CN103927511B - image identification method based on difference feature description - Google Patents
image identification method based on difference feature description Download PDFInfo
- Publication number
- CN103927511B CN103927511B CN201410063894.9A CN201410063894A CN103927511B CN 103927511 B CN103927511 B CN 103927511B CN 201410063894 A CN201410063894 A CN 201410063894A CN 103927511 B CN103927511 B CN 103927511B
- Authority
- CN
- China
- Prior art keywords
- image
- target image
- pixel
- region
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
Disclosed is an image identification method based on difference feature description. The method includes the steps that firstly, the gray values of image pixel points are converted into the subordinate degree through a selected subordinating degree function; on the basis of defining a difference value concept, each set of adjacent pixel points, with the same difference value, of an image serves as an area, and feature vectors of the area are defined; each area is represented through a feature vector point at the mass center; the mass centers of the adjacent areas are connected through a triangle, the inner corners of the triangle serve as the relation angle of the adjacent areas, and features of a target image are extracted; finally, the features of an image to be identified are extracted, target feature matching is carried out on the processed image, and then whether the image contains the target image or not is determined. The difference values and direction of the image area feature points are combined, the novel image identification method is provided, and the image identification method is simple, convenient to implement, easy to understand, complete in feature and wide in application range, and has the better combination property compared with a traditional method.
Description
Technical field
The present invention relates to a kind of carry out recognition target image using image difference as feature description and with this structural features amount
Method, belongs to technical field of data processing.
Background technology
Image recognition is the heat subject of picture research field, and it is based on the principal character of image, using calculating
Machine is processed to image, analyzed and is understood, to identify the target of various different modes and the technology of object.Under normal circumstances,
The target recognition of image, also known as the pattern recognition with regard to visual pattern it is intended to utilize the reason of image procossing and area of pattern recognition
By and method, determine and in image, whether there is target interested, if there is being then that target gives rational explanation, if necessary
Its position is also predefined.
Image recognition, as one of computer vision important field of research, is widely used to military and civilian many
Individual aspect, such as biomedicine, satellite remote sensing, robot vision, freight detection, target following, Autonomous Vehicles navigation, public security, bank,
Traffic, ecommerce and multimedia network communication etc..In security monitoring field, the analyzing and processing to goal description information, permissible
Apply in industrial product quality inspection, IC design and graphic designs it is also possible to be used in weather forecast, forest
The field such as fire and Geological Hazards Monitoring, air pollution forecasting.Picture can be locked in speech by human face detection tech in time
On the person, thus largely reduce the image transmitting ratio of video conferencing.Motor-driven in virtual reality, calculating
The applications such as picture, video notes and commentary, the target identification technology of image equally also plays irreplaceable effect.
At present, image recognition technology mainly has：Fingerprint recognition, recognition of face, Text region etc..For in image domains
Various particular problems, the biology barrier scheme that image recognition is adopted is all otherwise varied, needs particular problem is entered
Row concrete analysis.
The premise of image recognition is iamge description, with the correlation of each target in numeral or symbol table diagram picture or scenery
Relation between feature, or even target, finally gives target characteristic and the abstract expression of the relation between them.How to extract
Characteristics of image and then construct the success or not that suitable description determines images match identification.The retouching of image local feature region
Stating is one of study hotspot of computer vision and area of pattern recognition in recent years, in image registration, three-dimensional reconstruction, image inspection
All play an important role in the concrete applications such as rope, image mosaic, robot localization, object tracking and identification.
In the evolution of image recognition, there are four classes representational theoretical and method：Statistical picture recognition methodss, structure
Image-recognizing method, fuzzy image recognition method and the recognition methodss based on artificial intelligence.Statistical picture recognition methodss mainly collect
In architectural feature be have ignored on the statistical relationship of quantity；Structural images recognition methodss are then mainly with the structure of image and interior
Mutual relation between portion's subdivision is as basis of characterization, but when solving some complicated problem of image recognition, equally has
Limitation；Fuzzy Pattern Recognition has the characteristics that Information Pull is more abundant, identifies that good stability, inferential capability are strong, but simultaneously, such as
Where setting up more rational membership function in Fuzzy Pattern Recognition Method is the problem needing to solve further；Artificial intelligence
Recognition methodss have its unique advantage when solving the problems, such as some complicated image steganalysis, but there is also computation complexity
Greatly, physical significance is difficult to resolve and releases, the problems such as algorithm is not sufficiently stable.
