CN106127243A - A kind of image matching method describing son based on binaryzation SIFT - Google Patents

Abstract

The present invention relates to technical field of image processing, disclose a kind of image matching method describing son based on binaryzation SIFT.In order to enable to carry out rapidly and accurately images match, it is proposed that technical scheme.It is characterized in that: calculate 128 dimension SIFT feature points in image and describe the difference between the consecutive value of son, this difference is made comparisons with threshold value, its result two bits totally four kinds of state representation, and give the formula setting threshold value；For avoiding directly utilizing Hamming distance defection generation error result, defining a kind of distance metric method that 256 new BSIFT describe between son, the similarity carrying out feature point pairs further according to arest neighbors coupling judges.Beneficial effect: can realize images match fast and accurately, slowly and binaryzation SIFT describes the problem that sub-matching precision is the highest to solve current SIFT method matching speed, can be widely applied to the industries such as industrial detection, satellite navigation, safety monitoring.

Description

A kind of image matching method describing son based on binaryzation SIFT

Technical field

The present invention relates to computer graphic image processing technology field, describe son particularly to one based on binaryzation SIFT Image matching method.

Background technology

Image matching technology is widely used in the aspects such as image retrieval, image classification and object identification, in graphical analysis skill Art is occupied critical role.In prior art, many image matching methods apply Scale invariant special during representing image Levy conversion (Scale-invariant feature transform is called for short SIFT) feature and represent image.This method of SIFT Put forward by David Lowe the earliest, there is the robust stronger to rotation, yardstick, visual angle, brightness and smear out effect image Property.The ultimate principle of SIFT is: find characteristic point the most in the picture, then centered by characteristic point, will divide in region about Being 16 fritters, each fritter comprises 8 directions, and each direction gives a numerical value, is derived from description of 128 dimensions.

When utilizing SIFT feature to carry out images match, extract 128 dimensions of all characteristic points in two width images the most respectively Describe son, the 128 of wherein piece image dimensions are described sons take out one by one, calculate it and tie up with all the 128 of another piece image and retouch State the Euclidean distance between son, thus find the feature point pairs mated most.Owing to each image there are about thousands of SIFT spies Levy a little, and distance involves mean square root-sum square computing, therefore, for large-scale image storehouse, based on SIFT in calculating The images match of feature will be the most time-consuming.

In recent years, many methods accelerating SIFT feature coupling are suggested.These methods can be divided into two categories below:

The first kind is to describe number and the dimension of son by reducing SIFT, thus reduces the complexity of calculating.Alitappeh Et al. utilize cluster to remove similar SIFT feature point to reduce its number, when the method can reduce SIFT feature coupling Between, but cluster needs take some time, and therefore, total process time is not lowered.Ke et al. utilizes main composition to divide 128 dimension SIFT are described son and are down to lower dimension by analysis method PCA, thus reduce the characteristic matching time, but the method needs accurate in advance Standby substantial amounts of image is for training.

Equations of The Second Kind is to describe son by binaryzation SIFT, will describe son and be converted to two-value SIFT (Binary by SIFT SIFT, be called for short BSIFT) describe son, matching ratio relatively time, utilize Hamming distance (Hamming Distance) to calculate two BSIFT describes the distance between son, so, utilize bit arithmetic to substitute original mean square root-sum square computing so that coupling Amount of calculation relatively is substantially reduced.Due to BSIFT describes the Hamming distance between son and former SIFT describes between son euclidean away from From not necessarily keeping consistent, therefore, during SIFT is described sub-binaryzation, often improve with sacrifice matching accuracy rate Computational efficiency.

Summary of the invention

For solving the problem that Large Copacity images match speed is slow, matching precision is low present in prior art, the present invention carries Go out a kind of method that new SIFT describes sub-binaryzation.

A kind of image matching method describing son based on binaryzation SIFT that the present invention proposes, including step: S1: binaryzation Step；S2: images match step.

Described S1: binarization step includes:

S11: extract the SIFT feature of image pair to be matched respectively；

S12: calculate 128 dimension SIFT feature points and describe the difference between sub consecutive value；

S13: calculate each 128 dimensional feature points and describe standard deviation s of son；

S14: calculate each 128 dimensional feature points and describe the threshold value of son；

S15: 128 dimensional feature points are described the difference of son with threshold ratio relatively, carry out binarization operation.

Described S2: images match step includes:

S21: to image to be matched to carry out feature point pairs BSIFT describe son to similarity distance calculate；

S22: the similarity carrying out feature point pairs according to arest neighbors coupling judges；

S23: the image pair after output characteristic Point matching.

Wherein, the concretely comprising the following steps of described S11:

After image is normalized pretreatment, image is amplified twice pre-filtering cancelling noise；Variable dimension Gauss Convolution kernel G (x, y, σ) and input picture I (x, y) convolution obtains the metric space of image and is:

L (x, y, σ)=G (x, y, σ) * I (x, y)

$G(x,y,\mathrm{\σ})=\frac{1}{2{\mathrm{\π\σ}}^2}{e}^{-(x^2+y^2)/2{\mathrm{\σ}}^2}$

In formula, (x, y) is input picture to I, and L (x, y, σ) is the metric space of the image of definition, and G (x, y, σ) is changeable ruler Degree Gaussian convolution core, symbol * represents convolution, and (x, y) location of pixels of representative image, σ is the metric space factor；

Detect local extremum using as feature in two-dimensional image plane space and difference of Gaussian DOG metric space simultaneously Point, DOG operator is as follows:

D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ)

In formula, k is invariant；

After obtaining characteristic point, remove the low Edge Feature Points with poor stability of wherein contrast.

Wherein, the acquisition mode that described 128 dimension SIFT feature points describe son is: centered by characteristic point, by district about Territory is divided into 4 × 4 fritters, calculates the histogram of gradients in 8 directions comprised in each fritter, and each direction gives a number Value, obtains the vector of one 128 dimension, and 128 dimensions being derived from this SIFT feature point describe son.

Wherein, the concrete mode of the described difference between consecutive value is:

${\mathrm{AD}}_i=\left\{\begin{array}{cc}{D}_{i+1}-D_i& \begin{array}{cc}if& i<127\end{array}\\ D_0-D_{127}& otherwise\end{array}\right.$

Wherein, (D₀,D₁,...,D₁₂₇) represent that 128 dimension SIFT feature points describe son, AD_i(i=0,1 ..., 127) represent 128 dimension SIFT feature points describe the difference between the consecutive value of son.

Wherein, the concrete mode of the standard deviation that described calculating characteristic point describes son is:

Calculate each 128 dimensional feature points describe son average:

Calculate each 128 dimensional feature points describe son standard deviation:

Wherein, the concrete formula of threshold value T that each 128 dimensional feature points of described calculating describe son is: T=a × s+b, coefficient A, b are constant, and s is the standard deviation that corresponding 128 dimensional feature points describe son.In one preferred embodiment of the invention, a= 3.7, b=0.

Wherein, the concretely comprising the following steps of described binarization operation: with-T, 0, tri-threshold values of T number axis is divided into four sections: if Difference AD_iBe positioned at less than or equal to-T section, then corresponding description is encoded to 00；If difference AD_iIt is positioned at more than-T and less than 0 Section, then corresponding description is encoded to 01；If difference AD_iIt is positioned at more than or equal to 0 and less than T section, then corresponding description It is encoded to 10；If difference AD_iBe positioned at more than or equal to T section, then corresponding description is encoded to 11；Formula is expressed as:

$\left(b_{2*i,}{b}_{2*i+1}\right)=\left\{\begin{array}{ccc}(0,0)& if& {\mathrm{AD}}_i\&le;(-T)\\ (0,1)& if& (-T)<{\mathrm{AD}}_i<0\\ (1,0)& if& 0\&le;{\mathrm{AD}}_i<T\\ (1,1)& if& {\mathrm{AD}}_i\&GreaterEqual;T\end{array}\right.$

Thus obtaining 128 dimensions to describe the two-value numeric representation BSIFT of son and describe son, a length of 256, i.e. BSIFT goes here and there B= {b₀,b₁,...,b₂₅₅}；Wherein, (b_2*i,b_2*i+1) represent the son coding that i-th difference is corresponding.

Wherein, what described similarity distance R calculated concretely comprises the following steps: by the BSIFT of the feature point pairs of image pair to be matched String B¹,B²Respectively by 2ⁿ(0≤n≤7) are divided equally, and obtain Calculate B¹,B²Between distance, fromWithStart, compare B the most one by one¹,B²Distance between corresponding each element, if Hamming distance between the two is 0, then add 1 by enumerator counter, and otherwise enumerator counter keeps constant；Calculate institute Have 256/2ⁿAfter individual element, after being normalized by the numerical value of enumerator counter, obtain P, i.e. P ∈ [0,1], use anticosine Function arccos (P) represents that the similarity degree of feature point pairs is similarity distance R, R=arccos (P).

In one preferred embodiment of the invention, by the BSIFT string B of the feature point pairs of image pair to be matched¹,B²Respectively By 2ⁿ(0≤n≤7) are divided equally, and take n=2, will 256 be divided into 64 parts by 4, obtainCalculate B¹,B²Between distance, fromWithStart, point Compare B the most one by one¹,B²Distance between corresponding each element, if Hamming distance between the two is 0, then by enumerator Counter adds 1, and otherwise enumerator counter keeps constant；After having calculated all 64 elements, by enumerator counter divided by 64 are normalized, and are represented by normalized value P, i.e. P ∈ [0,1], use inverse cosine function arccos (P) to represent feature Point to similarity degree be similarity distance R, R=arccos (P).

Every part of reason taking 4 is: describe in son at 256 BSIFT, and every 4 reflections are that former 128 dimension SIFT describe son Some value D_iWith above numerical value D_i-1Numerical value D below_i+1Magnitude relationship；When two SIFT describe son similar if, then two Individual SIFT describes D corresponding between son_iWith D_i-1、D_iWith D_i+1The identical probability of magnitude relationship should be higher；This relevant for reflection Property, the BSIFT of a length of 256 is gone here and there by this preferred embodiment and is divided into 64 parts.

Inverse cosine function arccos (P) is used to represent similarity distance, this is because this similarity distance can be directly perceived Ground shows, works as B¹With B²When two binary strings are the most similar, enumerator counter is the biggest, and normalized value P is the biggest, similarity distance Arccos (P) is the least, otherwise, then similarity distance arccos (P) is the biggest, such result also allow for the present invention and other BSIFT method carries out experiment and compares.

Wherein, described Hamming distance is: two i.e. two-value strings of BSIFT string, and corresponding bit value is all 0 mutually, and difference is then 1, Then the number of calculating 1 is the size of Hamming distance.

The particular content of described arest neighbors coupling includes: described by the BSIFT corresponding to characteristic point A of wherein piece image Sub and the another piece image BSIFT corresponding to all characteristic points describes son and calculates similarity distance R one by one, by ascending for R row Sequence, then by minima compared with sub-minimum, if this ratio is less than preset value distratio, then it is assumed that the spy of piece image Levy the Feature Points Matching corresponding to similarity distance R minima in an A and another piece image, otherwise do not mate；Concrete formula is such as Under:

$matchornot?\left\{\begin{array}{cc}match& \begin{array}{cc}if& vals\left(1\right)<vals\left(2\right)*distratio\end{array}\\ notmatch& otherwise\end{array}\right.$

Wherein, vals (1), vals (2) are respectively the minima after pressing the ascending sequence of similarity distance R and sub-minimum, Distratio ∈ [0,1] is predetermined threshold value.In one preferred embodiment of the invention, the value of distratio is 0.84.

