CN106570475A - Purple clay teapot seal retrieval method - Google Patents
Purple clay teapot seal retrieval method Download PDFInfo
- Publication number
- CN106570475A CN106570475A CN201610955127.8A CN201610955127A CN106570475A CN 106570475 A CN106570475 A CN 106570475A CN 201610955127 A CN201610955127 A CN 201610955127A CN 106570475 A CN106570475 A CN 106570475A
- Authority
- CN
- China
- Prior art keywords
- image
- seal
- dark
- point
- matching
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a purple clay teapot seal retrieval method. The method is characterized by cutting and zooming a purple clay teapot seal image photographed by a mobile phone; through adjusting a mean value and a mean square error of the image, changing image brightness and contrast and acquiring a standard image; in the standard image, extracting an edge characteristic and a shape area characteristic of a purple clay teapot seal; after characteristic fusion is performed, using a trained SVM classifier to coarsely divide the seal into four types of a round shape, a triangle, rectangle and an irregular shape; extracting a SIFT characteristic of the purple clay teapot seal image, introducing a geometric constraint condition, and using an improved matching method based on belief propagation to match with the image in a template database; and simultaneously using a clustering method to optimize retrieval time and completing seal retrieval. By using the method of the invention, good robustness is possessed to uneven illumination and affine transformation of rotation, zooming, side looking and the like.
Description
Technical field
The invention belongs to Digital Image Processing and mode identification technology, and in particular to a kind of dark-red enameled pottery seal retrieval side
Method.
Background technology
The retrieval of seal is due to the multiformity of its font, the unstability of printed text carrier, the identification of current dark-red enameled pottery seal
Still mainly by perusal, by the minutia for comparing seal, control is overlapped, splicing such as compares at the manual method, and identification is purple
The producer of sand kettle.In the last few years, flourishing with Digital Image Processing and mode identification technology, is that computer generation replaces people
Work carries out the retrieval of dark-red enameled pottery seal and provides reliable theoretical basiss.
2008, Ni Qi et al. proposed a kind of seal recognizer (Ni Qi, king split based on gradient with characteristic matching
New match, Li Jian, etc. based on gradient segmentation and seal recognizer research [J] of characteristic matching. computer engineering and design,
2008,29(20):5400-5402.), first by image histogram equalization stretching seal gray scale and background gray scale difference value, then
Seal edge is slightly picked using shade of gray edge cataclysm principle, finally by template matching essence detection seal edge, completes to know
Not.The method simple and fast, but the shortcoming of template matching is limited to, not with rotational invariance and scale invariability.
2010, Lang Haitao et al. proposed a kind of falsification of seal based on feature line randomly generated by matching feature points and recognizes
Method (Lang Haitao, Lei Lan mono- is luxuriant and rich with fragrance. the method for identifying forgery seal based on feature line randomly generated by matching feature points:CN,
CN101894260A [P] .2010.), specific implementation method is to extract the characteristic point of altimetric image to be checked, builds data base, is contained
The information of each characteristic point position and other descriptors.Based on matching characteristic point generate at random discernible seal to be tested with
Reference seal characteristics of image line；According to the printed text that same seal is obtained under various circumstances, there is the image of making peace of image information one special
A similar principle of distribution is levied, judges whether altimetric image to be checked is falsification of seal image.The characteristics of the method has simple efficient,
But when being limited to light change and scaling, the disappearance of Partial Feature point may cause positive sample to be mistaken for falsification of seal image.
2007, Zhang Yi et al. proposed a kind of print identification control method (Zhang Yi, Lv Jiancheng, a power Fang, etc. seal
Identification system and its control method:CN, CN 101008985A [P] .2007.), carried with binaryzation, skeletal extraction, frame first
Take and extract printed text to be identified with printed text four operating procedures of extraction.Next in printed text step of registration, first rough registration, first will
Printed text to be identified is adjusted to roughly the same position and direction with masterplate printed text, and fine registration is further adjusted to two width printed texts
Almost identical position and direction；In printed text differentiates step, multistage recognition strategy and multiple features fusion for classification decision-making are employed
Method differentiates to printed text to be identified and masterplate printed text.The method can be realized well differentiating control to complete seal image.