From the point of view of the existing method of actual implementation, image recognition can be divided into calculus of finite differences, optical flow method, template matching method, feature again
Matching method, method based on Wavelet Transform with based on neutral net etc..A kind of individually method often pluses and minuses depositing, no
The requirements at the higher level to image recognition for the people can be met, current people more pay close attention to and multiple methods combine to reach more preferable identification
Effect.
It is a kind of fairly simple also the more commonly used image-recognizing method based on the identification of gradation of image matrix statistical nature,
This kind of method is processed calculating accordingly for gray matrix, to obtain the statistical nature information with regard to image, by feature
Select and adjust its weight coefficient to reconstruct required various features data it is established that the exclusive characteristic information of each image indexes
, the dynamic cataloging identification retrieval of picture material is realized further according to fuzzy clustering principle.
It will usually first coloured image is converted into gray level image in image processing process, although after so simplifying
Processing procedure, but also lost some information of image script, be likely to result in error.Under normal circumstances, colouring information
It is an important evidence of image recognition, by the color characteristic of statistical picture, the object of different colours can be identified,
Based on the hybrid coding method of color component, the coding result of target image and reference picture is compared as a kind of, finds it
Between linear relationship complete multicolor model identification.Certain methods are also had to be that from rgb space, coloured image is transformed into HSI sky
Between, more respectively three components are analyzed, it is identified based on morphology scheduling algorithm and classify.
Additionally, rectangular histogram method for expressing simplifies the picture element matrix of image it is possible to obtain probability distribution situation, have more
The representativeness of statistics.Same rectangular histogram can be divided into grey level histogram and two kinds of color histogram.By different types of information
The iamge description of the complexity of composition solve before some problems present in image recognition：Mix before identification and identify in the background
The object not gone out；Object not seen before is classified.From this research obtain one total it was concluded that with made in the past
The rectangular histogram of relatively low dimension is compared, the multiple clues that can be shown using the clue rectangular histogram of the complexity of more higher-dimension composition
Common situation of change, and have better image recognition performance.
Gray value and grey level histogram all can only represent the Global Information of image pixel, and the relation between neighbor pixel
Cannot describe.Textural characteristics are the concentrated expression of grey-level statistics, space distribution information and the structural information of image.It is by
The set with definite shape and size of pixel composition, is all intrinsic characteristic of nearly all imaging surface.As in image
Hold one of study hotspot of retrieval, the image-recognizing method species being currently based on textural characteristics is very many.The most frequently used surely belongs to ash
Degree co-occurrence matrix method, it to reflect the Global Information such as the relevant direction of gradation of image and interval with the probability in statistics.But
It is that gray level co-occurrence matrixes are generally basede on spatial domain picture, the description to textural characteristics is not careful.
Object of different shapes is also possible to identical color or gray value, so shape recognition is also essential
's.Identification for picture structure shape is broadly divided into based on boundary profile and is based on region two big class method, than more typical
Means have not bending moment, transform domain, geometric parameter model etc..
As classical alternative approach, various different squares, Fourier describe son, and small echo describes son, morphologic description exists
Carried out substantial amounts of research in 20 years of past.Bending moment is not the statistical property of image, meets translation, stretches, rotates all not
The invariance becoming, is now widely used for many aspects such as target recognition, scenery coupling, shape analysis and character recognition.Base
In Radon and Fourier-Mellin conversion yardstick and rotational invariance it is proposed that a kind of new will obscure, flexible, rotation,
The moment function that translation invariance combines is used as image recognition.The moment function obtaining in this way and moment invariants are to noise
There is robustness, the global information of image and the fixed information under conditions of scaling, translation and rotation, can be passed through this
Individual single descriptor shows.Bending moment does not have superiority, but to closing and open structure, due to can not directly calculate
The feature of square, thus also need to first structure realm, and because the calculating of this square is related to all pictures on intra-zone and border
Element, thus the time expended is more.