The image matching method difference by 128 dimension SIFT description describing son based on binaryzation SIFT that the present invention proposes Value is made comparisons with threshold value, its result two bits totally four kinds of state representation, and gives the formula setting threshold value；For keeping away Exempt to directly utilize Hamming distance defection and produce error result, thxe present method defines the distance that a kind of 256 new BSIFT describe between son Measure, it is achieved the most jumbo images match, solves that SIFT matching speed is slow and current BSIFT describes Sub-matching precision problem the most accurately, is conducive to the Rapid matching in current large capacity image data storehouse, can extensively apply In industries such as industrial detection, satellite navigation, safety monitorings, there is higher promotional value, wide market.

Accompanying drawing explanation

Fig. 1 is schematic flow sheet of the present invention；

Fig. 2 is that the image to be matched in the present embodiment is to schematic diagram；

Fig. 3 is the image pair to be matched extracting SIFT feature point in the present embodiment；

Fig. 4 is the rotary test image pair in the present embodiment；

Fig. 5 is the yardstick test image pair in the present embodiment；

Fig. 6 is the visual angle test image pair in the present embodiment；

Fig. 7 is the luminance test image pair in the present embodiment；

Fig. 8 is the fuzz testing image pair in the present embodiment；

Fig. 9 is the rotary test coupling image pair in the present embodiment；

Figure 10 is the yardstick test coupling image pair in the present embodiment；

Figure 11 is the visual angle test coupling image pair in the present embodiment；

Figure 12 is the luminance test coupling image pair in the present embodiment；；

Figure 13 is the fuzz testing coupling image pair in the present embodiment；

Figure 14 is the Euclidean distance in the present embodiment and Hamming distance comparison diagram.

Detailed description of the invention

The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention Premised on implement, give detailed embodiment and concrete operating process, but protection scope of the present invention be not limited to Following embodiment.

Embodiment 1

The present embodiment schematic flow sheet is as shown in Figure 1；Read in and there is rotationally-varying image pair to be matched, as shown in Figure 4； Extract the SIFT feature point of image pair to be matched respectively, as shown in Figure 3；Characteristic point owing to extracting is more, for purposes of illustration only, In the present embodiment, it is assumed that the left figure at coupling image pair only extracts A and B totally 2 characteristic points, and right figure only extracts a-j totally 10 spies Levy a little, as shown in Figure 2.It is as follows that the SIFT of features described above point describes son:

A:(0.0020,0,0.0098,0.0118,0.0039,0.2497,0.2497,0.0039,0.0393,0.0079, 0.0059,0.0374,0.0197,0.2281,0.2497,0.1003,0.0551,0.0216,0.0059,0.0098,0.0020, 0.0098,0.1691,0.2124,0.0098,0,0,0,0,0,0.1239,0.1318,0,0,0.1612,0.1573,0.0059, 0.0157,0.0079,0,0.0098,0.0197,0.0846,0.2497,0.0570,0.0472,0.0275,0.0059, 0.2497,0.1573,0.0551,0.0767,0.0079,0.0059,0.0157,0.0413,0.0905,0.0098,0,0,0, 0,0.0433,0.0531,0,0.0039,0.2497,0.1121,0.0059,0.0020,0,0,0.0197,0.0492, 0.0767,0.1357,0.0570,0.0256,0.0334,0.0413,0.2497,0.0570,0.0059,0.0118,0.0059, 0.0079,0.0295,0.1809,0.1495,0.0039,0,0,0,0.0020,0.0157,0.0413,0,0.0039, 0.0354,0.0216,0.0315,0.0728,0.0098,0,0.0079,0.1160,0.0924,0.0374,0.0118, 0.0039,0.0020,0.0059,0.2065,0.2497,0.0413,0.0039,0,0,0,0.0177,0.1534,0.0590, 0,0,0,0.0020,0,0.0039)；

B:(0.0667,0.0824,0,0,0,0,0,0,0.2532,0.1668,0,0,0,0,0,0,0.2650,0.1688, 0,0,0,0,0,0,0.2650,0.1138,0,0,0,0,0,0,0.0923,0.0157,0,0,0,0,0.0020,0.0020, 0.2650,0.0373,0,0,0,0,0,0.0079,0.2650,0.1138,0,0,0.0020,0,0,0.0039,0.2650, 0.1727,0,0,0,0,0,0,0.0765,0,0,0,0,0.0196,0.0294,0.0236,0.2650,0,0,0,0,0,0, 0.0805,0.1982,0.0393,0,0,0.0196,0.0137,0,0.0196,0.2650,0.1394,0,0,0,0,0,0, 0.0255,0,0,0,0,0.1629,0.1609,0.0471,0.2002,0,0,0,0.0020,0.0137,0.0432,0.2179, 0.0707,0.0118,0,0,0.0510,0.0294,0.0059,0.0294,0.2650,0.1040,0,0,0.0039,0,0, 0)；

a:(0.0197,0.0491,0.0079,0,0,0.0079,0.2673,0.1081,0.0885,0.0649, 0.0039,0.0098,0.0098,0.0256,0.2673,0.1946,0.1337,0.0334,0.0039,0.0039,0, 0.0020,0.0688,0.2025,0.0118,0.0020,0,0,0,0,0.0865,0.0747,0,0,0.0432,0.1356, 0.0059,0.0118,0.2359,0.0039,0.0138,0.0216,0.0393,0.2673,0.0727,0.0373,0.1356, 0.0157,0.2673,0.1455,0.0334,0.0590,0.0098,0.0020,0.0098,0.0393,0.0924,0.0098, 0,0,0,0,0.0236,0.0275,0,0.0020,0.1730,0.2280,0.0020,0,0,0,0.0197,0.0393, 0.0570,0.1730,0.0472,0.0177,0.0373,0.0472,0.2673,0.0668,0.0079,0.0098,0.0039, 0.0020,0.0216,0.1376,0.1337,0.0098,0,0,0,0,0.0039,0.0197,0,0.0020,0.1160, 0.0550,0.0079,0.0079,0,0,0.0039,0.0963,0.1750,0.0491,0.0020,0,0,0.0039, 0.1907,0.2673,0.0629,0.0039,0,0,0,0.0059,0.1317,0.0668,0,0,0,0,0,0)；

b:(0,0,0,0,0.0765,0.2512,0.0098,0,0.0589,0.1452,0.0039,0,0.1413, 0.2139,0,0,0.2296,0.1864,0,0,0.0039,0.0039,0,0,0.0196,0.0039,0,0,0.0059, 0.0078,0.0020,0,0,0,0,0.0098,0.1727,0.0589,0,0,0.0765,0.0157,0,0.0373,0.2630, 0.1236,0,0.0020,0.2630,0.0824,0,0.0039,0.0392,0.0078,0,0.0098,0.0569,0.0098, 0,0.0059,0.0059,0.0059,0,0,0,0,0,0.0589,0.2414,0.0373,0,0,0.0981,0.0039,0, 0.1845,0.2630,0,0,0.0118,0.2630,0.0078,0,0.0137,0.0294,0,0,0.0824,0.0530,0,0, 0.0098,0.0314,0.0078,0,0.0118,0,0,0.0608,0.2630,0.1040,0.0098,0,0.0020, 0.0922,0.0020,0.0039,0.2532,0.1177,0,0.0020,0.1374,0.2630,0,0,0.0039,0.0020, 0,0,0.1727,0.0196,0,0,0,0.0137,0.0137,0,0.0078)；

c:(0.0059,0,0,0.0236,0.2519,0.0059,0,0.0039,0.1260,0,0,0.0098,0.0689, 0.0039,0,0.0748,0.0236,0,0,0.0453,0.1122,0.0039,0.0020,0.0177,0.1397,0,0, 0.0236,0.0472,0,0,0.0531,0.0177,0.0020,0,0.0157,0.2519,0.0157,0,0.0020, 0.2401,0.0118,0,0.0059,0.0807,0.0157,0.0039,0.0433,0.0256,0,0,0.0177,0.2303, 0.0512,0.0039,0.0118,0.2519,0,0,0.0079,0.0551,0.0020,0,0.0610,0.0197,0.0039, 0,0.0039,0.2519,0.0354,0,0,0.2519,0.0394,0,0.0020,0.1043,0.0098,0.0020, 0.0079,0.0472,0.0039,0,0,0.1791,0.0709,0.1102,0.0709,0.1201,0,0.0039,0.0039, 0.0236,0.0138,0.1279,0.2519,0.0098,0.0020,0,0.0020,0.2519,0.0531,0,0,0.2519, 0.0157,0,0,0.0846,0.0157,0.0059,0.0650,0.1023,0,0,0.0217,0.0059,0.0157, 0.1535,0.2519,0,0.0039,0.0669,0.1791,0.0039,0,0.0905,0.1830)；

d:(0.0354,0,0,0.0236,0.0098,0,0,0.0393,0.0530,0.0039,0,0.0589,0.0413, 0,0.0020,0.0432,0.1611,0.0039,0,0.0039,0.0059,0.0020,0.0825,0.3144,0.0236, 0.0020,0.0118,0.0079,0.0079,0.0039,0.0688,0.3144,0.0157,0,0,0.0314,0.0648, 0.0098,0.0020,0.0236,0.1965,0.0452,0.0079,0.0668,0.0570,0,0,0.0177,0.3144, 0.1041,0,0,0,0.0020,0.0059,0.1434,0.1061,0.0236,0.0138,0.0157,0.0766,0.0806, 0.0157,0.0472,0.0039,0,0,0.0707,0.1002,0.0020,0,0.0039,0.2613,0.0039,0.0020, 0.0452,0.0707,0,0,0.1120,0.3144,0.0393,0,0,0,0,0,0.1945,0.1022,0.0648,0.0432, 0.0059,0.0098,0.0393,0.0020,0.0039,0.0039,0,0,0.0314,0.1140,0.0020,0,0.0020, 0.3144,0,0,0.0098,0.0295,0,0,0.1611,0.3144,0.0020,0,0.0039,0.0020,0,0,0.1906, 0.0393,0.0118,0.0255,0.0413,0.0059,0,0,0.0079)；

e:(0,0,0,0,0,0.1278,0.0963,0,0.3008,0.0138,0,0,0,0.0590,0.1042, 0.0629,0.3185,0.0708,0,0,0,0,0.0020,0.0374,0.0551,0.0157,0.0039,0.0197, 0.0138,0,0,0,0,0,0,0,0,0.1317,0.1180,0,0.2183,0.0079,0,0,0,0.0669,0.1239, 0.0688,0.3185,0.1258,0.0039,0,0,0,0.0079,0.0531,0.0669,0.0433,0.0433,0.0157, 0.0059,0,0,0,0,0.0393,0.0433,0,0,0.0826,0.1317,0,0.0649,0.0197,0.0059,0,0, 0.0570,0.2536,0.0924,0.3185,0.1140,0.0118,0,0,0,0.0374,0.1317,0.0452,0.0963, 0.1081,0,0,0,0,0,0.0118,0.2202,0.1829,0.0020,0,0,0.0020,0,0.0177,0.2183, 0.1789,0,0,0,0.0118,0.0157,0.1140,0.2635,0.0551,0,0,0,0.0020,0.0177,0.0472, 0.0747,0.0197,0,0,0,0.0079,0.0157)；