But when image has disappearance, the effect in image registration will drastically decline.
By finding current many methods to existing seal technology of identification analysis：Global registration method, is limited to image
The bottleneck of matching, not with the ability of anti-affine transformation；Due to the artistry of dark-red enameled pottery seal font, Chinese characters recognition method is not yet
Can be applied to well in seal retrieval；At present, the retrieval of dark-red enameled pottery seal is used for without a kind of framework of system.
The content of the invention
Goal of the invention：In order to overcome the deficiencies in the prior art, the present invention to provide a kind of dark-red enameled pottery seal retrieval side
Method, the method can reliably be quickly accomplished the retrieval of dark-red enameled pottery seal.
Technical scheme：For achieving the above object, the technical solution used in the present invention is：
Step 1, to the shooting image of dark-red enameled pottery seal pretreatment is carried out, and obtains the normalized images of detection zone；
The normalized images are carried out the adjustment of brightness and contrast by step 2, by adjusting the average of image and mean square
Difference, the brightness and contrast for making image is separately fixed at a particular value；
Step 3, by the shape Preliminary division of seal in image；
Step 4, to image contract SIFT feature to be retrieved, forms 128 and ties up SIFT description；And to template image and to be checked
Rope image carries out characteristic matching using improved BP-SIFT algorithms, by the matching rate for calculating match point, judges the affiliated mould of name
Classification in plate storehouse, completes the identification of seal name.
Further, step 1 is realized as follows：The shooting image is carried out into rim detection and marking area is carried
Take, obtain seal part；And zoom in and out, make edge retain the detection zone that 100*100 sizes are obtained after at least 5 pixel blank
The normalized images in domain.
Further, the concrete grammar of step 2 adjustment image is：
Step 2-1, by the average of image 0, g is adjusted to1=g-u1, it is in course of adjustment, mean square deviation is constant；Wherein, g1For
The pixel value of the image that the first step is obtained, g is original pixel value, u1For original image average；
Step 2-2, by the mean square deviation of image d is adjusted to0, g2=g1*d0/d1, be in course of adjustment, average is constant, still for
0；Wherein, d0For fresh target image mean-squared deviation, g2The pixel value of the image obtained for second step, g1For first step gained image
Pixel value, d1For artwork mean square deviation；
Step 2-3, by the average of image u is adjusted to0, g3=g2+u0, wherein, g3For target image pixel value, g2For second
The pixel value of step gained image, u0For target image average.
Further, in step 3, seal image to be retrieved is divided into four big class：Circular stamp image, rectangular seal image,
Triangle seal image, irregular seal image.
Further, it is to the concrete grammar that the shape of seal in image carries out Preliminary division in step 3：
Step 3-1, by the detection of canny operators the marginal information of image, the compartment of terrain on the edge contour for obtaining are obtained
Sampled, image centroid is calculated to the distance of marginal point, as contour description；
Step 3-2, by image centroid, tries to achieve maximum circumscribed circle；Sampling on this basis obtains the region shape of target
Feature；
Step 3-3, normalization edge feature and region shape feature, are input into after linear fusion feature as grader；
Step 3-4, using DAG rules in many classification, using N (N-1)/2 SVM classifier, grader forms one
Directed acyclic graph, when one sample of arrival, from top to down finely divides sample, until reaching leaf node, by image
Shape carry out Preliminary division, wherein, N be divide classification number.
Further, the sampling interval in step 3 beWherein, L is profile length.
