CN103136760B - A kind of multi-sensor image matching process based on FAST Yu DAISY - Google Patents

Abstract

The invention discloses image analysis technology field, particularly relate to a kind of multi-sensor image matching process based on FAST Yu DAISY.Its technical scheme is, first, respectively reference picture and image to be matched is carried out FAST Corner Detection, extracts characteristic point；Secondly, respectively reference picture and image zooming-out to be matched characteristic point out are done DAISY process, generate and describe subvector；Finally, use Euclidean distance to obtain initial matching point pair from characteristic point, then filter out accurate matching double points with RANSAC algorithm from initial matching point centering, thus realize reference picture and the coupling of image to be matched.Present invention substantially reduces the time of whole matching process and achieve preferable matching effect, the conceptual design of relevant issues is had certain reference.

Description

A kind of multi-sensor image matching process based on FAST Yu DAISY

Technical field

The invention belongs to image analysis technology field, particularly relate to a kind of multi-sensor image based on FAST Yu DAISY Matching process.

Background technology

Images match is as a kind of key technology of Pattern recognition and image processing, at remotely-sensed data analysis, medical image Process, and computer vision aspect has been obtained for being widely applied.

In simple terms, images match is exactly the technology carrying out the different images that Same Scene shoots aliging, and i.e. relates to From various visual angles, multi-time Scales and multi-modal etc. analysis.Imaging mode yet with different images sensor there are differences, and causes leading to There is obvious limitation and otherness, so multi-sensor image matching technique is in the view data crossing different sensors acquisition One particularly important problem.

The detection of characteristic point and description are steps particularly important in images match, therefore feature point detection and description algorithm Quality directly influence the effect of coupling.

FAST (Features from Accelerated Segment Test) is a kind of simple and quick feature spot check Method of determining and calculating, this algorithm can quickly determine angle point.And DAISY feature interpretation algorithm uses the histogram replacement of convolution to commonly use Histogram, reduces computing cost, it is achieved that the dense correspondence of wide baseline.At present, FAST and DAISY is the most respectively at single-sensor Images match has been applied.

Summary of the invention

In order to improve the accuracy and speed of multi-sensor image coupling, the present invention proposes a kind of based on FAST and DAISY Multi-sensor image matching process.

A kind of multi-sensor image matching process based on FAST Yu DAISY, it is characterised in that described method specifically includes Following steps:

Step 1: respectively reference picture and image to be matched are carried out FAST Corner Detection, extract characteristic point；

Step 2: reference picture and image zooming-out to be matched characteristic point out do DAISY process respectively, generates and describes Subvector；

Step 3: use Euclidean distance to obtain initial matching point pair from characteristic point, then with RANSAC algorithm from initial Join a centering and filter out accurate matching double points, thus realize reference picture and the coupling of image to be matched.

The process of described feature point extraction is:

Step 101: choose a candidate point C to be determined；

Step 102: check the point with a C as the center of circle, on the circle with R as radius；Consider 16 pixels on circumference, if There is the grey scale pixel value of continuous 9 and above point more than/grey scale pixel value less than some C and difference more than setting threshold value T, then recognize It is a characteristic point for a C.

The process that described generation describes subvector is:

Step 201: for each characteristic point, calculates its convolution in 8 directions；

Step 202: form characteristic vector with the convolution value cascade of 8 different directions calculating gained, and this vector is entered Row normalized obtains unit character vector；

Step 203:DAISY feature interpretation algorithm uses roundness mess method to obtain feature neighborhood of a point and supports point set, And utilize unit character vector to generate the description subvector of this point.

Described employing Euclidean distance obtains the process of initial matching point pair from characteristic point:

Step 301: the description subvector of certain characteristic point in reference picture is special with each in image to be matched respectively Levy description subvector a little and carry out Euclidean distance calculating；

Step 302: by the ratio of the minimum euclidean distance corresponding to this feature point that obtains and time minimum euclidean distance with The threshold value set compares；If the ratio of minimum euclidean distance and time minimum euclidean distance is less than the threshold value set, then should Characteristic point constitutes an initial matching point pair with the characteristic point in the image to be matched corresponding to minimum euclidean distance；Otherwise, recognize Coupling cannot be obtained for this feature point in reference picture；

Step 303: repeat step 301 and 302, characteristic point remaining in reference picture is carried out initial matching, until ginseng Examine till in image, all of characteristic point all completes initial matching.