In the application of a lot of computer visions, in order to improve accuracy rate and improve the robustness to noise through frequently with many points
Resolution analysis method.Because wavelet transformation provides multi-resolution representation, therefore application is wide.
The method of computational geometry intermediate cam subdivision is introduced form fit and retrieval by YiTao and William I. Grosky
In problem, first the delaunay triangulation with regard to its characteristic point is made to shape, obtain some trianglees；Secondly calculate these three
The angular histogram of angular sequence carries out similarity coupling as shape facility, and the difference between two objects is with its angle Nogata
Euclidean distance between figure is measuring.The method weak point is as the triangle number play increasing subdivision generation of characteristic point
Increase, leverage the efficiency of algorithm.Its innovatory algorithm is first Feature Points to be ranked up, only in characteristic point sequence
Carry out triangle subdivision, thus having saved the algorithm time in little Convex range.
Algorithm based on shape topological structure reduces the impact to image illumination, collection forming shape letter to a certain extent
Breath, texture information occur in that a kind of new face identification method is referred to as biview face recognition algorithms.Using sub-space learning side
It is to be made up of the curve chart setting up face image that method builds texture model and shape topology.The biview recognition of face side that will propose
Method is compared with the recognizer being based only on texture or shape information.Experiment is entered under the change of illumination, expression and yardstick
OK, result shows, the performance of biview face identification method is better than the algorithm based on texture and shape.
The method of above-mentioned image recognition respectively has its pluses and minuses, and have is limited only to some special applications, such as car plate
Identification, recognition of face, Text region etc..Searching one kind is simple and convenient at present, should be readily appreciated that, feature is relatively comprehensive, range of application
Wider, that is, the preferable image-recognizing method of combination property, it is still the direction that people make great efforts.
Content of the invention
Present invention aims to the drawback of prior art, provide a kind of simple and convenient, should be readily appreciated that, feature complete
Face, the image-recognizing method based on difference characteristic description of applied range.
Problem of the present invention is to be realized with following technical proposals：
A kind of image-recognizing method based on difference characteristic description, methods described will first with selected membership function
The grayvalue transition of image slices vegetarian refreshments becomes degree of membership；Then, on the basis of defining difference value concept, image had same difference
The characteristic vector gathering as a region and defining this region of the neighbor pixel of different value；Afterwards by each region with being located at
The characteristic vector point of barycenter represents；Again the barycenter triangle of adjacent area is connected and use triangle interior angle as adjacent area
Relational angle, extract the feature of target image, finally image to be identified carried out the process of feature extraction, after processing
Image carry out target characteristic coupling, so that it is determined that whether it contains target image.
The above-mentioned image-recognizing method based on difference characteristic description, is specifically carried out according to the following steps：
1.. the extraction of target image characteristics and description
A. select a membership function, whereinIt is target imagef(x,y)
The gray value of pixel,x,yIt is the position coordinateses of pixel, utilizeGray value by target image pixel
Be converted to degree of membership；
B. calculate the difference value of each pixel and surrounding pixel point；
C. region segmentation is carried out to image：Select threshold value q, by difference valueAll phases less than thresholding q
Mutually the pixel merger of next-door neighbour is same region, all splits all zoness of different of target image by this method, then will
The center of mass point of regional is as the central point in this region；
D. define regional form factor k, form factor k be equal to region area divided by area circumference square；
E. define the centroid vector r (D, B) of center of mass point, that is, determine modulus value D and the angle of centroid vector rB：Calculate each
The mean difference of all pixels point in region, wherein N is the sum of the pixel in the same area, then
D1With the product of region shape coefficient k as centroid vector r mould, i.e. D=kD1；The angle of centroid vector r is, whereinB iFor the angle value of pixel, it is each pixel（Except boundary point）8 pictures with its neighborhood
The difference value of vegetarian refreshments is compared this obtained orientation angle changing to difference value maximum point；
F. the center of mass point of each adjacent area is connected, constructs the triangular mesh of target image, use all trianglees
Interior angle value structural features matrix M；
G. the centroid vector r (D, B) of target image and eigenmatrix M is combined, as the feature of target image；
2. in images to be recognized g (x, y) target image identification
According to step a-e, same technical finesse is carried out to images to be recognized g (x, y), obtain images to be recognized g (x, y)
In all center of mass point vector, then carry out image recognition, identification step is as follows：
H. target imagefThe centroid vector of (x, y) is carried out one by one with all centroid vector of images to be recognized g (x, y)
Vector matching（In certain range of error）, find out the vector of all couplings；
I. the barycenter of all neighbouring vectors of coupling in images to be recognized g (x, y) is connected, draw triangle, then by
Triangle interior angle constitutive characteristic matrix E；
J. the eigenmatrix M of target image and E matrix is used to contrast, if the partly continuous element in E matrix and Metzler matrix phase
With（In certain error) it is determined that containing target image in images to be recognized g (x, y)f(x,y).