Further, the SVM classifier in step 3 uses RBF kernel functions, be defined as the Euclidean of space midpoint x to y points away from
From monotonic function, y for kernel function center：
Wherein, the influence factor of RBF cores SVM performances is punishment parameter C and nuclear parameter σ in space2；Using grid data service
Select the parameter：It is rightN and M value are taken respectively, and the wherein scope of C is set to Scope is set toStep pitch is 0.1；Combine all of N × M value and calculate SVM Generalization Abilities, will wherein cause SVM to promote best performance
Combination alternatively parameter.
Further, improved BP-SIFT algorithms include introducing the constraint of two space of points distance ratios, point of proximity in step 4
Cluster before search is reducing time complexity.
Further, the concrete grammar of improved BP-SIFT algorithms is in step 4：
Step 4-1, using classical SIFT algorithms to image contract SIFT feature to be retrieved, forms 128 dimension SIFT features and retouches
State son；
Step 4-2, is gathered all characteristic points in template image for C classes using KMEANS clusters according to coordinate position；For
All characteristic points in template image, calculate which class it belongs to, and find out nearest K point of proximity in the apoplexy due to endogenous wind；
Step 4-3, it is a constant to initialize all of confidence level, if n=1；To current signature point in template image and
The combination of the current signature point in image to be retrieved is right, iteratively updates its matching confidence level；
Step 4-4, is constrained and the Europe between two width image characteristic points by the geometric distance of characteristic point and its K point of proximity
The linear function combinational estimation of family name's distance calculates matching probability；
Then step 4-5, n ← n+1 goes to (4-4) step, until matching probability no longer changes or iterationses n reaches
It is maximum；
Step 4-6, if matching degree is less than predetermined threshold value, the two Point matchings success；Otherwise it is not that matching is right.
All template images are repeated above step by step 4-7, calculate image to be retrieved with each template image
Matching rate, is classified as the class of matching degree highest one.
Beneficial effect：The dark-red enameled pottery seal search method that the present invention is provided, compared with it presently, there are technology, has with following
Beneficial effect：Retrieval precision is higher, combines machine learning and image registration techniques, the Preliminary division big class before characteristic matching, energy
Enough reductions subsequently with the time of template matching, filter out some false retrievals；Spatial information and local message are combined, with good
Anti-rotation, the ability of scaling equiaffine conversion；Using cluster before search point of proximity, it is to avoid global extensive search, significantly
The time required to reducing retrieval.
Description of the drawings
Fig. 1 is the dark-red enameled pottery seal retrieval flow figure of the present invention.
Fig. 2 (a), Fig. 2 (b), Fig. 2 (c) are the result schematic diagrams for adjusting gray level image average and mean square deviation.
Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) are to extract edge feature schematic diagram.
Fig. 4 (a), Fig. 4 (b) are the successful schematic diagrams of retrieval.Black line segment connects the two of image to be matched and template image
Individual match point.Wherein 4 (a) contains rotation transformation.
Fig. 5 (a), Fig. 5 (b) are the schematic diagrams that such seal is not retrieved in template image.
Specific embodiment
The present invention is further described below in conjunction with the accompanying drawings.
The present invention is a kind of dark-red enameled pottery seal search method.The dark-red enameled pottery seal image for obtaining to mobile phone photograph first, enters
Row cutting, scaling, then change the brightness and contrast of image by adjusting the average and mean square deviation of image, obtain the figure of specification
Picture.The edge feature and shape area feature of dark-red enameled pottery seal are extracted in the image of specification, after carrying out Feature Fusion, using
The SVM classifier for training, is circle, triangle, rectangle, irregularly shaped four big class by seal thick division.Extract dark-red enameled pottery
The SIFT feature of seal image, introduces geometry constraint conditions, using the improved matching process based on belief propagation and template base
Middle images match；Simultaneously using clustering method optimization retrieval required time, seal retrieval is completed.The invention to uneven illumination and
Rotation, scaling, the conversion of side-looking equiaffine, there is good robustness.