f:(0.1101,0.2221,0.0452,0.0079,0.0059,0.0039,0,0,0.0138,0.0747, 0.0432,0.0413,0.0668,0.0452,0.0216,0.0039,0,0,0,0.0570,0.1671,0.0786,0.0393, 0.0020,0,0,0,0.0020,0.0747,0.0845,0.0079,0,0.1258,0.2221,0.0275,0,0,0,0,0, 0.2221,0.1376,0.0138,0.0118,0.0138,0.0197,0.0668,0.1317,0.0098,0.0020,0.0039, 0.0884,0.1494,0.1867,0.1828,0.0413,0,0,0.0020,0.0531,0.1356,0.1297,0.0216,0, 0.2221,0.1376,0.0334,0,0.0059,0.0157,0,0.0039,0.2221,0.2221,0.0649,0.0295, 0.0059,0.0039,0.0059,0.0452,0.0373,0.0570,0.0570,0.1690,0.1828,0.1140,0.0629, 0.0216,0,0,0.0020,0.0531,0.0570,0.1474,0.1612,0.0255,0.1297,0.0904,0.0059,0, 0.0550,0.0884,0.0059,0.0039,0.1219,0.2221,0.0550,0,0,0,0,0.0020,0.2162, 0.1867,0.0177,0.0020,0,0.0295,0.0491,0.0275,0.0373,0.0236,0,0,0.0098,0.2083, 0.1435,0.0216)；

g:(0,0,0,0.0059,0.3143,0.0393,0,0,0.2475,0.0413,0,0.0098,0.0471, 0.0157,0,0.0020,0.3143,0.0825,0,0,0,0,0,0.0020,0.0354,0.0020,0,0,0.0157, 0.0118,0,0,0,0,0,0.0079,0.3143,0.0629,0,0,0.2200,0.0432,0,0.0020,0.0805, 0.0688,0.0020,0.0098,0.3143,0.1847,0,0,0,0,0,0.0255,0.0707,0.0236,0,0,0.0059, 0.0039,0,0.0059,0,0,0,0.0727,0.2298,0.0511,0,0,0.1670,0.0157,0.0020,0.0334, 0.0648,0.0747,0.0079,0.0511,0.3143,0.0452,0,0,0,0,0.0020,0.1807,0.0766, 0.0039,0,0,0,0,0,0.0334,0,0,0,0.0530,0.1454,0.0413,0,0,0.1454,0.0452,0.0039, 0.0452,0.0707,0.0118,0,0.0118,0.3143,0.1375,0,0,0,0,0,0.0471,0.0629,0.0157,0, 0,0,0,0,0.0059)；

h:(0.0647,0,0,0.0020,0.0275,0,0,0.0020,0.3649,0,0,0,0,0,0,0.0334, 0.2590,0.0020,0,0.0039,0.0432,0.0255,0,0.0275,0.1236,0.0059,0,0.0020,0.0059, 0.0039,0,0.0334,0.0883,0,0,0.0020,0.0294,0.0020,0,0.0020,0.3649,0.0078,0,0,0, 0,0,0.0059,0.3296,0.0118,0,0.0137,0.0471,0.0098,0,0.0039,0.1805,0.0118,0, 0.0020,0.0098,0.0059,0.0020,0.0059,0.0765,0,0,0,0.0196,0.0157,0.0039,0.0078, 0.3649,0.0275,0,0,0,0,0,0.0078,0.3178,0.0216,0,0.0078,0.0196,0.0118,0.0039, 0.0177,0.1432,0.0294,0,0,0.0039,0.0020,0,0.0255,0.0432,0.0020,0,0,0.0216, 0.0098,0.0020,0.0039,0.3649,0.0687,0,0,0,0,0,0.0020,0.2570,0.0471,0.0078, 0.0294,0.0235,0.0059,0,0.0039,0.0687,0.0883,0.0059,0.0098,0.0059,0,0,0.0020)；

i:(0.0039,0.0296,0.0512,0,0,0.0059,0.0138,0,0.0158,0.0138,0.0059,0, 0.0059,0.0414,0.0394,0.0138,0,0,0,0.0473,0.1833,0.0631,0.0197,0.0059,0.0769, 0.0118,0.0099,0.0394,0.0493,0.0039,0.0059,0.0828,0.0256,0.0197,0.0197,0, 0.0236,0.1498,0.2168,0.0749,0.2148,0.0158,0.0020,0,0.0315,0.1104,0.0847, 0.1222,0.0197,0,0,0.0099,0.2168,0.2168,0.0926,0.0788,0.0256,0,0,0,0.0296, 0.0434,0.1931,0.2168,0.0197,0.0118,0.0276,0.1399,0.1872,0.1123,0.0670,0.0453, 0.2168,0.0512,0.0099,0.1025,0.1419,0.0512,0.0059,0.0197,0.1005,0.0867,0.0177, 0.2069,0.2128,0.0552,0.0197,0.0532,0.0926,0.1931,0.0355,0.0039,0.0039,0.0217, 0.1222,0.1379,0.0138,0.0158,0.0591,0.2168,0.0591,0.0118,0.0059,0.0039,0.1379, 0.0808,0.0394,0.1773,0.0453,0.0020,0.0079,0.0099,0.0611,0.0690,0.0276,0.2049, 0.0729,0,0.0020,0.0414,0.0315,0.2168,0.0650,0.0217,0.0158,0.0039,0.0059, 0.0217)；

j:(0.0118,0.0197,0.0611,0.1202,0.0631,0.0197,0.0197,0.0079,0.1183, 0.0946,0.0552,0.0158,0.0020,0.0059,0.0197,0.0572,0.0099,0.0039,0.0059,0.0256, 0.0138,0.0118,0.0217,0.0158,0,0.0079,0.0729,0.0335,0.0020,0,0.0020,0,0.0256, 0.0059,0.0079,0.0355,0.1084,0.1202,0.1616,0.1222,0.2247,0.0256,0.0020,0.0158, 0.0729,0.0690,0.0690,0.1912,0.0394,0.0020,0.0039,0.1853,0.2247,0.0374,0.0414, 0.0828,0.0138,0.0079,0.0670,0.0808,0.0177,0.0414,0.1715,0.1380,0.0256,0.0138, 0.0335,0.0434,0.2247,0.0335,0.0217,0.1064,0.2247,0.0887,0.0039,0.0512,0.2247, 0.0217,0.0059,0.0276,0.2247,0.0493,0.0020,0.1340,0.2247,0.0039,0.0020,0.0769, 0.2109,0.0611,0.0079,0.0118,0.0197,0.0453,0.0887,0.1774,0.0020,0.0276,0.1419, 0.0374,0.0572,0.0099,0.0020,0.0020,0.1064,0.0453,0.0611,0.0276,0.0907,0.0177, 0.0079,0.0355,0.0769,0.1084,0.0276,0.0335,0.2247,0.0493,0.0020,0,0.0710, 0.2247,0.1656,0.0374,0.0118,0.0158,0.0079,0.0020)；

Respectively son is described 128 dimensions of characteristic point a-j that right figure in characteristic point A of left figure extraction, B and Fig. 2 in Fig. 2 extracts Calculate the difference between the consecutive value describing son, corresponding standard deviation, threshold value, be then described sub-binaryzation.Set forth below It is the difference between the consecutive value of 128 dimension SIFT feature points description:

A:(-0.0020,0.0098,0.0020,-0.0079,0.2458,0,-0.2458,0.0354,-0.0315,- 0.0020,0.0315,-0.0177,0.2084,0.0216,-0.1495,-0.0452,-0.0334,-0.0157,0.0039,- 0.0079,0.0079,0.1593,0.0433,-0.2025,-0.0098,0,0,0,0,0.1239,0.0079,-0.1318,0, 0.1612,-0.0039,-0.1514,0.0098,-0.0079,-0.0079,0.0098,0.0098,0.0649,0.1652,- 0.1927,-0.0098,-0.0197,-0.0216,0.2438,-0.0924,-0.1023,0.0216,-0.0688,-0.0020, 0.0098,0.0256,0.0492,-0.0806,-0.0098,0,0,0,0.0433,0.0098,-0.0531,0.0039, 0.2458,-0.1377,-0.1062,-0.0039,-0.0020,0,0.0197,0.0295,0.0275,0.0590,- 0.0787,-0.0315,0.0079,0.0079,0.2084,-0.1927,-0.0511,0.0059,-0.0059,0.0020, 0.0216,0.1514,-0.0315,-0.1455,-0.0039,0,0,0.0020,0.0138,0.0256,-0.0413, 0.0039,0.0315,-0.0138,0.0098,0.0413,-0.0629,-0.0098,0.0079,0.1082,-0.0236,- 0.0551,-0.0256,-0.0079,-0.0020,0.0039,0.2006,0.0433,-0.2084,-0.0374,-0.0039, 0,0,0.0177,0.1357,-0.0944,-0.0590,0,0,0.0020,-0.0020,0.0039,-0.0019)；

B:(0.0157,-0.0824,0,0,0,0,0,0.2532,-0.0864,-0.1668,0,0,0,0,0,0.2650,- 0.0962,-0.1688,0,0,0,0,0,0.2650,-0.1511,-0.1138,0,0,0,0,0,0.0923,-0.0765,- 0.0157,0,0,0,0.0020,0,0.2630,-0.2277,-0.0373,0,0,0,0,0.0079,0.2571,-0.1511,- 0.1138,0,0.0020,-0.0020,0,0.0039,0.2611,-0.0923,-0.1727,0,0,0,0,0,0.0765,- 0.0765,0,0,0,0.0196,0.0098,-0.0059,0.2414,-0.2650,0,0,0,0,0,0.0805,0.1178,- 0.1590,-0.0393,0,0.0196,-0.0059,-0.0137,0.0196,0.2453,-0.1256,-0.1394,0,0,0, 0,0,0.0255,-0.0255,0,0,0,0.1629,-0.0020,-0.1138,0.1531,-0.2002,0,0,0.0020, 0.0118,0.0294,0.1747,-0.1472,-0.0589,-0.0118,0,0.0510,-0.0216,-0.0236,0.0236, 0.2355,-0.1609,-0.1040,0,0.0039,-0.0039,0,0,0.0667)；

a:(0.0295,-0.0413,-0.0079,0,0.0079,0.2595,-0.1592,-0.0197,-0.0236,- 0.0609,0.0059,0,0.0157,0.2418,-0.0727,-0.0609,-0.1003,-0.0295,0,-0.0039, 0.0020,0.0668,0.1337,-0.1907,-0.0098,-0.0020,0,0,0,0.0865,-0.0118,-0.0747,0, 0.0432,0.0924,-0.1297,0.0059,0.2241,-0.2320,0.0098,0.0079,0.0177,0.2280,- 0.1946,-0.0354,0.0983,-0.1199,0.2516,-0.1219,-0.1120,0.0256,-0.0491,-0.0079, 0.0079,0.0295,0.0531,-0.0826,-0.0098,0,0,0,0.0236,0.0039,-0.0275,0.0020, 0.1710,0.0550,-0.2261,-0.0020,0,0,0.0197,0.0197,0.0177,0.1160,-0.1258,- 0.0295,0.0197,0.0098,0.2202,-0.2005,-0.0590,0.0020,-0.0059,-0.0020,0.0197, 0.1160,-0.0039,-0.1238,-0.0098,0,0,0,0.0039,0.0157,-0.0197,0.0020,0.1140,- 0.0609,-0.0472,0,-0.0079,0,0.0039,0.0924,0.0786,-0.1258,-0.0472,-0.0020,0, 0.0039,0.1867,0.0767,-0.2044,-0.0590,-0.0039,0,0,0.0059,0.1258,-0.0649,- 0.0668,0,0,0,0,0,0.0197)；