The difference value of image area characteristics point is combined by the present invention with direction, constitutes a kind of new provincial characteristicss description arrow
Amount, proposes a kind of new image recognition algorithm on this basis, and the method is simple and convenient, should be readily appreciated that, feature is relatively comprehensive, application
Scope is wider, has preferable combination property compared with traditional method.
The present invention, using the difference value concept based on fuzzy membership, the gray scale value matrix of image is changed into difference value square
Battle array, extends low gray area, have compressed high gray area, the image of low gray area can be made more visiblely to show.As accompanying drawing 6 institute
Show it is assumed that the value of degree of membership is U, compareWithThe size of two functions, that is, compare a pixel and person in servitude
Genus degree is the relation between 1 postulated point, the amplification of difference measurement formula that can newly be defined, reduces a little.Known by Fig. 6：
Abscissa U between zero and one, two line intersection point U=0.13712886, when U ＜ intersection value, when that is, big with 1 difference ,-lgU ＞ 1-U, that is,
Difference value is exaggerated by the formula of new definition；When U ＞ intersection value, difference hour ,-lgU ＜ 1-U, that is, new formula is by difference
Value is reduced.It can be seen that method presented herein reduces to less difference value, larger difference value is carried out
Amplify, more meet the sense organ understanding of people.
Brief description
The invention will be further described below in conjunction with the accompanying drawings.
Fig. 1 a, Fig. 1 b are the example of single target identification（Fig. 1 a is target image, and Fig. 1 b is images to be recognized）；
Fig. 2 is the angle schematic diagram of pixel；
Fig. 3 a, Fig. 3 b are the target recognition example of multiple areas combine（Fig. 3 a is target image, and Fig. 3 b is figure to be identified
Picture）；
Fig. 4 a ~ Fig. 4 b is the triangle gridding of the target image in Fig. 3 and images to be recognized（Fig. 4 a is the triangle of target image
Grid, Fig. 4 b is the triangle gridding of images to be recognized）；
The invariance example that Fig. 5 describes for characteristics of image（Fig. 5 a is target image, and Fig. 5 b ratates 90 degrees for Fig. 5 a and to obtain
Image, Fig. 5 c extends one times of image obtaining, Fig. 5 d for Fig. 5 a, and e, f are respectively Fig. 5 a, b, c corresponding feature triangle net
Lattice）.
Fig. 6 is functionWithCurve comparison diagram.
In literary composition, each symbol inventory is：For membership function,f(x, y) is target image,For target image
The gray value of pixel,x,yIt is the position coordinateses of pixel,For difference value, q is the threshold value of difference value, r (D, B)
For centroid vector, D is the modulus value of centroid vector r,BFor the angle of centroid vector r, k is the form factor in region, D1For in region
The mean difference of all pixels point, N is the sum of the pixel in the same area, and M is the eigenmatrix of target image, g (x, y)
For images to be recognized, E is the eigenmatrix obtaining from images to be recognized.
Specific embodiment
Extract and the feature of description target image is the important step identifying, target characteristic should have uniqueness, so that area
Not in other targets.
The present invention proposes a kind of new image-recognizing method, on the basis of setting up difference concept, calculates different in image
The difference value in region, proposes image difference as feature description, carrys out recognition target image with this structural features amount.