After the dark-red enameled pottery seal image standardization processing that the present invention will be obtained from shooting or network, average is first adjusted equal
Variance pretreatment.Edge feature is extracted, region shape feature is extracted, is input into as SVM classifier after fusion, by seal image root
According to being divided into four big class at the beginning of geometry.SIFT feature is extracted again, it is registering with template image using improved BP-SIFT algorithms,
Complete seal retrieval.Flow process is as shown in Figure 1.
Step 1, the image to shooting or network originating is obtained carries out standardization processing, including effective extracted region, edge
Cutting and image size are adjusted.
Step 2, to normalized image the adjustment of brightness and contrast is carried out, by adjusting the average of image and mean square
Difference, the brightness and contrast that can make image is separately fixed at a particular value, strengthens robustness of the seal image to illumination effect.
As shown in Fig. 2 (a), Fig. 2 (b), Fig. 2 (c).
(1) average of image is adjusted to into 0.g1=g-u1.It is in course of adjustment, mean square deviation is constant.Wherein, g1For first
The pixel value of image that step is obtained, g is original pixel value, u1For original image average.
(2) mean square deviation of image is adjusted to into d0, g2=g1*d0/d1.It is in course of adjustment, average is constant, is still 0.Its
In, d0For fresh target image mean-squared deviation, g2The pixel value of the image obtained for second step, g1For the pixel of first step gained image
Value, d1For artwork mean square deviation.
(3) average of image is adjusted to into u0。g3=g2+u0.The present invention takes d0=50, u0=127.Wherein, g3For target
Image pixel value, g2For the pixel value of second step gained image, u0For target image average.
Step 3, to pretreated image zooming-out edge feature and region shape feature, is input into SVM classifier, according to right
The priori of seal is classified at the beginning of completing seal shape.As shown in Fig. 3 (a), Fig. 3 (b), Fig. 3 (c).
(3-1) marginal information of image is obtained by the detection of canny operators, then, is spaced on the edge contour for obtaining
Sampled, altogether sample 20 times, calculate image centroid to marginal point distance, as the sub- f of contour description1=[b1 b2 ...
b20].Including four steps：First, the convolution of Gaussian filter and image is sought, then, with the finite difference formulations of single order local derviation
The direction of gradient and amplitude；Then, non-maxima suppression (only retaining the maximum point of amplitude localized variation) is carried out to gradient magnitude；
Finally, false edge segment number is reduced using dual threashold value-based algorithm.
(3-2) lead to the centre of form for finding binary image, try to achieve maximum circumscribed circle.Sampling on this basis obtains the area of target
Domain shape facility.First with formula
Try to achieve image centroid.Wherein, (xc,yc) it is image centroid coordinate, m is collection point number.
With the centre of form as the center of circle, as radius, the maximum circumscribed circle of target is formed with apart from the maximum pixel distance of the centre of form.Will
Radius is equidistant to be divided into 5 parts, forms 5 concentric circulars.Start rotation from 0 to 360 degree, at intervals ofSampled, record
Rotary shaft and the point of intersection grey scale pixel value of concentric circular, form the shape descriptor that size is 5 × 20
Wherein L is circumference.
(3-3) normalization edge feature and region shape feature, are input into after linear fusion feature as grader.
To edge feature and the fusion of form matrix feature normalization：finew=(fi-fmin)/(fmax-fmin)。
Wherein, finewFor the feature after i-th normalization, fminAnd fmaxRespectively minimum and maximum characteristic quantity.Formed most
The Feature Descriptor of whole 6 × 20 dimension
(3-4) using DAG rules in many classification, with N (N-1)/2, the SVM classifier of N=4 of the present invention, i.e., 6, divide
Class device forms a directed acyclic graph, when one sample of arrival, from top to down gradually finely divides sample, until reaching
Leaf node, image is finally divided into rectangle, circular, triangle, irregularly shaped four big class.
Choose and be based on SVM classifier, using RBF kernel functions, be defined as the dullness of the Euclidean distance of space midpoint x to y points
Function, y is the center of kernel function.
