CN104008555A - Curve detecting method based on backtracking accumulation - Google Patents

Abstract

The invention discloses a curve detecting method based on backtracking accumulation. The method comprises the steps that the feature measurement of a curve is constructed by analyzing the gray feature of the curve; an energy function is constructed to minimize the energy of curve points; one point in the curve is selected as a starting point, and searching is carried out through a shortest path algorithm to obtain a feature image; backtracking accumulation is carried out on each point obtained through searching to obtain a backtracking accumulation diagram; a threshold value is selected from the backtracking accumulation diagram, and the final curve is obtained through extraction. The curve in the image can be fast and efficiently extracted under the condition of only using one starting point.

Description

A kind of based on recalling cumulative curve detection method

Technical field

The present invention relates to information science technology field, relate in particular to a kind of based on recalling cumulative curve detection method.

Background technology

At present, aspect the simple open curve of shortest path technology in detected image, there is reasonable application.Shortest path technology can be used for searching for a paths and make the gross energy minimum of search, can be expressed as mathematical problem and solve:

$U_{p_s}\left(p\right)=\underset{A(p_S,p)}{\min }\left\{E\left(l\right)\right\}=\underset{A(p_s,p)}{\min }\left\{{\∫}_{\mathrm{\Ω}}(P\left(l\left(s\right)\right)+\mathrm{\ω})\mathrm{ds}\right\}---\left(1\right)$

Wherein, be expressed as starting point p _sto the least energy of arbitrfary point p, A (p _s, p) represent starting point p _sto the set of paths of arbitrfary point p in image, E (l) is the energy of curve l, l (s) ∈ R ⁿbe the parameterized form in the time of arc length s, P (l (s)) represents potential energy, and ω is correction term.

Owing to using energy minimization criterion, short path technology is used to active margin model the earliest, extracts curve by minimizing interior energy and external enwergy, but the location comparison sensitivity of active margin model to initial active margin.Afterwards, other scholar, under the starting point and terminal prerequisite of setting curve, made gross energy minimum by separating Eikonal equation, thereby obtained curve, but because the correlativity of the method and active margin is easier to influenced in continuum.Also some scholars has proposed some and has used the method for less prior imformation recently, for example a kind of shortest-path method being called as based on critical point detection, only need to provide a starting point just can search out forward whole piece curve, but still need known length of a curve to stop whole search procedure.

Be not difficult to find that these methods need more priori, such as needing starting point and the terminal of given curve, the in the situation that of a given starting point only, also need length of a curve to be used as priori conditions.In the time that curve has a lot of bifurcated, need a distal point could obtain good testing result for each bifurcated.

Therefore, be just only needing, under the prerequisite of little priori, still to obtain good curve extraction effect in the main research aspect curve detection at present.

Summary of the invention

Technical matters to be solved by this invention is to overcome the dependence of existing curve detection method to priori, knows under the prerequisite of a starting point so propose a kind of need, the curve in can high efficiency extraction image based on recalling cumulative curve detection method.What is called is recalled cumulative referring to: after Shortest Path Searching, after oppositely dating back to starting point or certain counting for each point, curve point can be accessed by high frequency, but not accessed less this feature of curve point, thereby successfully curve point and non-curve point are separated, finally draw out curved profile.

For achieving the above object, the technical solution used in the present invention is:

The present invention is based on and recall cumulative curve detection method, comprise the following steps:

Step 1: analyze the gray feature of any point p in pending image I, the characteristic measure value M (p) of the curvilinear characteristic that structure description point p has, and curve construction point p _ccharacteristic measure value M (p _c) be greater than non-curve point p _non-ccharacteristic measure value M (p _non-c), i.e. M (p _c) > M (p _non-c);

Step 2: by characteristic measure value M (p) the structure energy function P (p) of structure, the energy function of structure meets curve point p _cenergy P (p _c) be less than non-curve point p _non-cenergy P (p _non-c),

Step 3: choose any point p on curve _sas starting point, p _s∈ p _c, use shortest path first to search for pending image I, search is through some p _psdcharacteristic measure value M (p _psd) form characteristic image I _f;

Step 4: the characteristic image I obtaining _fin, the some p of record searching process _psd, to the some p of each search process _psdrecall the cumulative cumulative figure I that obtains recalling _c;

Step 5: for recalling cumulative figure I _c, selected threshold λ _f, described threshold value λ _fscope is [10 ^-3, 10 ³], make I _cin be greater than threshold value λ _fpoint be curve point p _c, all curve point p _cform the curve map I finally obtaining _c.

Further, the inverse that the energy function P (p) constructing is characteristic measure value M (p).

Further, shortest path first is not specified search terminal to pending image I search.