b:(0,0,0,0.0765,0.1747,-0.2414,-0.0098,0.0589,0.0863,-0.1413,-0.0039, 0.1413,0.0726,-0.2139,0,0.2296,-0.0432,-0.1864,0,0.0039,0,-0.0039,0,0.0196,- 0.0157,-0.0039,0,0.0059,0.0020,-0.0059,-0.0020,0,0,0,0.0098,0.1629,-0.1138,- 0.0589,0,0.0765,-0.0608,-0.0157,0.0373,0.2257,-0.1393,-0.1236,0.0020,0.2610,- 0.1805,-0.0824,0.0039,0.0353,-0.0314,-0.0078,0.0098,0.0471,-0.0471,-0.0098, 0.0059,0,0,-0.0059,0,0,0,0,0.0589,0.1825,-0.2041,-0.0373,0,0.0981,-0.0942,- 0.0039,0.1845,0.0785,-0.2630,0,0.0118,0.2512,-0.2551,-0.0078,0.0137,0.0157,- 0.0294,0,0.0824,-0.0294,-0.0530,0,0.0098,0.0216,-0.0235,-0.0078,0.0118,- 0.0118,0,0.0608,0.2021,-0.1590,-0.0942,-0.0098,0.0020,0.0903,-0.0903,0.0020, 0.2492,-0.1354,-0.1177,0.0020,0.1354,0.1256,-0.2630,0,0.0039,-0.0020,-0.0020, 0,0.1727,-0.1531,-0.0196,0,0,0.0137,0,-0.0137,0.0078,-0.0078)；

c:(-0.0059,0,0.0236,0.2283,-0.2460,-0.0059,0.0039,0.1220,-0.1260,0, 0.0098,0.0590,-0.0650,-0.0039,0.0748,-0.0512,-0.0236,0,0.0453,0.0669,- 0.1083,-0.0020,0.0157,0.1220,-0.1397,0,0.0236,0.0236,-0.0472,0,0.0531,- 0.0354,-0.0157,-0.0020,0.0157,0.2362,-0.2362,-0.0157,0.0020,0.2382,-0.2283,- 0.0118,0.0059,0.0748,-0.0650,-0.0118,0.0394,-0.0177,-0.0256,0,0.0177,0.2126,- 0.1791,-0.0472,0.0079,0.2401,-0.2519,0,0.0079,0.0472,-0.0531,-0.0020,0.0610,- 0.0413,-0.0157,-0.0039,0.0039,0.2480,-0.2165,-0.0354,0,0.2519,-0.2126,- 0.0394,0.0020,0.1023,-0.0945,-0.0079,0.0059,0.0394,-0.0433,-0.0039,0,0.1791,- 0.1083,0.0394,-0.0394,0.0492,-0.1201,0.0039,0,0.0197,-0.0098,0.1142,0.1240,- 0.2421,-0.0079,-0.0020,0.0020,0.2500,-0.1988,-0.0531,0,0.2519,-0.2362,- 0.0157,0,0.0846,-0.0689,-0.0098,0.0590,0.0374,-0.1023,0,0.0217,-0.0157, 0.0098,0.1378,0.0984,-0.2519,0.0039,0.0630,0.1122,-0.1752,-0.0039,0.0905, 0.0925,-0.1771)；

d:(-0.0354,0,0.0236,-0.0138,-0.0098,0,0.0393,0.0138,-0.0491,-0.0039, 0.0589,-0.0177,-0.0413,0.0020,0.0413,0.1179,-0.1572,-0.0039,0.0039,0.0020,- 0.0039,0.0806,0.2318,-0.2908,-0.0216,0.0098,-0.0039,0,-0.0039,0.0648,0.2456,- 0.2986,-0.0157,0,0.0314,0.0334,-0.0550,-0.0079,0.0216,0.1729,-0.1513,-0.0373, 0.0589,-0.0098,-0.0570,0,0.0177,0.2967,-0.2102,-0.1041,0,0,0.0020,0.0039, 0.1375,-0.0373,-0.0825,-0.0098,0.0020,0.0609,0.0039,-0.0648,0.0314,-0.0432,- 0.0039,0,0.0707,0.0295,-0.0982,-0.0020,0.0039,0.2574,-0.2574,-0.0020,0.0432, 0.0255,-0.0707,0,0.1120,0.2024,-0.2751,-0.0393,0,0,0,0,0.1945,-0.0923,- 0.0373,-0.0216,-0.0373,0.0039,0.0295,-0.0373,0.0020,0,-0.0039,0,0.0314, 0.0825,-0.1120,-0.0020,0.0020,0.3124,-0.3124,0,0.0098,0.0196,-0.0295,0, 0.1611,0.1533,-0.3124,-0.0020,0.0039,-0.0020,-0.0020,0,0.1906,-0.1513,- 0.0275,0.0138,0.0157,-0.0354,-0.0059,0,0.0079,0.0275)；

e:(0,0,0,0,0.1278,-0.0315,-0.0963,0.3008,-0.2871,-0.0138,0,0,0.0590, 0.0452,-0.0413,0.2556,-0.2477,-0.0708,0,0,0,0.0020,0.0354,0.0177,-0.0393,- 0.0118,0.0157,-0.0059,-0.0138,0,0,0,0,0,0,0,0.1317,-0.0138,-0.1180,0.2183,- 0.2104,-0.0079,0,0,0.0669,0.0570,-0.0551,0.2497,-0.1927,-0.1219,-0.0039,0,0, 0.0079,0.0452,0.0138,-0.0236,0,-0.0275,-0.0098,-0.0059,0,0,0,0.0393,0.0039,- 0.0433,0,0.0826,0.0492,-0.1317,0.0649,-0.0452,-0.0138,-0.0059,0,0.0570, 0.1966,-0.1612,0.2261,-0.2045,-0.1022,-0.0118,0,0,0.0374,0.0944,-0.0865, 0.0511,0.0118,-0.1081,0,0,0,0,0.0118,0.2084,-0.0374,-0.1809,-0.0020,0, 0.0020,-0.0020,0.0177,0.2006,-0.0393,-0.1789,0,0,0.0118,0.0039,0.0983, 0.1494,-0.2084,-0.0551,0,0,0.0020,0.0157,0.0295,0.0275,-0.0551,-0.0197,0,0, 0.0079,0.0079,-0.0157)；

f:(0.1120,-0.1769,-0.0373,-0.0020,-0.0020,-0.0039,0,0.0138,0.0609,- 0.0314,-0.0020,0.0255,-0.0216,-0.0236,-0.0177,-0.0039,0,0,0.0570,0.1101,- 0.0884,-0.0393,-0.0373,-0.0020,0,0,0.0020,0.0727,0.0098,-0.0766,-0.0079, 0.1258,0.0963,-0.1946,-0.0275,0,0,0,0,0.2221,-0.0845,-0.1238,-0.0020,0.0020, 0.0059,0.0472,0.0649,-0.1219,-0.0079,0.0020,0.0845,0.0609,0.0373,-0.0039,- 0.1415,-0.0413,0,0.0020,0.0511,0.0825,-0.0059,-0.1081,-0.0216,0.2221,- 0.0845,-0.1042,-0.0334,0.0059,0.0098,-0.0157,0.0039,0.2182,0,-0.1572,- 0.0354,-0.0236,-0.0020,0.0020,0.0393,-0.0079,0.0197,0,0.1120,0.0138,-0.0688,- 0.0511,-0.0413,-0.0216,0,0.0020,0.0511,0.0039,0.0904,0.0138,-0.1356,0.1042,- 0.0393,-0.0845,-0.0059,0.0550,0.0334,-0.0825,-0.0020,0.1179,0.1002,-0.1671,- 0.0550,0,0,0,0.0020,0.2142,-0.0295,-0.1690,-0.0157,-0.0020,0.0295,0.0197,- 0.0216,0.0098,-0.0138,-0.0236,0,0.0098,0.1985,-0.0649,-0.1219,0.0884)；

g:(0,0,0.0059,0.3084,-0.2750,-0.0393,0,0.2475,-0.2063,-0.0413,0.0098, 0.0373,-0.0314,-0.0157,0.0020,0.3124,-0.2318,-0.0825,0,0,0,0,0.0020,0.0334,- 0.0334,-0.0020,0,0.0157,-0.0039,-0.0118,0,0,0,0,0.0079,0.3065,-0.2515,- 0.0629,0,0.2200,-0.1768,-0.0432,0.0020,0.0786,-0.0118,-0.0668,0.0079,0.3045,- 0.1297,-0.1847,0,0,0,0,0.0255,0.0452,-0.0471,-0.0236,0,0.0059,-0.0020,- 0.0039,0.0059,-0.0059,0,0,0.0727,0.1572,-0.1788,-0.0511,0,0.1670,-0.1513,- 0.0138,0.0314,0.0314,0.0098,-0.0668,0.0432,0.2632,-0.2691,-0.0452,0,0,0, 0.0020,0.1788,-0.1041,-0.0727,-0.0039,0,0,0,0,0.0334,-0.0334,0,0,0.0530, 0.0923,-0.1041,-0.0413,0,0.1454,-0.1002,-0.0413,0.0413,0.0255,-0.0589,- 0.0118,0.0188,0.3025,-0.1768,-0.1375,0,0,0,0,0.0471,0.0157,-0.0471,-0.0157,0, 0,0,0,0.0059,-0.0059)；

h:(-0.0647,0,0.0020,0.0255,-0.0275,0,0.0020,0.3630,-0.3649,0,0,0,0,0, 0.0334,0.2256,-0.2570,-0.0020,0.0039,0.0392,-0.0177,-0.0255,0.0275,0.0961,- 0.1177,-0.0059,0.0020,0.0039,-0.0020,-0.0039,0.0334,0.0549,-0.0883,0,0.0020, 0.0275,-0.0275,-0.0020,0.0020,0.3630,-0.3571,-0.0078,0,0,0,0,0.0059,0.3237,- 0.3178,-0.0118,0.0137,0.0334,-0.0373,-0.0098,0.0039,0.1766,-0.1687,-0.0118, 0.0020,0.0078,-0.0039,-0.0039,0.0039,0.0706,-0.0765,0,0,0.0196,-0.0039,- 0.0118,0.0039,0.3571,-0.3375,-0.0275,0,0,0,0,0.0078,0.3100,-0.2963,-0.0216, 0.0078,0.0118,-0.0078,-0.0078,0.0137,0.1256,-0.1138,-0.0294,0,0.0039,- 0.0020,-0.0020,0.0255,0.0177,-0.0412,-0.0020,0,0.0216,-0.0118,-0.0078,0.0020, 0.3610,-0.2963,-0.0687,0,0,0,0,0.0020,0.2551,-0.2099,-0.0392,0.0216,-0.0059,- 0.0177,-0.0059,0.0039,0.0647,0.0196,-0.0824,0.0039,-0.0039,-0.0059,0,0.0020, 0.0628)；

i:(0.0256,0.0217,-0.0512,0,0.0059,0.0079,-0.0138,0.0158,-0.0020,- 0.0079,-0.0059,0.0059,0.0355,-0.0020,-0.0256,-0.0138,0,0,0.0473,0.1360,- 0.1202,-0.0434,-0.0138,0.0709,-0.0650,-0.0020,0.0296,0.0099,-0.0453,0.0020, 0.0769,-0.0571,-0.0059,0,-0.0197,0.0236,0.1261,0.0670,-0.1419,0.1399,- 0.1990,-0.0138,-0.0020,0.0315,0.0788,-0.0256,0.0374,-0.1025,-0.0197,0,0.0099, 0.2069,0,-0.1241,-0.0138,-0.0532,-0.0256,0,0,0.0296,0.0138,0.1498,0.0236,- 0.1971,-0.0079,0.0158,0.1123,0.0473,-0.0749,-0.0453,-0.0217,0.1714,-0.1655,- 0.0414,0.0926,0.0394,-0.0906,-0.0453,0.0138,0.0808,-0.0138,-0.0690,0.1892, 0.0059,-0.1576,-0.0355,0.0355,0.0394,0.1005,-0.1576,-0.0315,0,0.0177,0.1005, 0.0158,-0.1241,0.0020,0.0434,0.1576,-0.1576,-0.0473,-0.0059,-0.0020,0.1340,- 0.0571,-0.0414,0.1379,-0.1320,-0.0434,0.0059,0.0020,0.0512,0.0079,-0.0414, 0.1773,-0.1320,-0.0729,0.0020,0.0394,-0.0099,0.1852,-0.1517,-0.0434,-0.0059,- 0.0118,0.0020,0.0158,-0.0177)；