The computing formula of described Euclidean distance is:

$d(x,y)=\sqrt{{(x_1-y_1)}^2+\·\·\·+{(x_p-y_p)}^2}$

Wherein, p represents the dimension in space；x=(x₁,…,x_p) it is with reference to the description subvector of characteristic point in reference picture；y= (y₁,…,y_p) it is the description subvector of characteristic point in image to be matched.

The present invention detects angle point first with FAST algorithm, then uses DAISY arthmetic statement, then realizes coupling.This Bright practical, greatly reduce the time of whole matching process and achieve preferable matching effect, the side to relevant issues Case is designed with certain reference.

Accompanying drawing explanation

Fig. 1 is the flow chart of a kind of based on FAST Yu DAISY multi-sensor image matching process that the present invention provides；

Fig. 2 is the experimental result picture mated remote sensing images by the inventive method；Wherein, (a) and (b) be respectively by The remote sensing images of two width difference spectrum of Daedalus scanner shooting；C () is through reducing postrotational image to (a)；(d) (e) it is the image after (c) and (b) detects angle point respectively；F () is the design sketch of accurate matching double points line；

Fig. 3 is the infrared image of transformer and visible images carries out the experimental result picture that mates；Wherein, (a) and (b) point It not infrared image and visible images；C () and (d) is (a) and (b) image after FAST Corner Detection respectively；E () is The accurately design sketch of matching double points line.

Detailed description of the invention

Below in conjunction with the accompanying drawings, preferred embodiment is elaborated.It should be emphasized that the description below is merely exemplary Rather than in order to limit the scope of the present invention and application thereof.

Fig. 1 is the flow chart of a kind of based on FAST Yu DAISY multi-sensor image matching process that the present invention provides.Figure In 1, the concrete steps of a kind of multi-sensor image matching process based on FAST with DAISY include:

Step 1: respectively reference picture and image to be matched are carried out FAST Corner Detection, extract characteristic point；

Step 2: reference picture and image zooming-out to be matched characteristic point out do DAISY process respectively, generates and describes Subvector；

Step 3: use Euclidean distance to obtain initial matching point pair from characteristic point, then with RANSAC algorithm from initial Join a centering and filter out accurate matching double points, thus realize reference picture and the coupling of image to be matched.

Set reference picture and have m characteristic point (description subvector), image to be matched have n characteristic point (description sub to Amount), then the detailed process of initial matching is as follows:

Step a: calculate the 1st characteristic point and the Euclidean distance of all characteristic points of image to be matched in reference picture, it is thus achieved that n Individual Euclidean distance；

Step b: the minimum euclidean distance of n Euclidean distance and the ratio of time minimum euclidean distance and the setting that will obtain Threshold value compares；If the ratio of minimum of a value and time minimum of a value is less than threshold value set in advance in this n Euclidean distance, then the Levy for 1 and a little constitute an initial matching point pair with the characteristic point in the image to be matched corresponding to minimum euclidean distance, otherwise recognize Coupling cannot be obtained for the in reference picture the 1st characteristic point；

Step c: repeating step a and b, m-1 characteristic point remaining to reference picture carries out initial matching.

Embodiment 1:

With multi-sensor image matching process based on FAST with DAISY, the remote sensing images of the two different spectrum of width are imitated True experiment.Original image is respectively as shown in Fig. 2 (a) and Fig. 2 (b), and wherein Fig. 2 (c) is that Fig. 2 (a) reduces 0.8 times, the most inverse Image after hour hands rotation 10 degree.Fig. 2 (c) and Fig. 2 (b) is mated by this experiment, first two width images is carried out FAST angle Point extracts, and wherein selects FAST9 detection, and threshold value is 30, and result is respectively as shown in Fig. 2 (d) and Fig. 2 (e)；The most respectively to Fig. 2 D the angle point of () and Fig. 2 (e) carries out DAISY description and generates description vectors, wherein the parameter of DAISY description vectors is set to: Central pixel point is 15 to the distance of outer outermost layer mesh point, and the number around central point annulus is 3, the histogram in a layer Quantity is 8, and the direction quantity of histogram gradient is 8；Then utilize Euclidean distance to obtain initial matching point pair, and calculate with RANSAC Method removes Mismatching point pair, accurate matching double points such as Fig. 2 (f).Can be seen that this method can successfully realize from the simulation experiment result The coupling of the remote sensing images of two width difference spectrum.

