance；
Since the first floor and last layer all lack an adjacent layer, searches for since every group of the second layer, be current with the second layer
Layer, first layer and third layer are done using third layer as current layer equally after the completion of search again respectively as the upper and lower level of cube
Search；So every layer of point search is twice；By difference of Gaussian function by current pixel 26 pixels adjacent with other into
If it is maximum value or minimum value that row, which compares it, that is, select the point as there are the potential features of scale and direction invariance
Point；Then by being fitted three-dimensional quadratic function accurately to determine key point position and scale.
4. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step3, remove unstable point；Point including low contrast and the point positioned at image border: it is compared by setting
Degree threshold value and Hessian matrix remove both unstable points.
5. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step4, calculate characteristic point direction: the gradient direction distribution characteristic using key point neighborhood territory pixel is each feature
Point assigned direction parameter, makes operator have rotational invariance, to realize image rotation invariance；When actually calculating, with spy
Sampling in neighborhood window centered on sign point, and with the gradient direction of statistics with histogram neighborhood territory pixel；The range of histogram of gradients
It is 0-360 °, histogram is divided into 8 directions, i.e., each characteristic point there are 8 Gradient direction informations；The peak value of histogram represents
The principal direction of neighborhood gradient, the i.e. direction as this feature point at this feature point；With away from the remoter field of central point, its is right
The contribution of histogram is also responsive to reduction；Histogram is carried out using Gaussian function simultaneously smoothly, the influence of mutation to be reduced, in gradient
In direction histogram, when there are the peak value that another is equivalent to 80% energy of main peak value, then this direction is considered this
The auxiliary direction of characteristic point；One characteristic point may be designated with multiple directions, a principal direction, more than one auxiliary direction,
For enhancing matched robustness.
6. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step5, description of image characteristic point is constructed, forms feature vector: for any one key point, i.e. feature
Point, scale space where it take 16 pixels centered on key point × 16 pixel sizes neighborhood, then by this neighborhood region
Uniformly it is divided into 4 × 4 sub-regions, each key point there are 8 gradient directions, therefore, share 4 × 4 × 8=128 data, i.e.,
128 dimension SIFT feature vectors.
7. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step6, after the SIFT feature vector of two images generates, it is special image registration: to calculate characteristic point in two images
The Euclidean distance of vector is levied as similarity determination measurement, it will be apart from the smallest characteristic point as initial matching point, and according to most
It is 0.8 that the ratio between Euclidean distance of neighbour and time neighbour, which is less than some proportion threshold value T, is determined as a pair of of match point, and remove mistake
Match point；The match point in reference picture and image subject to registration is connected with line again realizes image registration.
8. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step7, RANSAC algorithm finds out homography matrix: from image subject to registration to reference picture, appointing and 4 pairs of points is taken to obtain list
Matrix H is answered, the characteristic point in image subject to registration is then projected into reference picture with matrix H, the point and ginseng that detection projection obtains
It examines the distance between accurate characteristic point in image: setting a threshold value: the feature in image subject to registration and reference picture first
Point is to meet a linear regression relation by homography matrix H, and the sum of the distance of match point to the linear regression line of mistake is minimum
Value is fixed threshold value；It is judged as interior point if being less than the threshold value that this sets；The number put in statistics, if more than this
The threshold value set is then judged as available homography matrix, then recalculates homography matrix using all interior points；According to list
It answers matrix to do linear weighted function, the connection between two images is obtained by linear weighting function, be finally by two image co-registrations
Piece image.
9. the medical figure registration and fusion method according to claim 1 based on SIFT+RANSAC algorithm, feature exist
In: in the Step8, modify the image merged: in order to make the syncretizing effect of two images seem more natural, in correspondence
Color or luminance information around point carry out brightness to wherein piece image or color adjusts, and image adjusted is final
As a result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910303778.2A CN110111372A (en) | 2019-04-16 | 2019-04-16 | Medical figure registration and fusion method based on SIFT+RANSAC algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910303778.2A CN110111372A (en) | 2019-04-16 | 2019-04-16 | Medical figure registration and fusion method based on SIFT+RANSAC algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110111372A true CN110111372A (en) | 2019-08-09 |
Family
ID=67485488
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910303778.2A Pending CN110111372A (en) | 2019-04-16 | 2019-04-16 | Medical figure registration and fusion method based on SIFT+RANSAC algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110111372A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111899289A (en) * | 2020-06-19 | 2020-11-06 | 北京理工大学 | Infrared image and visible light image registration method based on image characteristic information |
| CN111915544A (en) * | 2020-07-03 | 2020-11-10 | 三峡大学 | Image fusion-based method for identifying running state of protection pressing plate |
| CN111967337A (en) * | 2020-07-24 | 2020-11-20 | 电子科技大学 | Pipeline line change detection method based on deep learning and unmanned aerial vehicle images |
| CN112102381A (en) * | 2020-09-11 | 2020-12-18 | 西安电子科技大学 | Hardware Trojan horse image registration method based on R-SIFT, storage medium and equipment |
| CN112288009A (en) * | 2020-10-29 | 2021-01-29 | 西安电子科技大学 | R-SIFT chip hardware Trojan horse image registration method based on template matching |
| CN112419374A (en) * | 2020-11-11 | 2021-02-26 | 北京航空航天大学 | Unmanned aerial vehicle positioning method based on image registration |
| CN114266703A (en) * | 2022-03-03 | 2022-04-01 | 凯新创达（深圳）科技发展有限公司 | Image splicing method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101714254A (en) * | 2009-11-16 | 2010-05-26 | 哈尔滨工业大学 | Registering control point extracting method combining multi-scale SIFT and area invariant moment features |
| CN106898019A (en) * | 2017-02-21 | 2017-06-27 | 广西大学 | Method for registering images and device based on Scale invariant Harris features |
| CN107169999A (en) * | 2017-06-19 | 2017-09-15 | 江南大学 | SIFT registration Algorithms based on fusion feature |
| CN107424181A (en) * | 2017-04-12 | 2017-12-01 | 湖南源信光电科技股份有限公司 | A kind of improved image mosaic key frame rapid extracting method |
| CN107491731A (en) * | 2017-07-17 | 2017-12-19 | 南京航空航天大学 | A kind of Ground moving target detection and recognition methods towards precision strike |
| CN108022228A (en) * | 2016-10-31 | 2018-05-11 | 天津工业大学 | Based on the matched colored eye fundus image joining method of SIFT conversion and Otsu |
-
2019
- 2019-04-16 CN CN201910303778.2A patent/CN110111372A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101714254A (en) * | 2009-11-16 | 2010-05-26 | 哈尔滨工业大学 | Registering control point extracting method combining multi-scale SIFT and area invariant moment features |
| CN108022228A (en) * | 2016-10-31 | 2018-05-11 | 天津工业大学 | Based on the matched colored eye fundus image joining method of SIFT conversion and Otsu |
| CN106898019A (en) * | 2017-02-21 | 2017-06-27 | 广西大学 | Method for registering images and device based on Scale invariant Harris features |
| CN107424181A (en) * | 2017-04-12 | 2017-12-01 | 湖南源信光电科技股份有限公司 | A kind of improved image mosaic key frame rapid extracting method |
| CN107169999A (en) * | 2017-06-19 | 2017-09-15 | 江南大学 | SIFT registration Algorithms based on fusion feature |
| CN107491731A (en) * | 2017-07-17 | 2017-12-19 | 南京航空航天大学 | A kind of Ground moving target detection and recognition methods towards precision strike |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111899289A (en) * | 2020-06-19 | 2020-11-06 | 北京理工大学 | Infrared image and visible light image registration method based on image characteristic information |
| CN111915544A (en) * | 2020-07-03 | 2020-11-10 | 三峡大学 | Image fusion-based method for identifying running state of protection pressing plate |
| CN111967337A (en) * | 2020-07-24 | 2020-11-20 | 电子科技大学 | Pipeline line change detection method based on deep learning and unmanned aerial vehicle images |
| CN112102381A (en) * | 2020-09-11 | 2020-12-18 | 西安电子科技大学 | Hardware Trojan horse image registration method based on R-SIFT, storage medium and equipment |
| CN112288009A (en) * | 2020-10-29 | 2021-01-29 | 西安电子科技大学 | R-SIFT chip hardware Trojan horse image registration method based on template matching |
| CN112419374A (en) * | 2020-11-11 | 2021-02-26 | 北京航空航天大学 | Unmanned aerial vehicle positioning method based on image registration |
| CN112419374B (en) * | 2020-11-11 | 2022-12-27 | 北京航空航天大学 | Unmanned aerial vehicle positioning method based on image registration |
| CN114266703A (en) * | 2022-03-03 | 2022-04-01 | 凯新创达（深圳）科技发展有限公司 | Image splicing method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110111372A (en) | Medical figure registration and fusion method based on SIFT+RANSAC algorithm | |
| Ma et al. | Infrared and visible image fusion via gradient transfer and total variation minimization | |
| Bogo et al. | FAUST: Dataset and evaluation for 3D mesh registration | |
| CN105869178B (en) | A kind of complex target dynamic scene non-formaldehyde finishing method based on the convex optimization of Multiscale combination feature | |
| Wang et al. | Multi-cue based tracking | |
| CN103155003B (en) | Posture estimation device and posture estimation method | |
| CN104835175B (en) | Object detection method in a kind of nuclear environment of view-based access control model attention mechanism | |
| Koschan et al. | Color active shape models for tracking non-rigid objects | |
| JP2019125057A (en) | Image processing apparatus, method thereof and program | |
| Alt et al. | Rapid selection of reliable templates for visual tracking | |
| Candemir et al. | Graph-cut based automatic lung boundary detection in chest radiographs | |
| Zeng et al. | View-invariant gait recognition via deterministic learning | |
| CN109523551B (en) | Method and system for acquiring walking posture of robot | |
| CN103503025A (en) | Determining model parameters based on transforming a model of an object | |
| CN112634125B (en) | Automatic face replacement method based on off-line face database | |
| Grewe et al. | Fully automated and highly accurate dense correspondence for facial surfaces | |
| CN108509925B (en) | Pedestrian re-identification method based on visual bag-of-words model | |
| CN107085728A (en) | The method and system effectively scored the probe in image using vision system | |
| JP2008204200A (en) | Face analysis system and program | |
| CN111695431A (en) | Face recognition method, face recognition device, terminal equipment and storage medium | |
| CN106909890A (en) | A kind of Human bodys' response method based on position cluster feature | |
| CN110020627A (en) | A kind of pedestrian detection method based on depth map and Fusion Features | |
| CN109035196A (en) | Image local fuzzy detection method based on conspicuousness | |
| St-Charles et al. | Online multimodal video registration based on shape matching | |
| Ataer-Cansizoglu et al. | Verification of very low-resolution faces using an identity-preserving deep face super-resolution network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190809 |
|
| RJ01 | Rejection of invention patent application after publication |