CN101916441A - Freeman chain code-based method for matching curves in digital image - Google Patents

Abstract

The invention relates to a Freeman chain code-based method for matching curves in a digital image, which is technically characterized by comprising the following steps of: encoding a characteristic curve with eight-direction Freeman chain codes first; detecting corner points of the curves by adopting a difference accumulated value and three-point chain code difference-based method for detecting the corner points of Freeman chain code curves; calculating the starting points and direction of the curves of a detected corner point sequence according to corner point distance information and information on the included angles between the midpoints of straight-line portions among the corner points and midpoints of the curves; rearranging the detected corner point sequence according to the starting points and direction of the curves to obtain a corner point sequence which is not changed with the rotation, translation and dimension change of the curves; and calculating a length sequence of the corner points and an included angle sequence of the corner points according to the corner point sequence, and performing similar judgment on the two curves through the length sequence of the corner points and the included angle sequence of the corner points to match the curves. The method has the advantages of no influence of rotation, scaling and translation of the curves, small calculated amount and easy realization, and is a practical method for matching the curves.

Description

In a kind of digital picture based on the Curve Matching method of Freeman chain code

Technical field

The present invention relates to the Curve Matching method in a kind of digital picture, relate in particular to a kind of images match and identification in the digital picture of belonging to, be suitable for having rotation, the matching operation between convergent-divergent and the translation curve based on the Freeman chain code.

Background technology

Object identification is very important field in the Flame Image Process, and one of method of object identification is exactly to extract the characteristic curve of object, these characteristic curvees is mated finish object identification then.Carried out a large amount of research at present both at home and abroad at the Curve Matching problem, had and use the semi-differential invariant to carry out Curve Matching, this method will be calculated the tangent vector of each point, and calculated amount is big, does not utilize the geometrical property of curve simultaneously; Also have the batten of employing to carry out Curve Matching, this method is to adopt batten to come approximating curve, has therefore determined its scope; Also the someone proposes to adopt straight line to approach to carry out Curve Matching, and this calculation of complex of approaching is difficult for realizing, can not solve the matching problem that has rotation and change of scale simultaneously.

Summary of the invention

The technical matters that solves

For fear of the deficiencies in the prior art part, the present invention proposes a kind of digital picture by the Curve Matching method based on the Freeman chain code, is a kind of practical and effective Curve Matching method.

Thought of the present invention is: curve is carried out can detecting corner point very easily behind the Freeman chain code coding, employing has order information based on the detected corner point of Freeman chain code curve corner point detection method of difference accumulated value and 3 chain code differences, if therefore taking a kind of appropriate method resequences to the detected next corner point sequence of Freeman chain code, obtain one not with the corner point sequence of the standard of rotation, convergent-divergent and translation transformation, such standard corner point sequence is carried out similarity between angle and distance judge and realize Curve Matching.

Technical scheme

Based on the Curve Matching method of Freeman chain code, it is characterized in that in a kind of digital picture:

Step 1: curve is carried out all directions obtain the Freeman chain code, adopt " based on the Freeman chain code curve corner point detection method of difference accumulated value and 3 chain code differences " that curve is carried out corner point and detect to Freeman chain code coding;

Step 2 is determined curve starting point and direction of a curve:

A: definite method of open curve:

The central point of curve and the central point on the curve string are connected into straight line, this straight line and curve string constitute two angles, the end points that two angles are compared pairing curve one end of the little angle of angle is a starting point, and direction of curve is to open curve opposite side end points from starting point;

If two angles equate, one section curve with arbitrary angle correspondence is that open curve repeats said method, unequal until two angles, be starting point with the end points of comparing pairing curve one end of little angle, direction of curve is to open curve opposite side end points from starting point;

B: definite method of closed curve:

Doing a straight line, closed curve is divided into two open curves at a distance of between two the longest corner points; Adopt definite method of above-mentioned open curve to obtain the minimum separately angle of two open curves respectively then, defining and comparing little angle between two angles is the direction calculating angle;

If the angle of two minimums equates, be that open curve repeats said method with one section curve of arbitrary angle correspondence, unequal until two angles, be the direction calculating angle to compare little angle;

Corner point on the line correspondence of direction calculating angle is a starting point, and direction of a curve is the short direction of curve distance from starting point to another limit, direction calculating angle and intersections of complex curve;

Be all the longest corner point apart if having two pairs in the closed curve corner point, between two pairs of corner points, do two straight lines, with two straight lines is that string calculates direction calculating angle separately, and calculated curve starting point and direction of a curve are come in the less real direction calculating angle of conduct, direction calculating angle;

Step 3: the curve starting point of determining with step 2 is the starting point of corner point sequence, and the direction of a curve that obtains according to step 2 sorts to the curve corner point that step 1 obtains, and obtains the corner point sequence;

Step 4 is calculated the length sequences { d between the adjacent corners point _i}:

(x wherein _i, y _i) be corner point g _iCoordinate, (x _I-1, y _I-1) be corner point g _I-1Coordinate, d _iBe a g _iTo a g _I-1Euclidean distance; g _iAnd g _I-1Be the corner point sequence G=(g after step 3 ordering ₁, g ₂, g ₃..., g _n) in corner point; I=(2,3,4,5,6 ... n);

Step 5 is calculated corner point angle sequence { θ _i}:

D wherein _I-1And d _iBe respectively length sequences { d between corner point _iI-1 and i point value;

Be a g _I-1To a g _I+1Euclidean distance;

Step 6 is calculated angle difference value T:

θ wherein _MiBe curve m angle sequence { θ _MiIn i angle value; θ _LiBe curve l angle sequence { θ _LiIn i angle value; M and l are two curves that mate calculating;

Step 7 computational length difference value S:

D wherein _MiBe curve m length sequences { d _MiIn the i point value; d _LiBe curve l length sequences { d _MiIn the i point value;

For the length sequences of curve m and;

For the length sequences of curve l and;

Step 6: as angle difference value T＜T _ThreWith difference in length value S＜S _ThreThe time, think that two curves are complementary, otherwise think and do not match; Described T _ThreFor the angle threshold value is 5～20; S _ThreBe length threshold, value is 3～15.

Described curve is that width is the digitizing two-value curve of a pixel.

Beneficial effect

Based on the Curve Matching method of Freeman chain code, only need that the corner point sequence is carried out similarity and judge that calculated amount is little in a kind of digital picture that the present invention proposes; Two, to a curve, through behind rotation, convergent-divergent and the translation transformation, do not change through the resulting corner point sequence of this method, therefore, this matching process is not subjected to the influence of curve rotation, convergent-divergent and translation.

Description of drawings

Fig. 1: the inventive method process flow diagram

Fig. 2: Freeman is from all directions to value

Fig. 3: the Freeman chain code points to

Fig. 4: the open curve starting point is calculated synoptic diagram

Fig. 5: open curve is the angle calculation synoptic diagram repeatedly

Fig. 6: closed curve starting point and direction of a curve calculate synoptic diagram

Fig. 7: use this method to finish the Curve Matching example

(a) curve model

(b) curve to be matched

(c) model corner point

(d) corner point that is complementary with the model corner point

Embodiment

Now in conjunction with the embodiments, accompanying drawing is further described the present invention:

Step 1:Freeman chain code

Digitized two-value contour curve can be eight kinds of possible direction values of adjacent two pixel lines to the Freeman chain code, as shown in Figure 2 with from all directions to Freeman chain representation from all directions.Article one, curve forms n bar chain after by grid discretization, and this curving chain code can be expressed as { c (i) } ⁿ, a direction in eight directions of every chain sensing, c (i) ∈ 0,1 ... 7}, i are the index value of pixel, and c (i) is the direction chain code that is pointed to pixel (i+1) by pixel (i), as shown in Figure 3.For example, if current pixel is p (i), b ₇Be the next pixel on the curve, then p (i) value is 7.The Curve Matching algorithm that this paper proposes just is being based on such coded system.

The detection of step 2:freeman chain code sequence corner point:

The present invention adopts the Freeman chain code curve corner point detection method based on difference accumulated value and 3 chain code differences to detect corner point, this method is the difference accumulated value of calculated curve freeman chain code at first, extract the real corner point of a part and some doubtful corner points by difference accumulated value, 3 chain codes that calculate doubtful corner point then are poor, if 3 chain code differences greater than some given threshold point, are then thought real corner point; At last detected all real corner points are put in order, obtain the corner point sequence of this curve, in this corner point sequence, if distance is very near between two corner points, distance is less than 3～8, then get 3 chain code differences bigger a bit as true corner point, 3 less corner points of chain code difference are deleted from the corner point sequence.This method is calculated simple, and accurate and effective is simultaneously consistent for the detected corner point of the curve of rotation, convergent-divergent and translation, is highly suitable for the present invention.

Step 3: calculated curve starting point and direction of a curve

For open curve, suppose that the P point is a curve

Mid point, the mid point of the curve is here got the central value of curve pixel, Q is the mid point of the formed line segment of AB line, connects QP, the angle of QP and QA is α, the angle of QP and QB is β, if α＜β thinks that then terminal A is a starting point, direction is from A to B; If α＞β thinks that then terminal B is a starting point, direction is from B to A; If α=β then continues to get curve

Mid point P _A1, curve

Mid point P _B1, calculate QP _A1Angle α with QA ₁, calculate P _B1Angle β with QB ₁If, α ₁＜β ₁, then terminal A is a starting point, if α ₁＞β ₁, then terminal B is a starting point; If α ₁=β ₁, then get

Mid point P _A2With

Mid point P _B2, proceed above-mentioned angle relatively, so circulation is till finding starting point.For some specific curves, it is still identical repeatedly to calculate its angle, thinks that then it is a symmetrical curve, and for symmetrical curve, getting any end points does not influence result of calculation as starting point.Fig. 4 is one and obtains the curve of starting point through angle calculation, and Fig. 5 is for obtaining the curve of starting point through twice angle calculation.

For closed curve, at first to calculate the line segment distance between per two corner points in the corner point sequence, to be designated as A, B apart from two corner points of maximum, connect A, B, line segment AB is partitioned into two open curves with closed curve, calculate the mid point of two open curves, be designated as Pm and Pn respectively, such two open curves can be designated as

With The mid point of line segment AB is designated as Q, connects QPm and QPn, calculates the acute angle that AB is become with QPm, is designated as θ, AB and QPn acutangulate are γ, compare the size of θ and γ, if θ＜γ, then calculate the angle as direction of curve, if γ＜θ then calculates the angle with angle γ as direction of curve with angle θ.After obtaining the direction calculating angle, can obtain curve starting point and direction of a curve by following judgement: corner point A must be wrapped in the direction calculating angle, among the B one, the corner point that the direction calculating angle is comprised is as the curve starting point, the direction that curve distance from the curve starting point to another limit, direction calculating angle and intersections of complex curve is lacked is as direction of a curve, γ＜θ as shown in Figure 5, then γ is that direction of a curve calculates the angle, angle γ has comprised corner point A, think that then corner point A is the starting point of this closed curve, direction of curve is the direction of γ angle, i.e. indicated direction among the figure.If θ=γ then proceeds two calculation procedures that angle equates in the open curve, up to finding a less angle to calculate the angle as direction of curve.

If one closed curve contains two and adjusts the distance and equate and the maximum corner point of distance simultaneously, line segment AB and A ' B ' distance equates and is all the point of distance maximum between this curve corner point as shown in Figure 5, need calculate the acute angle angle that line segment AB and A ' B ' are become with its corresponding mid point of curve this moment respectively, the acute angle of Fig. 5 angle is respectively θ, γ, θ ', γ ', find out θ, γ, θ ', one of the middle angle minimum of γ ' calculates the angle as direction of a curve, can find out the starting point and the direction of a curve of this class curve.

By top calculating, can obtain starting point and direction of a curve to open curve, thereby can obtain starting point and the direction of a curve that the direction calculating angle has also obtained curve to closed curve, according to starting point that obtains and direction of curve the corner point sequence in the curve is resequenced, obtain the corner point sequence of a standard.Because for same curve, its starting point and direction of curve are unique, therefore the corner point sequence that obtains also is unique.

Step 4: length sequences and corner point angle sequence between corner point

After obtaining the standardization corner point sequence that curve rearranges, can obtain length sequences and corner point angle sequence between corner point according to the standardization sequence.

1, Freeman chain code length sequences

Length of a curve sequence { d _iExpression, then have:

$d_i=\sqrt{{(x_i-x_{i-1})}^2+{(y_i-y_{i-1})}^2};$

d _iBe a g _iTo a g _I-1Euclidean distance, (x _i, y _i) corner point g _iCoordinate, (x _I-1, y _I-1) corner point g _I-1Coordinate, g _iAnd g _I-1Be the flex point sequence G=(g after the rearrangement ₁, g ₂, g ₃..., g _n) in point, i=(2,3,4,5,6 ... n).

2, angle sequence between corner point

Suppose rearrangement after the corner point sequence be K=(g ₁, g ₂, g ₃..., g _n), the line segment length sequence of corresponding adjacent 2 connections is { d _i, calculate g successively ₁And g ₂Angle theta ₁, calculate g ₂And g ₃Angle theta ₂, up to calculating g _N-1And g _nAngle theta _N-1With angle theta ₁θ ₂θ ₃θ _N-1Preserve with the corner point sequence is corresponding in order.

D wherein _I-1And d _iBe respectively length sequences { d between flex point _iI-1 and i point value; Be a g _I-1To a g _I+1Euclidean distance; I=(2,3,4,5,6 ... n-1); Then have:

$z_i^2=d_{i+1}^2+d_i^2-2d_{i+1}{d}_i\cos {\mathrm{\θ}}_i---\left(1\right)$

Have after the conversion:

${\mathrm{\&theta;}}_i=\arccos \left(\frac{d_{i+1}^2+d_i^2-{\mathrm{<figure-callout\; id="2"\; label="z\; i"\; filenames="HSA00000222767700012.png"\; state="\{\{state\}\}">z</figure-callout>}}_i^{}}{2d_{i+1}{d}_i}\right)---\left(2\right)$

Can obtain angle sequence { θ _i.

Step 5: Curve Matching

The angle difference degree of two curve m and l is measured by following formula.

$T=\underset{i=2}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&theta;}}_{\mathrm{mi}}-{\mathrm{\&theta;}}_{\mathrm{li}})---\left(3\right)$

T is the perspective curve difference value in the following formula; θ _MiBe curve m angle sequence { θ _MiI angle value; θ _LiBe curve l angle sequence { θ _LiI angle value.

Cause the mistake coupling in order to prevent that some curve corner point is less, length of a curve also carried out diversity factor weigh:

$S=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}(\frac{D_{\mathrm{mi}}}{D_{\mathrm{li}}}-\frac{D_m}{D_l})---\left(4\right)$

S is the length of curve difference value in the following formula;

For the length sequences of curve m and; For the length sequences of curve l and; d _MiBe curve m length sequences { d _MiIn the i point value; d _LiBe curve l length sequences { d _MiIn the i point value.

For the curve that has change of scale, the conversion ratio of its each corner point half interval contour length is consistent, what the computing method of top length similarity were calculated is the difference of a ratio, and therefore for the curve that has change of scale, top length similarity is judged also suitable.

Last set angle threshold value T _ThreWith length threshold S _Thre, general T _ThreValue is 5～20, S _ThreValue is 3～15, if T＜T _Thre, S＜S _Thre, think that two curves are complementary, otherwise think and do not match.

Claims (2)

1. In the digital picture based on the Curve Matching method of Freeman chain code, it is characterized in that:

   Step 1: curve is carried out all directions obtain the Freeman chain code, adopt " based on the Freeman chain code curve corner point detection method of difference accumulated value and 3 chain code differences " that curve is carried out corner point and detect to Freeman chain code coding;

   Step 2 is determined curve starting point and direction of a curve:

   A: definite method of open curve:

   The central point of curve and the central point on the curve string are connected into straight line, this straight line and curve string constitute two angles, the end points that two angles are compared pairing curve one end of the little angle of angle is a starting point, and direction of curve is to open curve opposite side end points from starting point;

   If two angles equate, one section curve with arbitrary angle correspondence is that open curve repeats said method, unequal until two angles, be starting point with the end points of comparing pairing curve one end of little angle, direction of curve is to open curve opposite side end points from starting point;

   B: definite method of closed curve:

   Doing a straight line, closed curve is divided into two open curves at a distance of between two the longest corner points; Adopt definite method of above-mentioned open curve to obtain the minimum separately angle of two open curves respectively then, defining and comparing little angle between two angles is the direction calculating angle;

   If the angle of two minimums equates, be that open curve repeats said method with one section curve of arbitrary angle correspondence, unequal until two angles, be the direction calculating angle to compare little angle;

   Corner point on the line correspondence of direction calculating angle is a starting point, and direction of a curve is the short direction of curve distance from starting point to another limit, direction calculating angle and intersections of complex curve;

   Be all the longest corner point apart if having two pairs in the closed curve corner point, between two pairs of corner points, do two straight lines, with two straight lines is that string calculates direction calculating angle separately, and calculated curve starting point and direction of a curve are come in the less real direction calculating angle of conduct, direction calculating angle;

   Step 3: the curve starting point of determining with step 2 is the starting point of corner point sequence, and the direction of a curve that obtains according to step 2 sorts to the curve corner point that step 1 obtains, and obtains the corner point sequence;

   Step 4 is calculated the length sequences { d between the adjacent corners point _i}:

   

   (x wherein _i, y _i) be corner point g _iCoordinate, (x _I-1, y _I-1) be corner point g _I-1Coordinate, d _iBe a g _iTo a g _I-1Euclidean distance; g _iAnd g _I-1Be the corner point sequence G=(g after step 3 ordering ₁, g ₂, g ₃..., g _n) in corner point; I=(2,3,4,5,6 ... n);

   Step 5 is calculated corner point angle sequence { θ _i}:

   

   D wherein _I-1And d _iBe respectively length sequences { d between corner point _iI-1 and i point value;

   

   Be a g _I-1To a g _I+1Euclidean distance;

   Step 6 is calculated angle difference value T:

   

   θ wherein _MiBe curve m angle sequence { θ _MiIn i angle value; θ _LiBe curve l angle sequence { θ _LiIn i angle value; M and l are two curves that mate calculating;

   Step 7 computational length difference value S:

   

   D wherein _MiBe curve m length sequences { d _MiIn the i point value; d _LiBe curve l length sequences { d _MiIn the i point value;

   

   For the length sequences of curve m and;

   

   For the length sequences of curve l and;

   Step 6: as angle difference value T＜T _ThreWith difference in length value S＜S _ThreThe time, think that two curves are complementary, otherwise think and do not match; Described T _ThreFor the angle threshold value is 5～20; S _ThreBe length threshold, value is 3～15.
2. 2. based on the Curve Matching method of Freeman chain code, it is characterized in that in the digital picture according to claim 1: described curve is that width is the digitizing two-value curve of a pixel.

Images