Sample can be mapped to a higher dimensional space by it, and linear kernel function is a special case of RBF kernel functions, thus not
With considering further that linear kernel function.And RBF kernel functional parameters are fewer than Polynomial kernel function, computation complexity is low.
Wherein, the influence factor of RBF cores SVM performances is punishment parameter C and nuclear parameter σ in space2；Using grid data service
Select the parameter：It is rightN and M value are taken respectively, and the wherein scope of C is set to Scope is set to
[2-10,23], step pitch is 0.1.Combine all of N × M value and calculate SVM Generalization Abilities, will wherein cause SVM to promote performance most
Excellent combination alternatively parameter.
Step 4, to image contract SIFT feature to be retrieved, forms 128 and ties up SIFT description.To template image and to be retrieved
Image carries out characteristic matching using improved BP-SIFT algorithms.By the matching rate for calculating match point, the affiliated template of name is judged
Classification in storehouse, completes the identification of seal name.The thought of BP-SIFT algorithms is as follows：Because traditional SIFT matchings lack space letter
Breath constraint, BP-SIFT is judged by the space geometry of the adjacent point of characteristic point apart from difference respectively in two images
The similarity of geometric distance.Improve the matching degree that the utilization BP-SIFT algorithms of traditional SIFT matchings are calculated between characteristic point.But
Weak point has at 2 points：One, there are scaling, mistake to cut equiaffine change in image and change, 2 distance differences of the correspondence in two width images
And it is unequal；Two, the time complexity for finding point of proximity is too big.More generally criterion is present invention employs, using distance ratio
As new geometry constraint conditions；Using cluster before global search point of proximity, being greatly reduced needs the time of search.
(1) using classics SIFT algorithms to image contract SIFT feature to be retrieved, 128 are formed and ties up SIFT feature description；
(2) all characteristic points in template image are gathered for C classes according to coordinate position using KMEANS clusters；For Prototype drawing
All characteristic points as in, calculate which class it belongs to, and find out nearest K point of proximity in the apoplexy due to endogenous wind；
(3) it is a constant to initialize all of confidence level, if n=1.To current signature point in template image and to be retrieved
The combination of the current signature point in image is right, iteratively updates its matching confidence level；
(4) constrained by the geometric distance of characteristic point and its K point of proximity and the Euclidean between two width image characteristic points away from
From linear function combinational estimation calculate matching probability；
(5) n ← n+1, then goes to (4th) step, until matching probability no longer changes or iterationses n reaches maximum；
(6) if matching degree is less than predetermined threshold value, the two Point matchings success；Otherwise it is not that matching is right.Such as Fig. 4 (a), figure
Shown in 4 (b), successful schematic diagram is retrieved, black line segment connects two match points of image to be matched and template image, wherein 4
A () contains rotation transformation.Fig. 5 (a), Fig. 5 (b) are the unsuccessful result schematic diagrams of retrieval.
(7) all template images are repeated with above step, calculating image to be retrieved is matched with each template image
Rate.It is classified as the class of matching degree highest one.
In sum, the present invention realizes the scheme that dark-red enameled pottery seal is retrieved with reference to SVM shapes rough sort and Feature Points Matching：
(1) average that adjusts to image and mean square deviation are processed, and brightness of image and contrast can be fixed on into a particular value,
Image is set to have good anti-light photograph interference performance.
(2) using SVM to seal shape facility rough sort, the big class of energy Preliminary division image to be retrieved, reduction is subsequently needed
The data volume of the template image to be matched, improves retrieval time.
(3) in matching using image characteristic point and around it point of proximity geological information, both remained local feature,
Global characteristics are combined, the degree of accuracy of retrieval is improve.
(4) point of proximity around characteristic point is clustered, it is to avoid global search, in the case where precision affects very little,
Improve the time of retrieval.