If piece imagef(x, y) is target image, another image to be identifiedgTarget image portion is contained in (x, y)
Point, in such as Fig. 1, Fig. 1 a is target image, and Fig. 1 b is the images to be recognized containing target image.To find out target from g (x, y)
Image can carry out technical finesse by two parts work.First is to extract target imagefThe feature of (x, y), second utilizes mesh
Target feature and g (x, y) carry out characteristic matching, thus identifying target.The target containing in images to be recognized g (x, y) is permissible
Be the translation of target image, rotation and flexible after image.The present invention comprises to extract characteristics of image and realizes image recognition two
Part：
Part I：Extract and describe target image characteristics
Extracting and describe target image characteristics is the work first having to when realizing image recognition complete, and the method for the present invention is
Extract target imagefThe centroid vector of (x, y) and eigenmatrix are as the feature of target image.Comprise the following steps that：
（1）Select a membership function, whereinIt is target image pixel
Gray value,x,yIt is the position coordinateses of pixel, pass throughCalculate the degree of membership of each pixel of target image and preserve,
Namely by the gray value of target image pixelBe converted to degree of membership through membership function；
（2）Again by the degree of membership of target image pixelUse logarithmRepresent, and preserve；
（3）Calculate the difference value of each pixel and surrounding pixel point,
And preserve.
（4）Region segmentation is carried out to image.Select threshold value q, by difference valueAll phases less than thresholding q
Mutually the pixel merger of next-door neighbour is same region, all splits all zoness of different of target image by this method.Meanwhile,
Again using the center of mass point of regional as this region central point；
（5）Define regional form factor k, form factor k be equal to region area divided by area circumference square；
（6）Define the centroid vector r (D, B) of center of mass point, that is, determine modulus value D and the angle of centroid vector rB.To each
All pixels point in regionCarry out average, calculate mean difference, wherein N is the same area
The sum of interior pixel, then D1With the product of region shape coefficient k as centroid vector mould, i.e. D=kD1, then same
In region, the angle value of all pixels point carries out mean value calculation, as the angle of centroid vector, the angle value of pixelB iIt is
Refer to each pixel（Except boundary point）This point obtained by being compared with the difference value of 8 pixels of its neighborhood is to difference
The orientation angle of different value maximum point change, as shown in Fig. 2 the angle of centroid vector r；
（7）The center of mass point of each adjacent area is connected, constructs the triangular mesh of target image, use all trianglees
Interior angle value the relative position relation between adjacent area is described；
（8）Determine the eigenmatrix M of target image.Interior angle value structural features matrix with each triangle；
（9）The centroid vector r (D, B) of the target image being eventually found and eigenmatrix M is exactly the feature of target image.
Part II：Target image identification method in images to be recognized g (x, y)
Process step according to Part I（1）-（6）Same technical finesse is carried out to images to be recognized g (x, y), obtains
The centroid vector of all center of mass point in images to be recognized g (x, y), then carries out image recognition, and identification step is as follows：
（10）Target imagefAll centroid vector of the centroid vector of (x, y) and images to be recognized g (x, y) carry out by
Individual vector matching（In certain range of error）, find out the vector of all couplings；
（11）The barycenter of all neighbouring vectors of coupling in images to be recognized g (x, y) is connected, draws triangle, then
By triangle interior angle constitutive characteristic matrix E；
（12）Contrasted with the eigenmatrix M and E matrix of target image, if the partly continuous element in E matrix and Metzler matrix
Identical（In certain error, it is determined as containing target image in images to be recognized g (x, y)f(x,y).
The core of the present invention of above step description is：Select a membership function, using membership function by image
The grayvalue transition of pixel becomes degree of membership；Define the concept of difference value, difference calculating can be carried out, and by human eye to image ash
The non-linear relation of degree contrast sensitivity is converted to linear relationship；Determine that image has identical（Less than certain thresholding）Difference value
The set of neighbor pixel constitutes a region, defines the characteristic vector in this region, i.e. the average difference values in region and region shape
The product of shape coefficient calculates the barycenter in region as modulus value, the meansigma methodss of area pixel point angle as the angle value of vector,
Each region is represented with the characteristic vector point positioned at barycenter；The barycenter of adjacent area is connected with triangle, with triangle
Angle is as the relational angle of adjacent area, thus extracting the feature of target image.For images to be recognized, through similar to spy
Levying the processing procedure of extraction, the image after processing being carried out target characteristic coupling, thus find out in images to be recognized whether containing
Target image.
Present invention application example 1：
As shown in figure 1, in the case of all not considering background, Fig. 1 a is the target image being only made up of a region, Fig. 1 b
For images to be recognized, b figure can be divided into 5 regions（By from top to bottom, from left to right label）.Each is obtained by feature extraction
The feature description vector matrix in regionRepresent,（D is the modulus value of centroid vector r,BAngle for centroid vector r）.
Choose the error threshold of suitable centroid vector modulus value and angleth1Withth2So that：, with
When,（Th1=20 in this example, th2=1）.
Can it is concluded that：The 4th Region Matching in target image and testing image.
a:, b:,
The computation example 2 of the present invention：
Assign the b in Fig. 1 as an entirety, it is made up of 5 zonules, as shown in figure 3, wherein Fig. 3 a is target figure
As it is desirable to identify whole target in complicated testing image Fig. 3 b.
First pass through and calculate and each region modulus value of target image, the Euclidean distance of deflection respectively, in testing image
Find the regional matching with target image（a1~b4,a2~b5,a3~b6,a4~b7,a5~b8）.
a:b:,
By the Based on Feature Points positioned at barycenter for the regional of target image, connect the characteristic point construction of adjacent area
Delaunay triangular mesh as shown in fig. 4 a, and the length of side of each triangle and angle value is charged to eigenmatrix（M2a,M3a）
In.In the same manner mated with target image in testing image 5 regions are also constructed triangular mesh, recording feature matrix.
（Matrix M2a、M2bIn row vector be three sides of a triangle long, M3a、M3bIn row vector be triangle three angle values）,
,
,
By the Euclidean distance between row vector corresponding in calculating matrix, when threshold selection is suitable, checking two width figures
Triangle gridding also mates.Thus provable have identified target in testing image.
This recognizer can identify to the translation of target image, rotation and scaling：
Assign Fig. 5 a as a target image, be made up of 5 regions, Fig. 5 b ratates 90 degrees, for Fig. 5 a, the image obtaining, figure
5c extends one times of image obtaining for Fig. 5 a, obtains the respective eigenmatrix of 3 width figures by the inventive method, Fig. 5 d, 5e, 5f divide
Fig. 5 a, 5b, 5c Wei respectively not scheme corresponding feature triangle grid respectively.
,
,
,
A and b compares, and the order of characteristic point changes, and the modulus value of each characteristic point is more or less the same, and angle has all changed greatly
About 90 degree, and each triangle length of side of characteristic point construction is almost unchanged with angle.A is compared with c, the modulus value of characteristic point and angle
And the angle of triangle is more or less the same, it is only that the length of side of each feature triangle becomes for original 2 times.
The modulus value of characteristic point has the flexible invariance of rotation translation, and direction has the flexible invariance of translation, feature triangle
The length of side has rotation translation invariance, and angle has the flexible invariance of rotation translation simultaneously.These conclusions comprehensive, as long as select close
Suitable judgement standard just can identify translation scaling or the target that have rotated in testing image.
Claims (2)
1. a kind of image-recognizing method based on difference characteristic description, is characterized in that, methods described is subordinate to first with selected
The grayvalue transition of image slices vegetarian refreshments is become degree of membership by degree function；Then, on the basis of defining difference value concept, image is had
There is the characteristic vector gathering as a region and defining this region of the neighbor pixel of same difference value, described difference value is
Refer to the difference value of each pixel and surrounding pixel point；Afterwards each region is represented with the characteristic vector point positioned at barycenter；
Again the barycenter triangle of adjacent area is connected and use triangle interior angle as the relational angle of adjacent area, extract target
Image to be identified is finally carried out the process of feature extraction by the feature of image, and the image after processing is carried out target characteristic
Join, so that it is determined that whether it contains target image.
2. the image-recognizing method based on difference characteristic description according to claim 1, is characterized in that, methods described press with
Lower step is carried out：
1.. the extraction of target image characteristics and description
A. select a membership functionWherein I (x, y) is target image f (x, y) pixel
Gray value, x, y are the position coordinateses of pixel, are converted to gray value I (x, y) of target image pixel using u (x, y)
Degree of membership；
B. calculate difference value e (x, y) of each pixel and surrounding pixel point=| log u (xi,yj)-log u(xi+1,yj+1)
|；
C. region segmentation is carried out to image：Select threshold value q, by difference value ei(x, y) is less than all mutually tight of thresholding q
Adjacent pixel merger is same region, all splits all zoness of different of target image by this method, then by each
The center of mass point in region is as the central point in this region；
D. define regional form factor k, form factor k be equal to region area divided by area circumference square；
E. define the centroid vector r (D, B) of center of mass point, that is, determine modulus value D and the angle B of centroid vector r：Calculate regional
The mean difference of interior all pixels pointWherein N is the sum of the pixel in the same area, then D1
With the product of region shape coefficient k as centroid vector r mould, i.e. D=kD1；The angle of centroid vector r is
Wherein BiFor the angle value of pixel, refer to the difference of each pixel in addition to boundary point and 8 pixels of its neighborhood
Different value is compared the orientation angle that this obtained pixel changes to difference value maximum point；
F. the center of mass point of each adjacent area is connected, constructs the triangular mesh of target image, with all trianglees
Angle value structural features matrix M；
G. the centroid vector r (D, B) of target image and eigenmatrix M is combined, as the feature of target image；
2. in images to be recognized g (x, y) target image identification
According to step a-e, same technical finesse is carried out to images to be recognized g (x, y), obtain in images to be recognized g (x, y)
The vector of all center of mass point, then carries out image recognition, and identification step is as follows：
H. the centroid vector of target image f (x, y) and all centroid vector of images to be recognized g (x, y) are carried out vector one by one
Coupling, finds out the vector of all couplings；
I. the barycenter of all neighbouring vectors of coupling in images to be recognized g (x, y) is connected, draw triangle, then by triangle
Shape interior angle constitutive characteristic matrix E；
J. the eigenmatrix M of target image and E matrix is used to contrast, if the partly continuous element in E matrix is identical with Metzler matrix,
Then determine in images to be recognized g (x, y) and contain target image f (x, y).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410063894.9A CN103927511B (en) | 2014-02-25 | 2014-02-25 | image identification method based on difference feature description |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410063894.9A CN103927511B (en) | 2014-02-25 | 2014-02-25 | image identification method based on difference feature description |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103927511A CN103927511A (en) | 2014-07-16 |
| CN103927511B true CN103927511B (en) | 2017-02-15 |
Family
ID=51145727
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410063894.9A Expired - Fee Related CN103927511B (en) | 2014-02-25 | 2014-02-25 | image identification method based on difference feature description |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103927511B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105117712A (en) * | 2015-09-15 | 2015-12-02 | 北京天创征腾信息科技有限公司 | Single-sample human face recognition method compatible for human face aging recognition |
| CN105825228B (en) | 2016-03-14 | 2019-04-30 | 百度在线网络技术（北京）有限公司 | Image-recognizing method and device |
| CN106446750B (en) * | 2016-07-07 | 2018-09-14 | 深圳市华汉伟业科技有限公司 | A kind of bar code read method and device |
| CN107067426B (en) * | 2017-04-17 | 2018-02-06 | 东华理工大学 | A kind of image texture characteristic extracting method based on nested three-legged structure |
| CN108491498B (en) * | 2018-03-20 | 2021-06-25 | 济南神博信息技术有限公司 | Bayonet image target searching method based on multi-feature detection |
| CN108846821B (en) * | 2018-04-27 | 2022-01-11 | 电子科技大学 | Wavelet transform-based cellular neural network hot region fusion method |
| CN108921020A (en) * | 2018-05-29 | 2018-11-30 | 北大青鸟环宇消防设备股份有限公司 | A kind of fire image-recognizing method and its device |
| CN110032964B (en) * | 2019-04-08 | 2023-07-14 | 腾讯科技(成都)有限公司 | Image processing method, device, equipment and storage medium for identifying visual angle |
| CN112116723B (en) * | 2020-08-10 | 2021-06-22 | 中标慧安信息技术股份有限公司 | License plate recognition-based non-inductive charging method and device |