j:(0.0079,0.0414,0.0591,-0.0572,-0.0434,0,-0.0118,0.1104,-0.0237,- 0.0394,-0.0394,-0.0138,0.0039,0.0138,0.0374,-0.0473,-0.0059,0.0020,0.0197,- 0.0118,-0.0020,0.0099,-0.0059,-0.0158,0.0079,0.0650,-0.0394,-0.0315,-0.0020, 0.0020,-0.0020,0.0256,-0.0197,0.0020,0.0276,0.0729,0.0118,0.0414,-0.0394, 0.1025,-0.1991,-0.0237,0.0138,0.0572,-0.0039,0,0.1222,-0.1518,-0.0374,0.0020, 0.1813,0.0394,-0.1872,0.0039,0.0414,-0.0690,-0.0059,0.0591,0.0138,-0.0631, 0.0237,0.1301,-0.0335,-0.1123,-0.0118,0.0197,0.0099,0.1813,-0.1912,-0.0118, 0.0847,0.1183,-0.1360,-0.0847,0.0473,0.1734,-0.2030,-0.0158,0.0217,0.1971,- 0.1754,-0.0473,0.1320,0.0907,-0.2207,-0.0020,0.0749,0.1340,-0.1498,-0.0532, 0.0039,0.0079,0.0256,0.0434,0.0887,-0.1754,0.0256,0.1143,-0.1045,0.0197,- 0.0473,-0.0079,0,0.1045,-0.0611,0.0158,-0.0335,0.0631,-0.0729,-0.0099,0.0276, 0.0414,0.0315,-0.0808,0.0059,0.1912,-0.1754,-0.0473,-0.0020,0.0710,0.1537,- 0.0591,-0.1281,-0.0256,0.0039,-0.0079,-0.0059,0.0099)；

Characteristic point a-that in Fig. 2, in characteristic point A of left figure extraction, B and Fig. 2, right figure extracts is tried to achieve according to mean value computation formula The 128 dimensional feature points of j describe the average of sonIt is respectively as follows: 0.0499,0.0411,0.0491,0.0435,0.0474,0.0439, 0.0457,0.0558,0.0407,0.0341,0.0579,0.0590；

According to standard deviation computing formula try to achieve characteristic point A, B, and a-j 128 dimension describe son standard deviation s be respectively as follows: 0.0729,0.0783,0.0735,0.0769,0.0746,0.0767,0.0756,0.0686,0.0784,0.0815,0.0668, 0.0658；

Calculate 128 dimensional feature points and describe threshold value T=3.7*s+0 of son, obtain threshold value and be respectively as follows: 0.2699,0.2895, 0.2720,0.2847,0.2761,0.2838,0.2799,0.2538,0.2902,0.3017,0.2470,0.2435；

According to threshold value T 128 dimensions to characteristic point a-j that right figure in characteristic point A of left figure extraction, B and Fig. 2 in Fig. 2 extracts Characteristic point describes son and carries out binaryzation respectively, obtains following binaryzation result:

A:(01 10 10 01 10 10 01 10 01 01 10 01 10 10 01 01 01 01 10 01 10 10 10 01 01 10 10 10 10 10 10 01 10 10 01 01 10 01 01 10 10 10 10 01 01 01 01 10 01 01 10 01 01 10 1010 01 01 10 10 10 10 10 01 10 10 01 01 01 01 10 10 10 10 10 01 01 10 10 10 01 01 10 01 10 10 10 01 01 01 10 10 10 10 10 01 10 10 01 10 10 01 01 10 10 01 01 01 01 01 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 01 10 01)；

B:(10 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 01 10 10 10 01 01 10 10 10 10 10 10 01 10 10 10 10 10 01 10 01 10 10 10 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 10 10 10 01 10 10 10 10 01 01 10 01 10 10 10 10 10 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 10)；

a:(10 01 01 10 10 10 01 01 01 01 10 10 10 10 01 01 01 01 10 01 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 01 10 10 01 10 10 10 10 01 01 10 01 10 01 01 10 01 01 10 10 10 01 01 10 10 10 10 10 01 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 01 01 10 01 01 10 10 01 01 01 10 10 10 10 10 01 10 10 01 01 10 01 10 10 10 10 01 01 01 10 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 10 10 10)；

b:(10 10 10 10 10 01 01 10 10 01 01 10 10 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 10 01 01 10 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 01 10 10 10 10 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 10 01 10 10 01 01 10 10 10 01 01 10 01 10 10 10 01 01 01 10 10 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 01 01 10 10 10 10 01 10 01)；

c:(01 10 10 10 01 01 10 10 01 10 10 10 01 01 10 01 01 10 10 10 01 01 10 10 01 10 10 10 01 10 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 01 01 10 10 10 01 01 10 10 01 10 10 10 01 01 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 01 10 01 10 10 10 01 10 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 01 10 10 10 01 10 10 10 01 01 10 10 01)；

d:(01 10 10 01 01 10 10 10 01 01 10 01 01 10 10 10 01 01 10 10 01 10 10 00 01 10 01 10 01 10 10 00 01 10 10 10 01 01 10 10 01 01 10 01 01 10 10 11 01 01 10 10 10 10 10 01 01 01 10 10 10 01 10 01 01 10 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 10 10 10 10 01 01 01 01 10 10 01 10 10 01 10 10 10 01 01 10 11 00 10 10 10 01 10 10 10 00 01 10 01 01 10 10 01 01 10 10 01 01 10 10 10)；

e:(10 10 10 10 10 01 01 11 00 01 10 10 10 10 01 10 01 01 10 10 10 10 10 10 01 01 10 01 01 10 10 10 10 10 10 10 10 01 01 10 01 01 10 10 10 10 01 10 01 01 01 10 10 10 10 10 01 10 01 01 01 10 10 10 10 10 01 10 10 10 01 10 01 01 01 10 10 10 01 10 01 01 01 10 10 10 10 01 10 10 01 10 10 10 10 10 10 01 01 01 10 10 01 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 01)；

f:(10 01 01 01 01 01 10 10 10 01 01 10 01 01 01 01 10 10 10 10 01 01 01 01 10 10 10 10 10 01 01 10 10 01 01 10 10 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 10 01 01 01 10 10 10 10 01 01 01 10 01 01 01 10 10 01 10 10 10 01 01 01 01 10 10 01 10 10 10 10 01 01 01 01 10 10 10 10 10 10 01 10 01 01 01 10 10 01 01 10 10 01 01 10 10 10 10 10 01 01 01 01 10 10 01 10 01 01 10 10 10 01 01 10)；

g:(10 10 10 11 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 10 01 01 10 10 10 10 10 11 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 10 01 01 10 01 10 10 10 10 01 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 10 10 10 01 01 01 10 10 10 10 10 01 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 01)；

h:(01 10 10 10 01 10 10 11 00 10 10 10 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 11 00 01 10 10 10 10 10 11 00 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 11 00 01 10 10 10 10 10 11 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 01 01 01 10 10 10 01 10 01 01 10 10 10)；

i:(10 10 01 10 10 10 01 10 01 01 01 10 10 01 01 01 10 10 10 10 01 01 01 10 01 01 10 10 01 10 10 01 01 10 01 10 10 10 01 10 01 01 01 10 10 01 10 01 01 10 10 10 10 01 01 01 01 10 10 10 10 10 10 01 01 10 10 10 01 01 01 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 01 01 10 10 10 10 01 10 10 10 01 01 01 01 10 01 01 10 01 01 10 10 10 10 01 10 01 01 10 10 01 10 01 01 01 01 10 10 01)；

j:(10 10 10 01 01 10 01 10 01 01 01 01 10 10 10 01 01 10 10 01 01 10 01 01 10 10 01 01 01 10 01 10 01 10 10 10 10 10 01 10 01 01 10 10 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 01 10 10 01 01 01 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 10 10 01 10 10 01 10 01 01 10 10 01 10 01 10 01 01 10 10 10 01 10 10 01 01 01 10 10 01 01 01 10 01 01 10)；

Take n=2, image to be matched is described substring B to the BSIFT of characteristic point A Yu a～j¹,B²Divide equally by 4 respectively, will 256 are divided into 64 parts by 4, and correspondence position is carried out XOR, obtain following result:

(A,a):(1111 1111 0000 0011 0000 0011 0000 0000 0000 0000 0000 0000 0011 0000 0000 1100 0000 1100 0011 0000 0000 0000 0011 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 1100 0011 0000 0000 0000 0000 0000 0000 0000 1100 00000000 0000 0000 0000 0000 0011 0000 1100 0011 0000 0011 0000 0000 0000 0000 0000 0000 0000 0011 0011)；

(A,b):(1100 0011 0011 0000 1100 1111 0011 1111 0000 0011 0011 0011 0011 0000 0011 1111 0000 1111 1100 1100 1111 0011 0000 1100 0000 0011 0011 0000 0000 0000 0011 0011 0000 1111 0000 0000 1111 0011 0000 0000 0000 0011 1100 0000 0011 0000 1111 0000 0000 1111 1100 1100 1111 1100 0011 0000 1111 1100 1100 0011 0011 0000 0000 0000)；

(A,c):(0000 0011 1111 1100 0011 0011 1111 1100 0011 0011 1111 0011 0000 0000 1100 0000 1111 1111 1100 1100 1111 0011 0000 1111 0011 0011 0011 0000 0011 0000 1111 0000 1111 1111 0000 0000 1111 0011 0011 0000 0000 0011 1100 1111 0011 0000 1100 0000 1111 1100 1100 1100 1100 1111 0000 0000 1111 1100 0000 0011 1111 0011 1111 0000)；

(A,d):(0000 0000 1100 1100 0000 0000 1100 1111 0000 0011 1100 0001 0000 1100 1100 0001 1100 1111 1100 1100 1111 0000 0011 1101 0000 0011 1100 0011 0000 0000 0011 0000 1100 1111 0000 0000 1111 0011 0000 0000 0000 0011 0000 0000 0000 1100 0011 0011 1100 1100 1100 1101 1011 1111 0011 0000 1000 1100 1100 0011 0011 0011 1111 0011)；

(A,e):(1100 0011 0011 0001 0100 0011 0000 0011 0000 0011 0000 0011 0011 0011 1100 0011 0000 1111 0000 0000 1111 0011 1111 0000 0000 1111 1100 0000 0011 1111 1100 0011 0000 0011 1111 1100 1111 1111 1100 1100 0000 1111 0000 0000 1111 1100 0000 0011 0011 0011 0011 0000 0000 0011 1111 0000 0000 0011 0000 0000 1100 1100 0011 0000)；

(A,f):(1111 1100 1111 1100 1100 1111 1111 0000 1111 0011 1111 1100 1100 0000 0011 1111 0011 0011 0011 1100 1111 1111 1111 1111 0011 0011 1111 1111 1111 0000 1111 1111 1111 0011 1100 0000 0011 1100 0000 0011 1111 0011 1111 1100 1111 0000 0000 1111 1111 0000 0000 0000 0000 0011 1111 0000 1100 0000 0000 1100 0000 0000 0000 1111)；