Tested by 500 samples, the present invention reaches more than 95.6% to the classification accuracy of seal shape.To technologist
Matching degree reach more than 91.2%.Compared with traditional method with character recognition or image registration retrieval seal, to illumination,
The robustness of the changes such as rotation, scaling is higher.
The above is only the preferred embodiment of the present invention, it should be pointed out that：For the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (9)
1. a kind of dark-red enameled pottery seal search method, it is characterised in that：Comprise the steps：
(1) pretreatment is carried out to the shooting image of dark-red enameled pottery seal, obtains the normalized images of detection zone；
(2) normalized images are carried out with the adjustment of brightness and contrast, by the average and mean square deviation that adjust image, figure is made
The brightness and contrast of picture is separately fixed at a particular value；
(3) by the shape Preliminary division of seal in image；
(4) to image contract SIFT feature to be retrieved, 128 are formed and ties up SIFT description；And to template image and image to be retrieved
Characteristic matching is carried out using improved BP-SIFT algorithms, by the matching rate for calculating match point, in judging the affiliated template base of name
Classification, complete seal name identification.
2. dark-red enameled pottery seal search method according to claim 1, it is characterised in that：Image semantic classification in step 1
Method is：The shooting image is carried out into rim detection and marking area is extracted, obtain seal part；And zoom in and out, make side
Edge retains the normalized images of the detection zone that 100*100 sizes are obtained after at least 5 pixel blank.
3. dark-red enameled pottery seal search method according to claim 1, it is characterised in that：Step (2) the adjustment image
Concrete grammar is：
(201) average of image is adjusted to into 0, g1=g-u1, it is in course of adjustment, mean square deviation is constant；Wherein, g1For the first step
The pixel value of the image for obtaining, g is original pixel value, u1For original image average；
(202) mean square deviation of image is adjusted to into d0, g2=g1*d0/d1, it is in course of adjustment, average is constant, is still 0；Wherein,
d0For fresh target image mean-squared deviation, g2The pixel value of the image obtained for second step, g1The pixel value of image obtained by the first step,
d1For artwork mean square deviation.
(203) average of image is adjusted to into u0, g3=g2+u0, wherein, g3For target image pixel value, g2For second step gained
The pixel value of image, u0For target image average.
4. dark-red enameled pottery seal search method according to claim 1, it is characterised in that：In the step (3), print to be retrieved
Chapter image is divided into four big class：Circular stamp image, rectangular seal image, triangle seal image, irregular seal image.
5. the dark-red enameled pottery seal search method according to claim 1 or 4, it is characterised in that：The step (3) is in image
The shape of seal carries out the concrete grammar of Preliminary division, comprises the following steps：
(301) marginal information of image is obtained by the detection of canny operators, compartment of terrain is adopted on the edge contour for obtaining
Sample, calculates image centroid to the distance of marginal point, as contour description；
(302) by image centroid, maximum circumscribed circle is tried to achieve；Sampling on this basis obtains the region shape feature of target；
(303) normalization edge feature and region shape feature, are input into after linear fusion feature as grader；
(304) using DAG rules in many classification, using N (N-1)/2 SVM classifier, grader forms a directed acyclic
Figure, when one sample of arrival, from top to down finely divides sample, until reaching leaf node, the shape of image is entered
Row Preliminary division, wherein, N is the classification number for dividing.
6. dark-red enameled pottery seal search method according to claim 5, it is characterised in that：In the step (301), between sampling
It is divided intoWherein, L is profile length.
7. dark-red enameled pottery seal search method according to claim 5, it is characterised in that：In the step (304), SVM point
Class device uses RBF kernel functions, is defined as the monotonic function of the Euclidean distance of space midpoint x to y points, and y is the center of kernel function：
Wherein, the influence factor of RBF cores SVM performances is punishment parameter C and nuclear parameter σ in space2；Selected using grid data service
The parameter：It is rightN and M value are taken respectively, and the wherein scope of C is set to [2-10,27],Scope is set to [2-10,23], step pitch is 0.1；Combine all of N × M value and calculate SVM Generalization Abilities, will wherein cause SVM to promote best performance
Combination alternatively parameter.