| CN113284152B (en) * | 2021-04-01 | 2023-06-27 | 西安理工大学 | Light field characteristic oil painting identification method |
| CN115641455B (en) * | 2022-09-16 | 2024-01-09 | 杭州视图智航科技有限公司 | Image matching method based on multi-feature fusion |
| CN116152530B (en) * | 2023-04-21 | 2023-10-03 | 青岛尘元科技信息有限公司 | Image difference determining method and device, storage medium and electronic equipment |
| CN117522534B (en) * | 2024-01-08 | 2024-03-29 | 深圳市卖点科技股份有限公司 | Intelligent commodity display method and system based on Internet of things |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1193148A (en) * | 1997-03-12 | 1998-09-16 | 萧培墉 | Method for identifying finger print by collecting high-order characteristic point group |
| US6393159B1 (en) * | 1998-06-29 | 2002-05-21 | The Regents Of The University Of California | Multiscale characterization and analysis of shapes |
| CN1912889A (en) * | 2005-08-11 | 2007-02-14 | 中国科学院自动化研究所 | Deformed fingerprint identification method based on local triangle structure characteristic collection |
| CN101398689A (en) * | 2008-10-30 | 2009-04-01 | 中控科技集团有限公司 | Real-time color auto acquisition robot control method and the robot |
-
2014
- 2014-02-25 CN CN201410063894.9A patent/CN103927511B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1193148A (en) * | 1997-03-12 | 1998-09-16 | 萧培墉 | Method for identifying finger print by collecting high-order characteristic point group |
| US6393159B1 (en) * | 1998-06-29 | 2002-05-21 | The Regents Of The University Of California | Multiscale characterization and analysis of shapes |
| CN1912889A (en) * | 2005-08-11 | 2007-02-14 | 中国科学院自动化研究所 | Deformed fingerprint identification method based on local triangle structure characteristic collection |
| CN101398689A (en) * | 2008-10-30 | 2009-04-01 | 中控科技集团有限公司 | Real-time color auto acquisition robot control method and the robot |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103927511A (en) | 2014-07-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103927511B (en) | image identification method based on difference feature description | |
| CN104915636B (en) | Remote sensing image road recognition methods based on multistage frame significant characteristics | |
| CN106909902B (en) | Remote sensing target detection method based on improved hierarchical significant model | |
| Gould et al. | Decomposing a scene into geometric and semantically consistent regions | |
| CN104063702B (en) | Three-dimensional gait recognition based on shielding recovery and partial similarity matching | |
| CN105574534A (en) | Significant object detection method based on sparse subspace clustering and low-order expression | |
| CN102509104B (en) | Confidence map-based method for distinguishing and detecting virtual object of augmented reality scene | |
| Herdiyeni et al. | Combination of morphological, local binary pattern variance and color moments features for indonesian medicinal plants identification | |
| CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
| CN109543632A (en) | A kind of deep layer network pedestrian detection method based on the guidance of shallow-layer Fusion Features | |
| Wang et al. | Head pose estimation with combined 2D SIFT and 3D HOG features | |
| CN109635726B (en) | Landslide identification method based on combination of symmetric deep network and multi-scale pooling | |
| CN109902585A (en) | A kind of three modality fusion recognition methods of finger based on graph model | |
| CN105354555B (en) | A kind of three-dimensional face identification method based on probability graph model | |
| CN107392187A (en) | A kind of human face in-vivo detection method based on gradient orientation histogram | |
| CN106844739A (en) | A kind of Remote Sensing Imagery Change information retrieval method based on neutral net coorinated training | |
| CN110516533A (en) | A kind of pedestrian based on depth measure discrimination method again | |
| CN104268833A (en) | New image fusion method based on shift invariance shearlet transformation | |
| CN110334656A (en) | Multi-source Remote Sensing Images Clean water withdraw method and device based on information source probability weight | |
| CN105975906B (en) | A kind of PCA static gesture identification methods based on area features | |
| CN103399863B (en) | Image search method based on the poor characteristic bag of edge direction | |
| CN106407975B (en) | Multiple dimensioned layering object detection method based on space-optical spectrum structural constraint | |
| CN114511012A (en) | SAR image and optical image matching method based on feature matching and position matching | |
| CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
| CN113011359B (en) | Method for simultaneously detecting plane structure and generating plane description based on image and application |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170215 Termination date: 20200225 |