Further, recall cumulative figure I for described _cin any point p _bk, from characteristic image I _fin the some p that searches _psdrecall through p to starting point _bk, p _bkrecalling cumulative figure I _cvalue I _c(p _bk) recall through p for all _bkp _psdcharacteristic measure value M (p _psd) sum.

Compared with prior art, the beneficial effect that the present invention has is:

The inventive method is first according to the gray distribution features curve construction feature representation of curve, and then express the potential energy of each point in image, then use shortest path first, the in the situation that of given starting point, search for, the characteristic pattern obtaining according to Shortest Path Searching is recalled each point, cumulative each eigenwert that is dated back to starting point, a last given suitable less point of threshold value filtering accumulated value, thus obtain final curve.The present invention can only need a starting point, and under the prerequisite of other prior imformations, fast and effeciently detects and extract the curve in image.

Brief description of the drawings

Fig. 1 is pending image simulation figure of the present invention;

Fig. 2 is the characteristic image that the present invention uses the simulation drawing that shortest path first obtains;

Fig. 3 the present invention is based on the curve I recalling in accumulation method extraction simulation drawing _c;

Fig. 4 is the actual crack pattern of the pending image of the present invention;

Fig. 5 is the characteristic image that the present invention uses the actual crack pattern that shortest path first obtains;

Fig. 6 the present invention is based on to recall cumulative method and extract the curve in actual crack pattern.

Embodiment

Below in conjunction with specific embodiment, further illustrate the present invention, should understand these embodiment is only not used in and limits the scope of the invention for the present invention is described, after having read the present invention, those skilled in the art all fall within the application's claims limited range to the amendment of the various equivalent form of values of the present invention.

As shown in Figure 1, illustrate specifically based on recalling cumulative curve detection method with pending image simulation figure, comprise the following steps:

Step 1: the difference of the gray feature of curve and background in the pending image simulation figure of analysis chart 1, definition curvilinear characteristic tolerance M (p) is used for describing the curvilinear characteristic that in Fig. 1, arbitrfary point p has; Utilize the difference of curve gray feature and background gray feature to carry out structural attitude tolerance M (p), make curve point p _ccurvilinear characteristic metric M (p _c) be greater than non-curve point p _non-ccurvilinear characteristic metric, because the gray scale of curve point is lower than the average gray value of background; And curve is a tiny line, the ratio of occupying at certain window inner curve point is less, and background area does not possess this feature, therefore curve point and background area can be separated.

Further, it is w that the wide of window is set, for a known arbitrfary point p _i,j, with p _i,jcentered by, in the window of w × w, calculate p _i,jwith the gray scale difference value of other point in window and ratio (p reciprocal _i,j), for curve point p _cthere is ratio (p _c) > ratio (p _non-c); Wherein, m, n is two variablees, I (p _i,j) expression p _i,jgrey scale pixel value, accordingly, I (p _{i+m, j+n}) interior other p of expression window _{i+m, j+n}gray-scale value; So definition curvilinear characteristic metric is M (p)=I (p) ratio (p).

Step 2: make curve point p by described curvilinear characteristic metric M (p) structure energy function P (p) _cenergy minimum;

By M (p _c) > M (p _non-c), in order to make curve point p _cbe more prone to searched and arrive, in construct image, the potential energy P (p) of arbitrfary point p is the inverse of characteristic measure M (p),

Step 3: choose the known any point p in curve _sas starting point, p _s∈ p _c, use shortest path first to search for pending image I, the some p of search process _psdcurvilinear characteristic metric M (p _psd) form characteristic image I _f;

Concrete, as shown in Figure 1, a in Fig. 1, b, c is the point on curve, i.e. a ∈ p _c, b ∈ p _c, c ∈ p _cget an a as starting point, carry out Shortest Path Searching with the dijkstra's algorithm in shortest path first, when search, in using a point as the neighborhood at center, carry out, record the accumulative total cost of this point to starting point when some when searching certain, select the point of an accumulative total Least-cost to proceed to calculate at every turn; The point p of search process _psdcurvilinear characteristic metric M (p _psd) form characteristic image, as shown in Figure 2, the gray-scale value in Fig. 2 is greater than 0 point and is the some p searching _psd, for example, put b, d, e; And the point that gray-scale value is 0 is the some p not searching _non-psd, for example f point in figure.