Embodiment 2:

With multi-sensor image matching process based on FAST and DAISY to from on-the-spot infrared image and visible ray figure As processing.Fig. 3 (a) and Fig. 3 (b) is respectively infrared image and the visible images of transformer.The most respectively to two width figures As carrying out FAST Corner Detection, wherein selecting FAST9 detection, threshold value is 30, shown in result such as Fig. 3 (c) and Fig. 3 (d)；Then Respectively the angle point of Fig. 3 (c) He Fig. 3 (d) is carried out DAISY description and generates description vectors, wherein the parameter of DAISY description vectors It is set to: the distance of central pixel point to outer outermost layer mesh point is 15, and the number around central point annulus is 3, at one layer In histogram quantity be 8, the direction quantity of histogram gradient is 8；Euclidean distance is finally utilized to obtain initial matching point pair, and Mismatching point pair is removed, accurately shown in the line effect such as Fig. 3 (e) after coupling with RANSAC algorithm.Can from experimental result Go out this method and the matching treatment from on-the-spot different sensor images can be obtained preferable matching effect.

The above, the only present invention preferably detailed description of the invention, but protection scope of the present invention is not limited thereto, Any those familiar with the art in the technical scope that the invention discloses, the change that can readily occur in or replacement, All should contain within protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of the claims It is as the criterion.

Claims (2)

1. a multi-sensor image matching process based on FAST Yu DAISY, it is characterised in that described method specifically include with Lower step:

Step 1: if reference picture is the two different spectrum of width shot by Daedalus scanner with image to be matched respectively One of them remote sensing images is reduced 0.8 times and rotate 10 degree relative to another remote sensing images by remote sensing images, then to reducing Two width images after Zhuaning carry out FAST Corner Detection, extract characteristic point；

If reference picture and image to be matched are respectively infrared image and the visible images of transformer, respectively to two width images Carrying out FAST Corner Detection, wherein select FAST 9 to detect son, threshold value is 30, extracts characteristic point；

Step 2: respectively reference picture and image zooming-out to be matched characteristic point out are done DAISY process, generate describe son to Amount；

Step 3: use Euclidean distance to obtain initial matching point pair from characteristic point, then with RANSAC algorithm from initial matching point Centering filters out accurate matching double points, thus realizes reference picture and the coupling of image to be matched；

The process of described feature point extraction is:

Step 101: choose a candidate point C to be determined；

Step 102: check the point with a C as the center of circle, on the circle with R as radius；Consider 16 pixels on circumference, if had even The grey scale pixel value of continuous 9 and above point more than/grey scale pixel value less than some C and difference more than setting threshold value T, then it is assumed that point C is a characteristic point；

The process that described generation describes subvector is:

Step 201: for each characteristic point, calculates its convolution in 8 directions；

Step 202: form characteristic vector with the convolution value cascade of 8 different directions calculating gained, and this vector is returned One change processes and obtains unit character vector；

Step 203:DAISY feature interpretation algorithm uses roundness mess method to obtain feature neighborhood of a point and supports point set, and profit The description subvector of this point is generated with unit character vector；The parameter of DAISY description vectors is set to: central pixel point is to outward The distance of outermost layer mesh point is 15, and the number around central point annulus is 3, and the histogram quantity in one layer is 8, histogram The direction quantity of gradient is 8；

Described employing Euclidean distance obtains the process of initial matching point pair from characteristic point:

Step 301: by reference picture certain characteristic point describe subvector respectively with each characteristic point in image to be matched Description subvector carry out Euclidean distance calculating；

Step 302: by the minimum euclidean distance corresponding to this feature point that obtains and the ratio of time minimum euclidean distance and setting Threshold value compare；If the ratio of minimum euclidean distance and time minimum euclidean distance is less than the threshold value set, then this feature Point constitutes an initial matching point pair with the characteristic point in the image to be matched corresponding to minimum euclidean distance；Otherwise it is assumed that ginseng Examine this feature point in image and cannot obtain coupling；

Step 303: repeat step 301 and 302, characteristic point remaining in reference picture is carried out initial matching, until with reference to figure Till in Xiang, all of characteristic point all completes initial matching.

A kind of multi-sensor image matching process based on FAST Yu DAISY the most according to claim 1, its feature exists In, the computing formula of described Euclidean distance is:

$d(x,y)=\sqrt{{(x_1-y_1)}^2+\mathrm{...}+{(x_p-y_p)}^2}$

Wherein, p represents the dimension in space；X=(x₁,…,x_p) it is the description subvector of characteristic point in reference picture；Y= (y₁,…,y_p) it is the description subvector of characteristic point in image to be matched.