(A,g):(1100 0010 1111 1100 0000 0011 1111 1110 0000 0011 0000 0011 0011 0000 1111 0011 0000 1110 1100 1100 1111 0011 0000 1101 0000 0011 1100 0000 0000 0000 1111 0000 0000 1111 0000 0000 1111 0011 1111 0000 0000 0011 0000 0000 0000 0000 0000 0000 0000 1100 1100 1100 1100 1111 0000 0001 1100 1111 0000 0000 0000 0000 0011 0000)；

(A,h):(0000 0011 1100 1101 0111 0011 0000 1111 0000 0011 1111 0011 0011 0000 1111 0011 1100 1111 1100 1101 1011 0011 1111 1101 0100 0011 0011 0000 0000 0000 1111 0011 1100 1111 0000 0001 1011 0011 1100 0001 0000 0011 1111 0011 0000 0000 1111 0011 1111 1100 1100 1101 1100 1111 1111 0000 1100 1100 1111 0000 1100 0011 1111 0011)；

(A,i):(1100 1111 0000 0000 0000 1111 0011 0000 1111 0011 1111 1111 0011 0000 1100 0000 1100 0011 0011 0000 1111 1111 1100 1111 0011 0011 1111 1111 0011 0000 0000 0000 1100 1111 0000 1100 1111 0011 0011 0000 0000 0011 1111 0011 1100 1100 0000 0000 0000 1111 1100 0000 1100 1100 0011 0000 0000 1100 1100 0011 1100 1111 1111 0000)；

(A,j):(1100 0000 1100 0000 0000 1100 0000 1100 0011 0000 1100 1100 1100 1111 1100 1111 1100 1111 0011 0000 1111 0011 0011 1111 0011 0011 0000 0011 0011 0011 0000 1100 1100 1111 0000 0000 1111 0011 0011 0000 0000 0011 1111 0011 0000 0000 0000 0000 0000 0000 1100 1100 1111 0011 0000 0000 0000 1111 1111 1100 1100 1111 0000 1111)；

By the above results by calculate every 41 number, i.e. calculate Hamming distance, respectively obtain following result:

(A,a):(4 4 0 2 0 2 0 0 0 0 0 0 2 0 0 2 0 2 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 2 2 0 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 2 2 0 2 0 0 0 0 0 0 0 2 2)；

(A,b):(2 2 2 0 2 4 2 4 0 2 2 2 2 0 2 4 0 4 2 2 4 2 0 2 0 2 2 0 0 0 2 2 0 4 0 0 4 2 0 0 0 2 2 0 2 0 4 0 0 4 2 2 4 2 2 0 4 2 2 2 2 0 0 0)；

(A,c):(0 2 4 2 2 2 4 2 2 2 4 2 0 0 2 0 4 4 2 2 4 2 0 4 2 2 2 0 2 0 4 0 4 4 0 0 4 2 2 0 0 2 2 4 2 0 2 0 4 2 2 2 2 4 0 0 4 2 0 2 4 2 4 0)；

(A,d):(0 0 2 2 0 0 2 4 0 2 2 1 0 2 2 1 2 4 2 2 4 0 2 3 0 2 2 2 0 0 2 0 2 4 0 0 4 2 0 0 0 2 0 0 0 2 2 2 2 2 2 3 3 4 2 0 1 2 2 2 2 2 4 2)；

(A,e):(2 2 2 1 1 2 0 2 0 2 0 2 2 2 2 2 0 4 0 0 4 2 4 0 0 4 2 0 2 4 2 2 0 2 4 2 4 4 2 2 0 4 0 0 4 2 0 2 2 2 2 0 0 2 4 0 0 2 0 0 2 2 2 0)；

(A,f):(4 2 4 2 2 4 4 0 4 2 4 2 2 0 2 4 2 2 2 2 4 4 4 4 2 2 4 4 4 0 4 4 4 2 2 0 2 2 0 2 4 2 4 2 4 0 0 4 4 0 0 0 0 2 4 0 2 0 0 2 0 0 0 4)；

(A,g):(2 1 4 2 0 2 4 3 0 2 0 2 2 0 4 2 0 3 2 2 4 2 0 3 0 2 2 0 0 0 4 0 0 4 0 0 4 2 4 0 0 2 0 0 0 0 0 0 0 2 2 2 2 4 0 1 2 4 0 0 0 0 2 0)；

(A,h):(0 2 2 3 3 2 0 4 0 2 4 2 2 0 4 2 2 4 2 3 3 2 4 3 1 2 2 0 0 0 4 2 2 4 0 1 3 2 2 1 0 2 4 2 0 0 4 2 4 2 2 3 2 4 4 0 2 2 4 0 2 2 4 2)；

(A,i):(2 4 0 0 0 4 2 0 4 2 4 4 2 0 2 0 2 2 2 0 4 4 2 4 2 2 4 4 2 0 0 0 2 4 0 2 4 2 2 0 0 2 4 2 2 2 0 0 0 4 2 0 2 2 2 0 0 2 2 2 2 4 4 0)；

(A,j):(2 0 2 0 0 2 0 2 2 0 2 2 2 4 2 4 2 4 2 0 4 2 2 4 2 2 0 2 2 2 0 2 2 4 0 0 4 2 2 0 0 2 4 2 0 0 0 0 0 0 2 2 4 2 0 0 0 4 4 2 2 4 0 4)；

Obtain: the counter value between characteristic point A and another 10 characteristic points a-j is respectively as follows: 46,23,18,20,21,17, 29,13,20,22；P value between characteristic point A and another 10 characteristic points a-j is respectively as follows: 0.7188,0.3594,0.2813, 0.3125,0.3281,0.2656,0.4531,0.2031,0.3125,0.3438；Between characteristic point A and another 10 characteristic points a-j Arccos (P) value be respectively as follows: 0.7687,1.2032,1.2856,1.2530,1.2365,1.3020,1.1006,1.3663, 1.2530,1.2198；The ascending sequence of arccos (P) value between characteristic point A and another 10 characteristic points a-j is respectively as follows: 0.7687,1.1006,1.2032,1.2198,1.2365,1.2530,1.2530,1.2856,1.3020,1.3663； Distratio is set to 0.84, because 0.7687 < 1.1006*distratio, and the spy of arccos (P) minima 0.7687 correspondence Levying is some a, therefore, as in figure 2 it is shown, characteristic point a that characteristic point A that left figure extracts is extracted with right figure is mated, uses thin white threads bar Connection features point A and characteristic point a, represent that feature point pairs A with a mates.

Take n=2, image is described substring B to the BSIFT of characteristic point B Yu a-j¹,B²Divide equally by 4 respectively, will 256 press 4 are divided into 64 parts, correspondence position carries out XOR, and by the above results by the number every 4 calculating 1, i.e. calculates the Chinese Prescribed distance, respectively obtains following result:

(B,a):(0 2 0 4 0 0 0 4 0 2 0 2 0 0 0 4 4 2 0 2 4 2 2 2 0 2 0 0 0 0 0 2 2 2 2 2 2 2 2 0 0 2 2 2 0 0 0 2 2 4 0 2 2 4 2 2 2 4 4 0 0 0 2 0)；

(B,b):(2 0 2 2 2 2 2 0 0 0 2 0 0 0 2 2 4 0 4 0 0 0 4 0 0 0 2 0 0 0 2 0 2 0 4 2 2 0 2 0 0 0 2 2 2 0 4 2 2 2 2 2 0 2 2 2 2 2 2 2 2 0 4 2)；

(B,c):(4 0 4 0 2 0 4 2 2 0 4 0 2 0 2 2 0 0 4 0 0 0 4 2 2 0 2 0 2 0 4 2 2 0 4 2 2 0 4 0 0 0 2 2 2 0 2 2 2 0 2 2 2 0 4 2 2 2 4 2 4 2 0 2)；

(B,d):(4 2 2 0 0 2 2 0 0 0 2 1 2 2 2 1 2 0 4 0 0 2 2 1 0 0 2 2 0 0 2 2 0 0 4 2 2 0 2 0 0 0 4 2 0 2 2 0 0 0 2 3 1 0 2 2 1 2 2 2 2 2 0 0)；

(B,e):(2 0 2 3 1 0 0 2 0 0 0 0 0 2 2 0 4 0 2 2 0 0 0 2 0 2 2 0 2 4 2 0 2 2 0 0 2 2 0 2 0 2 4 2 4 2 0 0 4 4 2 0 4 2 0 2 2 2 4 0 2 2 2 2)；

(B,f):(0 4 4 0 2 2 4 4 4 0 4 4 4 0 2 2 2 2 0 0 0 2 0 2 2 0 4 4 4 0 4 2 2 2 2 2 4 4 2 2 4 0 0 4 4 0 0 2 2 2 0 0 4 2 0 2 0 4 4 2 0 0 4 2)；

(B,g):(2 1 4 0 0 0 4 1 0 0 0 0 0 0 4 0 4 1 4 0 0 0 4 1 0 0 2 0 0 0 4 2 2 0 4 2 2 0 2 0 0 0 4 2 0 0 0 2 2 0 2 2 2 0 4 1 0 0 4 0 0 0 2 2)；

(B,h):(4 0 2 1 3 0 0 0 0 0 4 0 0 0 4 0 2 0 4 1 1 0 0 1 1 0 2 0 0 0 4 0 0 0 4 3 3 0 0 1 0 0 0 0 0 0 4 0 2 0 2 3 2 0 0 2 0 2 0 0 2 2 0 0)；

(B,i):(2 2 0 2 0 2 2 4 4 0 4 2 0 0 2 2 2 2 0 2 0 2 2 2 2 0 4 4 2 0 0 2 0 0 4 0 2 0 4 0 0 0 0 0 2 2 0 2 2 2 2 0 2 2 2 2 2 2 2 2 2 4 0 2)；

(B,j):(2 2 2 2 0 4 0 2 2 2 2 4 4 4 2 2 2 0 0 2 0 0 2 2 2 0 0 2 2 2 0 4 0 0 4 2 2 0 4 0 0 0 0 0 0 0 0 2 2 2 2 2 0 2 4 2 2 0 0 2 2 4 4 2)；

It can thus be appreciated that: the counter value between characteristic point B and another 10 characteristic points a-j is respectively as follows: 27,26,23,25, 25,20,33,37,22,23；P value between characteristic point B and another 10 characteristic points a-j is respectively as follows: 0.4219,0.4063, 0.3594,0.3906,0.3906,0.3125,0.5156,0.5781,0.3438,0.3594；Characteristic point B and another 10 characteristic points Arccos (P) value between a-j is respectively as follows: 1.1353,1.1524,1.2032,1.1695,1.1695,1.2530,1.0291, 0.9544,1.2198,1.2032；The ascending sequence of arccos (P) value between characteristic point B and another 10 characteristic points a-j divides It is not: 0.9544,1.0291,1.1353,1.1524,1.1695,1.1695,1.2032,1.2032,1.2198,1.2530； Distratio is set to 0.84, because 0.9544 > 1.0291*distratio, characteristic point B is not mated with characteristic point a-j.

Embodiment 2

The present embodiment idiographic flow is as shown in Figure 1.As in figure 2 it is shown, for have rotationally-varying image pair to be matched, extract The SIFT feature point A of left figure, 128 dimension SIFT feature of characteristic point A describe son (D₀,D₁,...,D₁₂₇) it is:

(0.0020,0,0.0098,0.0118,0.0039,0.2497,0.2497,0.0039,0.0393,0.0079, 0.0059,0.0374,0.0197,0.2281,0.2497,0.1003,0.0551,0.0216,0.0059,0.0098,0.0020, 0.0098,0.1691,0.2124,0.0098,0,0,0,0,0,0.1239,0.1318,0,0,0.1612,0.1573,0.0059, 0.0157,0.0079,0,0.0098,0.0197,0.0846,0.2497,0.0570,0.0472,0.0275,0.0059, 0.2497,0.1573,0.0551,0.0767,0.0079,0.0059,0.0157,0.0413,0.0905,0.0098,0,0,0, 0,0.0433,0.0531,0,0.0039,0.2497,0.1121,0.0059,0.0020,0,0,0.0197,0.0492, 0.0767,0.1357,0.0570,0.0256,0.0334,0.0413,0.2497,0.0570,0.0059,0.0118,0.0059, 0.0079,0.0295,0.1809,0.1495,0.0039,0,0,0,0.0020,0.0157,0.0413,0,0.0039, 0.0354,0.0216,0.0315,0.0728,0.0098,0,0.0079,0.1160,0.0924,0.0374,0.0118, 0.0039,0.0020,0.0059,0.2065,0.2497,0.0413,0.0039,0,0,0,0.0177,0.1534,0.0590, 0,0,0,0.0020,0,0.0039)；

Obtain 128 dimension SIFT feature points and describe difference AD between sub consecutive value_i(i=0,2 ..., 127), and according to average Computing formulaTry to achieve 128 dimensional feature points describe son average: According to standard deviation computing formula Trying to achieve 128 dimensional feature point A and describing sub standard deviation s is 0.0729；Calculate 128 dimensional feature points and describe the threshold value T=a × s+b of son =3.7 × s+0=0.2699；Thus obtain 256 BSIFT of characteristic point A and describe son be:

(01 10 10 01 10 10 01 10 01 01 10 01 10 10 01 01 01 01 10 01 10 10 10 01 01 10 10 10 10 10 10 01 10 10 01 01 10 01 01 10 10 10 10 01 01 01 01 10 01 01 10 01 01 10 1010 01 01 10 10 10 10 10 01 10 10 01 01 01 01 10 10 10 10 10 01 01 10 10 10 01 01 10 01 10 10 10 01 01 01 10 10 10 10 10 01 10 10 01 10 10 01 01 10 10 01 01 01 01 01 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 01 10 01)；

In like manner, extract such as the SIFT feature point a of figure right in Fig. 2；256 BSIFT of characteristic point a describe son and are: (10 01 01 10 10 10 01 01 01 01 10 10 10 10 01 01 01 01 10 01 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 01 10 10 01 10 10 10 10 01 01 10 01 10 01 01 10 01 01 10 10 10 01 01 10 10 10 10 10 01 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 01 01 10 01 01 10 10 01 01 01 10 10 10 10 10 01 10 10 01 01 10 01 10 10 10 10 01 01 01 10 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 10 10 10)；

Taking n=2, by image pair to be matched, the BSIFT of characteristic point A and characteristic point a being positioned at right figure that are positioned at left figure retouches State substring B¹, B²Divide equally by 4 respectively, will 256 be divided into 64 parts by 4, correspondence position is carried out XOR, obtains following Result:

(1111 1111 0000 0011 0000 0011 0000 0000 0000 0000 0000 0000 0011 0000 0000 1100 0000 1100 0011 0000 0000 0000 0011 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 1100 0011 0000 0000 0000 0000 0000 0000 0000 1100 0000 0000 0000 0000 0000 0000 0011 0000 1100 0011 0000 0011 0000 0000 0000 0000 0000 0000 0000 0011 0011)；

By the above results by calculate every 41 number, i.e. calculate Hamming distance, obtain following result:

(4 4 0 2 0 2 0 0 0 0 0 0 2 0 0 2 0 2 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 2 2 0 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 2 2 0 2 0 0 0 0 0 0 0 2 2)；

In the above results, have 46 0, thus obtain, counter=46；Then P=counter/64=46/64= 0.7188；R=arccos (P)=arcos (0.7188)=0.7687.

Seek in the left figure of Fig. 2 the similarity distance R between a-j totally 10 characteristic points in characteristic point B and right figure respectively, and by similar Property the ascending sequence of distance R, then according to arest neighbors coupling carry out feature point pairs similarity judge.Left figure characteristic point B and It is as follows that the BSIFT of 10 characteristic points a-j of right figure describes son:

B:(10 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 01 10 10 10 01 01 10 10 10 10 10 10 01 10 10 10 10 10 01 10 01 10 10 10 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 10 10 10 01 10 10 10 10 01 01 10 01 10 10 10 10 10 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 10)；

a:(10 01 01 10 10 10 01 01 01 01 10 10 10 10 01 01 01 01 10 01 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 01 10 10 01 10 10 10 10 01 01 10 01 10 01 01 10 01 01 10 10 10 01 01 10 10 10 10 10 01 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 01 01 10 01 01 10 10 01 01 01 10 10 10 10 10 01 10 10 01 01 10 01 10 10 10 10 01 01 01 10 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 10 10 10)；

b:(10 10 10 10 10 01 01 10 10 01 01 10 10 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 10 01 01 10 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 01 10 10 10 10 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 10 01 10 10 01 01 10 10 10 01 01 10 01 10 10 10 01 01 01 10 10 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 01 01 10 10 10 10 01 10 01)；

c:(01 10 10 10 01 01 10 10 01 10 10 10 01 01 10 01 01 10 10 10 01 01 10 10 01 10 10 10 01 10 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 01 01 10 10 10 01 01 10 10 01 10 10 10 01 01 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 01 10 01 10 10 10 01 10 10 01 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 01 10 10 10 01 10 10 10 01 01 10 10 01)；

d:(01 10 10 01 01 10 10 10 01 01 10 01 01 10 10 10 01 01 10 10 01 10 10 00 01 10 01 10 01 10 10 00 01 10 10 10 01 01 10 10 01 01 10 01 01 10 10 11 01 01 10 10 10 10 10 01 01 01 10 10 10 01 10 01 01 10 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 10 10 10 10 01 01 01 01 10 10 01 10 10 01 10 10 10 01 01 10 11 00 10 10 10 01 10 10 10 00 01 10 01 01 10 10 01 01 10 10 01 01 10 10 10)；

e:(10 10 10 10 10 01 01 11 00 01 10 10 10 10 01 10 01 01 10 10 10 10 10 10 01 01 10 01 01 10 10 10 10 10 10 10 10 01 01 10 01 01 10 10 10 10 01 10 01 01 01 10 10 10 10 10 01 10 01 01 01 10 10 10 10 10 01 10 10 10 01 10 01 01 01 10 10 10 01 10 01 01 01 10 10 10 10 01 10 10 01 10 10 10 10 10 10 01 01 01 10 10 01 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 10 01 01 10 10 10 10 01)；

f:(10 01 01 01 01 01 10 10 10 01 01 10 01 01 01 01 10 10 10 10 01 01 01 01 10 10 10 10 10 01 01 10 10 01 01 10 10 10 10 10 01 01 01 10 10 10 10 01 01 10 10 10 10 01 01 01 10 10 10 10 01 01 01 10 01 01 01 10 10 01 10 10 10 01 01 01 01 10 10 01 10 10 10 10 01 01 01 01 10 10 10 10 10 10 01 10 01 01 01 10 10 01 01 10 10 01 01 10 10 10 10 10 01 01 01 01 10 10 01 10 01 01 10 10 10 01 01 10)；

g:(10 10 10 11 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 10 01 01 10 10 10 10 10 11 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 10 01 01 10 01 10 10 10 10 01 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 10 10 10 01 01 01 10 10 10 10 10 01 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 10 10 10 10 01)；

h:(01 10 10 10 01 10 10 11 00 10 10 10 10 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 11 00 01 10 10 10 10 10 11 00 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 10 10 10 01 01 10 11 00 01 10 10 10 10 10 11 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 11 01 01 10 10 10 10 10 10 01 01 10 01 01 01 10 10 10 01 10 01 01 10 10 10)；

I:(10 10 01 10 10 10 01 10 01 01 01 10 10 01 01 01 10 10 10 10 01 01 01 10 01 01 10 10 01 10 10 01 01 10 01 10 10 10 01 10 01 01 01 10 10 01 10 01 01 10 10 10 10 01 01 01 01 10 10 10 10 10 10 01 01 10 10 10 01 01 01 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 01 01 10 10 10 10 01 10 10 10 01 01 01 01 10 01 01 10 01 01 10 10 10 10 01 10 01 01 10 10 01 10 01 01 01 01 10 10 01)；

j:(10 10 10 01 01 10 01 10 01 01 01 01 10 10 10 01 01 10 10 01 01 10 01 01 10 10 01 01 01 10 01 10 01 10 10 10 10 10 01 10 01 01 10 10 01 10 10 01 01 10 10 10 01 10 10 01 01 10 10 01 10 10 01 01 01 10 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 01 01 10 10 10 10 10 01 10 10 01 10 01 01 10 10 01 10 01 10 01 01 10 10 10 01 10 10 01 01 01 10 10 01 01 01 10 01 01 10)；

Obtain: the counter value between characteristic point B and another 10 characteristic points a-j is respectively as follows: 27,26,23,25,25,20, 33,37,22,23；P value between characteristic point B and another 10 characteristic points a-j is respectively as follows: 0.4219,0.4063,0.3594, 0.3906,0.3906,0.3125,0.5156,0.5781,0.3438,0.3594；Between characteristic point B and another 10 characteristic points a-j Arccos (P) value be respectively as follows: 1.1353,1.1524,1.2032,1.1695,1.1695,1.2530,1.0291,0.9544, 1.2198,1.2032；The ascending sequence of arccos (P) value between characteristic point B and another 10 characteristic points a-j is respectively as follows: 0.9544,1.0291,1.1353,1.1524,1.1695,1.1695,1.2032,1.2032,1.2198,1.2530； Distratio is set to 0.84 because 0.9544 > 1.0291*distratio, so, in fig. 2, characteristic point B in left figure with Characteristic point a-j in right figure is not mated.

Seek in the left figure of Fig. 2 the similarity distance R between characteristic point a-j totally 10 characteristic points in characteristic point A and right figure respectively, and By similarity distance R by ascending sequence, then carry out the similarity judgement of feature point pairs according to arest neighbors coupling.Obtain: Counter value between characteristic point A and another 10 characteristic points a-j is respectively as follows: 46,23,18,20,21,17,29,13,20,22； P value between characteristic point A and another 10 characteristic points a-j is respectively as follows: 0.7188,0.3594,0.2813,0.3125,0.3281, 0.2656,0.4531,0.2031,0.3125,0.3438；Arccos (P) value between characteristic point A and another 10 characteristic points a-j It is respectively as follows: 0.7687,1.2032,1.2856,1.2530,1.2365,1.3020,1.1006,1.3663,1.2530,1.2198； The ascending sequence of arccos (P) value between characteristic point A and another 10 characteristic points a-j is respectively as follows: 0.7687,1.1006, 1.2032,1.2198,1.2365,1.2530,1.2530,1.2856,1.3020,1.3663；Distratio is set to 0.84, because of It is 0.7687 < 1.1006*distratio, and arccos (P) minima 0.7687 characteristic of correspondence point is a, so, at Fig. 2 In, characteristic point A in left figure is mated with characteristic point a in right figure, represents A and a of coupling in Fig. 2 with white lines.From Fig. 2 It can be seen that characteristic point A in left figure lays respectively at rabbit ears and position, head boundary in figure with characteristic point a in right figure, Therefore, feature point pairs A and a is correct matching double points.

For checking, the present invention proposes the matching performance of method, we choose from experiment sample 5 groups have rotation, yardstick, The coupling image of visual angle, brightness and smear out effect, to mating, is then used the present invention to carry out images match by computer, and Feature point pairs line to coupling, wherein, represents correct coupling and erroneous matching respectively with thin white threads bar and heavy black line bar Feature point pairs, see Fig. 4-Figure 13, wherein, Fig. 4 is rotary test image pair；Fig. 5 is that yardstick tests image pair；Fig. 6 is visual angle Test image pair；Fig. 7 is luminance test image pair；Fig. 8 is fuzz testing image pair；Fig. 9 is that rotary test mates image pair；Figure 10 is yardstick test coupling image pair；Figure 11 is visual angle test coupling image pair；Figure 12 is that luminance test mates image pair；Figure 13 It is that fuzz testing mates image pair.

For further illustrating the effect of the present invention, now the present invention is compared with following three kinds of methods.Table 1 is D.G.Lowe.Distinctive image features from scale-invariant keypoints[J] .International Journal of Computer Vision, the SIFT method in 2004,60 (2): 91-110. mono-literary compositions The experimental data (hereinafter referred to as SIFT method) obtained；Table 2 is Chun-Che Chen, Shang-Lin Hsieh, Using binarization and hashing for efficient SIFT matching[J].Journal of Visual Communication and Image Representation, the binarization method in 30 (2015) 86 93. one literary compositions combines this The similarity of invention judges the experimental data (hereinafter referred to as Chen ' method) obtained；Table 3 is Wengang Zhou, Houqiang Li.BSIFT:Toward data-independent codebook for large scale image search[J] .IEEE Transactions on Image Processing, the binarization method in 2015,24 (03): 967 979. one literary compositions The experimental data (hereinafter referred to as Zhou ' method) obtained in conjunction with the similarity judgement of this method；Table 4 is for utilize the present invention to obtain Experimental data.Images match result is as follows:

Table 1:

Table 2:

Table 3:

Table 4:

Table 1-table 4 lists SIFT method, Chen ' method, Zhou ' method and the present invention respectively for Fig. 4, Fig. 5, Fig. 6, figure The SIFT feature extracted in 7 and Fig. 8 is counted out, the number of matching characteristic point pair, the number of error matching points pair, matching accuracy rate With experimental results such as process times, wherein, the computing formula of matching accuracy rate is:

The number of matching accuracy rate=(number-error matching points of matching characteristic point pair is to number)/matching characteristic point pair

Data from table 1-table 4, it can be seen that the present invention is compared to SIFT method, are finding the same of more features point pair Time, there has been obvious reduction the process time；It is essentially identical that the present invention and Zhou ' method process the time, but slightly above Chen ' method, Reason is that two valued description that the present invention and Zhou ' method generate is 256, and Chen ' method only has 128；At figure 5, in Fig. 7 and Fig. 8, the matching accuracy rate of the present invention is all high than SIFT method, in Fig. 4 and Fig. 6, and the present invention and SIFT method Matching accuracy rate is close；In addition in Fig. 7, the matching accuracy rate of the present invention is slightly below Zhou ' method, all in all, the present invention's Matching accuracy rate is above Zhou ' method and Chen ' method.The above-mentioned experimental result explanation present invention solves SIFT method coupling Speed is slow and current binaryzation SIFT describes the problem that sub-matching precision is the highest.

It is said that in general, use Euclidean distance (Euclidean Distance) to calculate 128 dimensions of SIFT feature point Describe the distance between son, and for the distance described between son after binaryzation, then use Hamming distance, for retouching after making binaryzation State son and still keep original 128 separating capacities stronger when tieing up, it is often desired to Hamming distance keeps consistent with Euclidean distance Property, i.e. it is the biggest that 128 dimensions describe Euclidean distance between son, and the Hamming distance that the BSIFT of its correspondence describes son is the biggest, otherwise As the same.For proving that the present invention proposes the effectiveness of threshold formula, we test the Hamming between its feature point pairs to great amount of images Distance and Euclidean distance, the Hamming distance that the binarization method of the present invention obtains and Euclidean distance keep one substantially Cause.As shown in figure 14, for the comparison diagram of the Hamming distance that obtained by binarization method in the present embodiment with Euclidean distance, Both are basic keeps consistent.As can be seen here, the threshold formula that the present invention proposes can preferably keep Hamming distance and euclidean Concordance between Ju Li, the binaryzation that thus the checking present invention obtains describes son and still keeps original 128 dimensions to describe what son was had Stronger separating capacity.

Being described in detail the specific embodiment of the present invention above, but it is intended only as example, the present invention does not limit It is formed on particular embodiments described above.To those skilled in the art, any equivalent modifications that the present invention is carried out and Substitute the most all among scope of the invention.Therefore, the impartial conversion made without departing from the spirit and scope of the invention and Amendment, all should contain within the scope of the invention.

Claims (10)

1. the image matching method describing son based on binaryzation SIFT, it is characterised in that include step:

S1: binarization step；S2: images match step；Described S1: binarization step includes:

S11: extract the SIFT feature of image pair to be matched respectively；

S12: calculate 128 dimension SIFT feature points and describe son；

S13: calculate 128 dimension SIFT feature points and describe the difference between sub consecutive value；

S14: calculate each 128 dimensional feature points and describe standard deviation s of son；

S15: calculate each 128 dimensional feature points and describe the threshold value of son；

S16: 128 dimensional feature points are described the difference of son with threshold ratio relatively, carry out binarization operation；

Described S2: images match step includes:

S21: to image to be matched to carry out feature point pairs BSIFT describe son to similarity distance calculate；

S22: the similarity carrying out feature point pairs according to arest neighbors coupling judges；

S23: the image pair after output characteristic Point matching.

A kind of method the most according to claim 1, it is characterised in that described S11 concretely comprises the following steps: return image After one changes pretreatment, image is amplified twice pre-filtering cancelling noise；Variable dimension Gaussian convolution core G (x, y, σ) and input Image I (x, y) convolution obtains the metric space of image and is:

L (x, y, σ)=G (x, y, σ) * I (x, y)

$G(x,y,\mathrm{\&sigma;})=\frac{1}{2{\mathrm{\&pi;\&sigma;}}^2}{e}^{-(x^2+y^2)/2{\mathrm{\&sigma;}}^2}$

In formula, (x, y) is input picture to I, and L (x, y, σ) is the metric space of the image of definition, and G (x, y, σ) is that variable dimension is high This convolution kernel, symbol * represents convolution, and (x, y) location of pixels of representative image, σ is the metric space factor；

Detect local extremum using as characteristic point, DOG in two-dimensional image plane space and difference of Gaussian DOG metric space simultaneously Operator is as follows:

D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ)

In formula, k is invariant；

After obtaining characteristic point, remove the low Edge Feature Points with poor stability of wherein contrast.

A kind of method the most according to claim 1, it is characterised in that described 128 dimension SIFT feature points describe the acquisition of son Mode is: centered by characteristic point, will be divided into 4 × 4 fritters in region about, and calculate 8 directions comprised in each fritter Histogram of gradients, each direction gives a numerical value, obtain one 128 dimension vector, be derived from this SIFT feature point 128 dimensions describe son.

A kind of method the most according to claim 1, it is characterised in that the concrete mode of the described difference between consecutive value For:

${\mathrm{AD}}_i=\left\{\begin{array}{cc}{D}_{i+1}-D_i& \begin{array}{cc}if& i<127\end{array}\\ D_0-D_{127}& otherwise\end{array}\right.$

Wherein, (D₀,D₁,...,D₁₂₇) represent that 128 dimension SIFT feature points describe son, AD_i(i=0,1 ..., 127) represent 128 dimensions SIFT feature point describes the difference between the consecutive value of son.

A kind of method the most according to claim 1, it is characterised in that described calculating characteristic point describes the standard deviation of son Concrete mode is:

Calculate each 128 dimensional feature points describe son average:

Calculate each 128 dimensional feature points describe son standard deviation:

A kind of method the most according to claim 1, it is characterised in that each 128 dimensional feature points of described calculating describe son The concrete formula of threshold value T is: T=a × s+b, and wherein, coefficient a, b are constant, and s is the mark that corresponding 128 dimensional feature points describe son Accurate poor.

A kind of method the most according to claim 1, it is characterised in that concretely comprising the following steps of described binarization operation:

With-T, 0, tri-threshold values of T number axis is divided into four sections: if difference AD_iIt is positioned at less than or equal to-T section, then corresponding retouches State son and be encoded to 00；If difference AD_iBe positioned at more than-T and less than 0 section, then corresponding description is encoded to 01；If difference AD_i Be positioned at more than or equal to 0 and less than T section, then corresponding description is encoded to 10；If difference AD_iIt is positioned at more than or equal to T section, Then corresponding description is encoded to 11；Formula is expressed as:

$\left(b_{2*i,}{b}_{2*i+1}\right)=\left\{\begin{array}{ccc}(0,0)& if& {\mathrm{AD}}_i\&le;(-T)\\ (0,1)& if& (-T)<{\mathrm{AD}}_i<0\\ (1,0)& if& 0\&le;{\mathrm{AD}}_i<T\\ (1,1)& if& {\mathrm{AD}}_i\&GreaterEqual;T\end{array}\right.$

Thus obtaining 128 dimensions to describe the two-value numeric representation BSIFT of son and describe son, a length of 256, i.e. BSIFT goes here and there B={b₀, b₁,...,b₂₅₅}；Wherein, (b_2*i,b_2*i+1) represent i-th difference AD_iCorresponding coding.

A kind of method the most according to claim 1, it is characterised in that what described similarity distance R calculated concretely comprises the following steps: By the BSIFT string B of the feature point pairs of image pair to be matched¹,B²Respectively by 2ⁿ(0≤n≤7) are divided equally, and obtainCalculate B¹,B²Between distance, fromWithOpen Begin, compare BSIFT string B the most one by one¹,B²Distance between corresponding position, if Hamming distance between the two is 0, then will counting Device counter adds 1, and otherwise enumerator counter keeps constant；Calculate all 256/2ⁿAfter individual element, by enumerator The numerical value of counter obtains P after being normalized, i.e. P ∈ [0,1], uses inverse cosine function arccos (P) to represent characteristic point To similarity degree be similarity distance R, R=arccos (P).

A kind of method the most according to claim 8, it is characterised in that described Hamming distance is: two BSIFT strings, corresponding Bit value is all 0 mutually, and difference is then 1, and then the number of calculating 1 is the size of Hamming distance.

A kind of method the most according to claim 1, it is characterised in that the particular content of described arest neighbors coupling includes: will Wherein the BSIFT corresponding to characteristic point A of piece image describes the sub and another piece image BSIFT corresponding to all characteristic points Describe son and calculate similarity distance R one by one, by ascending for R sequence, then by minima compared with sub-minimum, if this ratio is less than pre- First setting value distratio, then it is assumed that in characteristic point A of piece image and another piece image corresponding to similarity distance R minima Feature Points Matching, otherwise do not mate；Concrete formula is as follows:

$matchornot?\left\{\begin{array}{cc}match& \begin{array}{cc}if& vals\left(1\right)<vals\left(2\right)*distratio\end{array}\\ notmatch& otherwise\end{array}\right.$

Wherein, vals (1), vals (2) are respectively the minima after pressing the ascending sequence of similarity distance R and sub-minimum, Distratio ∈ [0,1] is predetermined threshold value.