8. dark-red enameled pottery seal search method according to claim 1, it is characterised in that：Improved BP- in the step (4)
SIFT algorithms include introducing the constraint of two space of points distance ratios, and the cluster closed on before point search is reducing time complexity.
9. the dark-red enameled pottery seal search method according to claim 1 or 8, it is characterised in that：It is improved in the step (4)
The concrete grammar of BP-SIFT algorithms, comprises the following steps：
(401) using classics SIFT algorithms to image contract SIFT feature to be retrieved, 128 are formed and ties up SIFT feature description；
(402) all characteristic points in template image are gathered for C classes according to coordinate position using KMEANS clusters；For template image
In all characteristic points, calculate which class it belongs to, and nearest K point of proximity is found out in the apoplexy due to endogenous wind；
(403) it is a constant to initialize all of confidence level, if n=1；To current signature point and figure to be retrieved in template image
The combination of the current signature point as in is right, iteratively updates its matching confidence level；
(404) constrained by the geometric distance of characteristic point and its K point of proximity and the Euclidean distance between two width image characteristic points
Linear function combinational estimation calculate matching probability；
(405) n ← n+1, then goes to (404th) step, until matching probability no longer changes or iterationses n reaches maximum；
(406) if matching degree is less than predetermined threshold value, the two Point matchings success；Otherwise it is not that matching is right；
(407) all template images are repeated with above step, the matching rate of image to be retrieved and each template image is calculated,
It is classified as the class of matching degree highest one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610955127.8A CN106570475B (en) | 2016-11-03 | 2016-11-03 | A kind of dark-red enameled pottery seal search method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610955127.8A CN106570475B (en) | 2016-11-03 | 2016-11-03 | A kind of dark-red enameled pottery seal search method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106570475A true CN106570475A (en) | 2017-04-19 |
| CN106570475B CN106570475B (en) | 2019-08-02 |
Family
ID=58535449
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610955127.8A Active CN106570475B (en) | 2016-11-03 | 2016-11-03 | A kind of dark-red enameled pottery seal search method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106570475B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108536827A (en) * | 2018-04-11 | 2018-09-14 | 南京理工大学 | A kind of similar frequency spectrum image searching method |
| CN109101979A (en) * | 2018-07-11 | 2018-12-28 | 中国艺术科技研究所 | A kind of dark-red enameled pottery bottom inscriptions distinguishing method between true and false |
| CN109767436A (en) * | 2019-01-07 | 2019-05-17 | 深圳市创业印章实业有限公司 | A kind of method and device that the seal true and false identifies |
| CN112016563A (en) * | 2020-10-17 | 2020-12-01 | 深圳神目信息技术有限公司 | Method for identifying authenticity of circular seal |
| CN113255686A (en) * | 2021-07-15 | 2021-08-13 | 恒生电子股份有限公司 | Method and device for identifying seal in image, processing equipment and storage medium |
| CN113361547A (en) * | 2021-06-30 | 2021-09-07 | 深圳证券信息有限公司 | Signature identification method, device, equipment and readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101576958A (en) * | 2009-06-15 | 2009-11-11 | 西安交通大学 | Extraction method of rectangle seal graphics with border based on form feature |
| KR101382163B1 (en) * | 2013-03-14 | 2014-04-07 | 국방과학연구소 | Ground target classification method, and ground target classification apparatus using the same |
| US20160012317A1 (en) * | 2014-07-09 | 2016-01-14 | Ditto Labs, Inc. | Systems, methods, and devices for image matching and object recognition in images using template image classifiers |
-
2016
- 2016-11-03 CN CN201610955127.8A patent/CN106570475B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101576958A (en) * | 2009-06-15 | 2009-11-11 | 西安交通大学 | Extraction method of rectangle seal graphics with border based on form feature |