Step 4: the characteristic image I obtaining _fin, from each search process point p _psdstart, recall the cumulative cumulative figure I that obtains recalling to starting point _c;

Concrete, for the characteristic image I obtaining in step 3 _f(Fig. 2), because we are described for curvilinear characteristic, so have larger characteristic measure M (p as the b in Fig. 1, c curve point _c) value and less potential energy P (p _c); For the p of the accessed mistake obtaining from step 3 _psdpoint, as some b, d, e in Fig. 2, while recalling, by its characteristic measure value M (p _psd) be added to the some p of process in trace-back process _bkon obtain recalling cumulative figure I _c,

I _c(p _bk)＝ΣI _f(p _bk) (3)

Wherein, a process point p while recalling _bkbelong to the arbitrfary point p of searched process _psd.For example d point, recalls while carrying out reverse search through d to starting point a ₁, d ₂, d ₃arrive b point and return a point, similarly, process e when e point is recalled to starting point a ₁, e ₂, e ₃, e ₄arrive b point and return a point, curve point a, b will be accessed in the time recalling more continually than non-curve point d and e point as can be seen here, therefore have I _c(p _c) > I _c(p _non-c).

Step 5: recalling cumulative figure I _cmiddle selected threshold λ _f, described threshold value λ _fscope is [10 ^-3, 10 ³], I _cin be greater than threshold value λ _fpoint be curve point p _c, all curve point p _cform the curve map I finally obtaining _c, as shown in Figure 3.

Concrete, due to I _c(p _c) > I _c(p _non-c) be curve point p _cthe I at place _c(p _c) value will be larger, setting threshold λ _f, make I _c(p _c) > λ _f> I _c(p _non-c), therefore at given threshold value λ _funder obtain final curve point set figure I _c, as black curve in Fig. 3 indicates, wherein I _c=I _c> λ _f.

Recruitment evaluation

First use a simulation drawing that has curve, as shown in Figure 1, carry out march line drawing, wherein, in simulation drawing, the average gray value of curve is lower than background value, and background is by random generation, obtains curve map I by the method for the invention _c, as shown in Figure 3, in Fig. 3, use black line to indicate the curve detecting.In order to verify the effect of the present invention in actual conditions, use real crack pattern, as Fig. 4 tests, crack pattern is except the character that has simulation drawing and have, and picture quality is also subject to the impact of many textures, shade, illumination etc., therefore more has challenge.With the starting point p shown in Fig. 4 _sfor starting point, utilize the energy function P (p) constructing, use shortest path first to search for and obtain characteristic image I _f, as shown in Figure 5; Then to the characteristic image I shown in Fig. 5 _fin each point recall cumulatively, obtain recalling cumulative figure I _c, certain threshold value λ is set _f, make I _c(p _c) > λ _f> I _c(p _non-c), thereby obtain final curve map I _c, as shown in Figure 6, the curve that Fig. 6 uses black line to indicate to detect.Experimental result shows, the extraction efficiency that method of the present invention all has in composite diagram and crack pattern and quality.

Visual assessment: by observing the extraction result of Fig. 3 and Fig. 6, we can see and use the inventive method can fast and effeciently extract the curve in publishing picture under the situation that a starting point is only provided.

Claims (4)

1. based on recalling a cumulative curve detection method, it is characterized in that, comprise the following steps:

Step 1: analyze the gray feature of any point p in pending image I, the curvilinear characteristic metric M (p) of the curvilinear characteristic that structure description point p has, and curve construction point p _ccurvilinear characteristic metric M (p _c) be greater than non-curve point p _non-ccurvilinear characteristic metric M (p _non-c), i.e. M (p _c) > M (p _non-c);

Step 2: by curvilinear characteristic metric M (p) the structure energy function P (p) of structure, the energy function of structure meets curve point p _cenergy P (p _c) be less than non-curve point p _non-cenergy P (p _non-c), i.e. P (p _c) < P (p _non-c);

Step 3: choose any point p on curve _sas starting point, p _s∈ p _c, use shortest path first to search for pending image I, search is through some p _psdcurvilinear characteristic metric M (p _psd) form characteristic image I _f;

Step 4: the characteristic image I obtaining _fin, the some p of record searching process _psd, to the some p of each search process _psdrecall the cumulative cumulative figure that obtains recalling

Step 5: for recalling cumulative figure selected threshold λ _f, described threshold value λ _fscope is [10 ^-3, 10 ³], make in be greater than threshold value λ _fpoint be curve point p _c, all curve point p _cform the curve map I finally obtaining _c.

2. curve detection method according to claim 1, is characterized in that, the inverse that the energy function P (p) constructing is curvilinear characteristic metric M (p).

3. curve detection method according to claim 1, is characterized in that, shortest path first is not specified search terminal to pending image I search.

4. curve detection method according to claim 1, is characterized in that, recalls cumulative figure for described in any point p _bk, from characteristic image I _fin the some p that searches _psdrecall through p to starting point _bk, p _bkrecalling cumulative figure value recall through p for all _bkp _psdcurvilinear characteristic metric M (p _psd) sum.