| KR101382163B1 (en) * | 2013-03-14 | 2014-04-07 | 국방과학연구소 | Ground target classification method, and ground target classification apparatus using the same |
| US20160012317A1 (en) * | 2014-07-09 | 2016-01-14 | Ditto Labs, Inc. | Systems, methods, and devices for image matching and object recognition in images using template image classifiers |
Non-Patent Citations (3)
| Title |
|---|
| 徐华: "印鉴识别算法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 李灵杰 等: "一种面向交通场景的快速图像增强算法", 《计算机与现代化》 * |
| 熊琰铖 等: "基于几何信息的特征匹配改进算法", 《小型微型计算机系统》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108536827A (en) * | 2018-04-11 | 2018-09-14 | 南京理工大学 | A kind of similar frequency spectrum image searching method |
| CN108536827B (en) * | 2018-04-11 | 2021-09-03 | 南京理工大学 | Similar spectrum picture searching method |
| CN109101979A (en) * | 2018-07-11 | 2018-12-28 | 中国艺术科技研究所 | A kind of dark-red enameled pottery bottom inscriptions distinguishing method between true and false |
| CN109767436A (en) * | 2019-01-07 | 2019-05-17 | 深圳市创业印章实业有限公司 | A kind of method and device that the seal true and false identifies |
| CN112016563A (en) * | 2020-10-17 | 2020-12-01 | 深圳神目信息技术有限公司 | Method for identifying authenticity of circular seal |
| CN112016563B (en) * | 2020-10-17 | 2021-07-13 | 深圳神目信息技术有限公司 | Method for identifying authenticity of circular seal |
| CN113361547A (en) * | 2021-06-30 | 2021-09-07 | 深圳证券信息有限公司 | Signature identification method, device, equipment and readable storage medium |
| CN113255686A (en) * | 2021-07-15 | 2021-08-13 | 恒生电子股份有限公司 | Method and device for identifying seal in image, processing equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106570475B (en) | 2019-08-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570475A (en) | Purple clay teapot seal retrieval method | |
| Wang et al. | A multiscale and hierarchical feature extraction method for terrestrial laser scanning point cloud classification | |
| CN109271991B (en) | License plate detection method based on deep learning | |
| CN101398894B (en) | Automobile license plate automatic recognition method and implementing device thereof | |
| CN107103317A (en) | Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN104951940B (en) | A kind of mobile payment verification method based on personal recognition | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN108280460B (en) | SAR vehicle target identification method based on improved convolutional neural network | |
| CN109902590A (en) | Pedestrian's recognition methods again of depth multiple view characteristic distance study | |
| CN105678231A (en) | Pedestrian image detection method based on sparse coding and neural network | |
| Zhang et al. | Road recognition from remote sensing imagery using incremental learning | |
| CN104751142A (en) | Natural scene text detection algorithm based on stroke features | |
| CN104809481A (en) | Natural scene text detection method based on adaptive color clustering | |
| CN104182763A (en) | Plant type identification system based on flower characteristics | |
| CN106503694B (en) | Digit recognition method based on eight neighborhood feature | |
| CN106909902A (en) | A kind of remote sensing target detection method based on the notable model of improved stratification | |
| Wang et al. | Traffic sign detection using a cascade method with fast feature extraction and saliency test | |
| CN105975968A (en) | Caffe architecture based deep learning license plate character recognition method | |
| CN107330027A (en) | A kind of Weakly supervised depth station caption detection method | |
| CN108629286A (en) | A kind of remote sensing airport target detection method based on the notable model of subjective perception | |
| Huang et al. | Text detection and recognition in natural scene images | |
| CN111339932B (en) | Palm print image preprocessing method and system | |
| CN106599924A (en) | Classifier construction method based on three-way decision | |
| CN107357834A (en) | Image retrieval method based on visual saliency fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |