CN103106265B - Similar image sorting technique and system - Google Patents
Similar image sorting technique and system Download PDFInfo
- Publication number
- CN103106265B CN103106265B CN201310037741.2A CN201310037741A CN103106265B CN 103106265 B CN103106265 B CN 103106265B CN 201310037741 A CN201310037741 A CN 201310037741A CN 103106265 B CN103106265 B CN 103106265B
- Authority
- CN
- China
- Prior art keywords
- image
- identified
- response
- template
- response diagram
- 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
Abstract
The present invention proposes a kind of similar image sorting technique and system.Wherein, method comprises the following steps: inputs image to be identified and obtains the shape facility of image to be identified, Gradient Features, color characteristic and textural characteristics；Being split by training sample in image library and generate multiple various sizes of local area image, and carry out change of scale, it is thus achieved that image template collection, image template is concentrated and is included multiple image template；Analyze the image template in image template set, and obtain the shape facility of image template, Gradient Features, color characteristic and textural characteristics；The character pair of image to be identified with image template collection image is mated, and processes, it is thus achieved that the image detail information of image to be identified；By image representative data and utilize Bagging grader to obtain the classification of image to be identified.Method according to embodiments of the present invention, by carrying out extraction and the coupling of shape facility, Gradient Features, color characteristic and textural characteristics to input picture, it is achieved that the correct classification of similar image.
Description
Technical field
The present invention relates to image identification technical field, particularly to a kind of similar image sorting technique and system.
Background technology
Classification is substantially the cascade form of a kind of tree, in such an embodiment, near the high first nodes of root
Describe and merge classification (inclusive class), the most upper classification (super-ordinate class), such as, vehicle
Deng.Intermediate layer, also makes basic class (basic class) node describe more specific classification, such as, motorcycle or automobile etc..
And near between the low node layer of leaf node, also referred to as sub-categories (subordinate class), the most generally capture object more
Add fine distinction, such as, motor sport or multifunctional motorcycle, passenger vehicle, truck or car etc..Similar image classification is
Refer to classify, same basic class or the aspect such as shape and visual appearance and close object thereof namely to sub-categories
Object is classified, and such as, distinguishes different types of mushroom, automobile etc..
Along with developing rapidly of computer technology, artificial intelligence technology and sensor technology, and the mankind are in work, study
With the urgent needs of sphere of life, Pattern classification techniques has had evolved to a brand-new stage.But, current expert and
Person has focused largely on focus in the classification work to Primary layer (basic level) object, and seldom mentions similar image, i.e.
The classification of subordinate layer (subordinate level) object.
Traditional method for classifying modes is applied in the similar image classification that similarity is high can be often failed, and it is main
Reason has following several:
First, the commonly used codebook mode of the sorting technique of some classics extracts feature at present, and this " dictionary " is usual
Build by non-supervisory method.Use k average or gauss hybrid models cluster " entry " that obtain although sometimes in certain space
Region high probability occurs, but in fact classification judgement is not necessarily useful information.Second, when detected area maps becomes
During " dictionary entry " form, a lot of detailed information are lost.3rd, " dictionary " method needs some ginsengs of manually regulation cluster
Number, the most loaded down with trivial details, the most not necessarily select the most suitable.And in contrast to this, method based on annotations compensate for a great extent
The deficiency of above-mentioned use codebook approach, and recognition effect is also very good, but huge cost of labor makes it develop to be received
The biggest restriction.
Linked method without Supervisory Shell effectively solves Primary layer classification problem, but cannot distinguish between the subordinate that dependency is the biggest
Layer classification.Recognition methods based on attribute also show the biggest advantage.These technology are typically to utilize with attribute tags
Training data study identification model, the perceptual property that then Applied Learning model evaluation test image is appropriate.These methods pair
In identifying similar fur, the attribute such as point or four lower limbs is effective, but for distinguishing between subordinate layer object fine distinction but
Seem not satisfactory.
Summary of the invention
The purpose of the present invention is intended at least solve one of above-mentioned technological deficiency.
For reaching above-mentioned purpose, the embodiment of one aspect of the present invention proposes a kind of similar image sorting technique, including following
Step: S1: input image to be identified and obtain shape facility, Gradient Features, color characteristic and the texture of described image to be identified
Feature；S2: the training sample in image library is split and generates multiple various sizes of local area image, and carry out yardstick change
Changing, it is thus achieved that image template collection, described image template is concentrated and is included multiple image template；S3: analyze what described image template was concentrated
Image template, and obtain the shape facility of image template, Gradient Features, color characteristic and textural characteristics；S4: by described to be identified
Image mates with the character pair of described image template collection image, and processes, it is thus achieved that the figure of described image to be identified
As detailed information；S5: by described image representative data and utilize Bagging grader to obtain the classification of image to be identified.
Method according to embodiments of the present invention, by input picture is carried out shape facility, Gradient Features, color characteristic and
The extraction of textural characteristics and coupling, effectively overcome the factor images to classification results such as affine, illumination, ensure that image simultaneously
The integrity of expression information, rich and discriminability, it is ensured that the correct classification of similar image.
In an example of the present invention, described step S2 specifically includes: S21: split by the training sample in image library raw
Become multiple various sizes of local area image, constitute the first image template collection；S22: described first image template is concentrated
Each image template carries out change of scale, it is thus achieved that the image template of different scale, pie graph is as template set.
In an example of the present invention, described step S4 specifically includes: S41: by described image to be identified and described image
In template set, the character pair of image mates, it is thus achieved that the characteristic response figure of image to be identified, wherein, and described image to be identified
Characteristic response figure include shape response diagram, gradient response diagram, color response figure and texture response diagram；S42: by described to be identified
Numerical value in every width response diagram of the shape response diagram of image, gradient response diagram, color response figure and texture response diagram is by from greatly
To little sequence, then take multiple data before in sorted numerical value, form the fisrt feature collection of this width response diagram；S43: by described
Every width response diagram of the shape response diagram of image to be identified, gradient response diagram, color response figure and texture response diagram is respectively divided
For multiple regions, and respectively by the numerical value in each region in described every width response diagram by sorting from big to small, then distinguish
Take multiple data before in sorted numerical value, in certain sequence the data that all regions are taken out are formed the second of this width response diagram
Feature set；S44: fisrt feature collection and the second feature collection of described every width response diagram are connected, composition characteristic vector, respectively
The characteristic vector of all shape response diagrams is connected, the characteristic vector series connection of all gradient response diagrams, all colours response diagram
The characteristic vector of the textured response diagram of characteristic vector series connection and institute is connected, obtain the shape blending feature of image to be identified to
Amount, gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector；S45: by described image to be identified
Shape blending characteristic vector, gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector are carried out again
Merge, generate final fusion feature vector, be used for representing image detail information to be identified.
In an example of the present invention, described step S5 specifically includes: S51: the target letter of design Bagging grader
Number；S52: by described object function and utilize described image detail information obtain characteristic vector weight sets；S53: according to multiple
Grader determines the classification of described image to be identified to the confidence level of described characteristic vector weight sets and exports.
For reaching above-mentioned purpose, on the other hand embodiments of the invention propose a kind of similar image categorizing system, including: defeated
Enter module, for inputting image to be identified and obtaining shape facility, Gradient Features, color characteristic and the stricture of vagina of described image to be identified
Reason feature；Divide module, generate multiple various sizes of local area image for being split by the training sample in image library, and
Local area image carrying out change of scale respectively, it is thus achieved that image template collection, described image template is concentrated and is included multiple image mould
Plate；Analyze module, for analyzing the image template that described image template is concentrated, and obtain the shape facility of image template, gradient
Feature, color characteristic and textural characteristics；Matching module, for concentrating image by described image to be identified with described image template
Character pair mates, and processes, it is thus achieved that the image detail information of described image to be identified；Obtain module, be used for leading to
Cross described image representative data and utilize Bagging grader to obtain the classification of image to be identified.
Method according to embodiments of the present invention, by input picture is carried out shape facility, Gradient Features, color characteristic and
The extraction of textural characteristics and coupling, effectively overcome the factor images to classification results such as affine, illumination, ensure that image simultaneously
The integrity of expression information, rich and discriminability, it is ensured that the correct classification of similar image.
In an example of the present invention, described division module includes: extraction unit, for by the training sample in image library
Segmentation generates multiple various sizes of local area image, constitutes the first image template collection；Change of scale unit, for described
Each image template that first image template is concentrated carries out change of scale, it is thus achieved that the image template of different scale, pie graph is as mould
Plate collection.
In an example of the present invention, described matching module includes: matching unit, for by described image to be identified and institute
The character pair stating image template collection image mates, and obtains the characteristic response figure of image to be identified respectively, wherein, described in treat
Identify that the characteristic response figure of image includes shape response diagram, gradient response diagram, color response figure and texture response diagram；Search single
Unit, for ringing every width of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified
Should numerical value in figure by sorting from big to small, then take multiple data before in sorted numerical value, form the of this width response diagram
One feature set；Division unit, for by shape response diagram, gradient response diagram, color response figure and the stricture of vagina of described image to be identified
Every width response diagram of reason response diagram is respectively divided into multiple region, and respectively by each region in described every width response diagram
Numerical value, by sorting from big to small, takes multiple data before in sorted numerical value the most respectively, is taken in all regions in certain sequence
The data gone out form the second feature collection of this width response diagram；Integrated unit, for described all features are carried out effective integration, obtains
Obtain final fusion feature vector, be used for representing image detail information to be identified；First integrated unit, for responding described every width
Fisrt feature collection and the second feature collection of figure are connected, generate characteristic vector, respectively by the feature of all shape response diagrams to
Amount series connection, the characteristic vector series connection of all gradient response diagrams, the characteristic vector series connection of all colours response diagram and institute are textured
The characteristic vector series connection of response diagram, obtains the shape blending characteristic vector of image to be identified, gradient fusion feature vector, color melts
Close characteristic vector and grain table characteristic vector；And second integrated unit, for by the shape blending of described image to be identified
Characteristic vector, gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector merge again, generate
Final fusion feature vector, is used for representing image detail information to be identified.
In an example of the present invention, described acquisition module includes: design cell, for designing Bagging grader
Object function；Obtain unit, be used for by described object function and utilize described image detail information to obtain characteristic vector weight
Collection；Determine output unit, for determining described to be identified according to multiple graders to the confidence level of described characteristic vector weight sets
The classification of image also exports.
Aspect and advantage that the present invention adds will part be given in the following description, and part will become from the following description
Obtain substantially, or recognized by the practice of the present invention.
Accompanying drawing explanation
The present invention above-mentioned and/or that add aspect and advantage will become from the following description of the accompanying drawings of embodiments
Substantially with easy to understand, wherein:
Fig. 1 is the flow chart of the similar image sorting technique according to one embodiment of the invention；
Fig. 2 is according to the comparison diagram of difference between difference and class in the class of one embodiment of the invention；
Fig. 3 is the frame diagram of the similar image categorizing system according to one embodiment of the invention；
Fig. 4 is the structured flowchart of the matching module according to one embodiment of the invention.
Detailed description of the invention
Embodiments of the invention are described below in detail, and the example of embodiment is shown in the drawings, the most identical
Or similar label represents same or similar element or has the element of same or like function.Retouch below with reference to accompanying drawing
The embodiment stated is exemplary, is only used for explaining the present invention, and is not construed as limiting the claims.
Fig. 1 is the flow chart of the similar image sorting technique according to one embodiment of the invention.As it is shown in figure 1, according to this
The similar image sorting technique of inventive embodiments, comprises the following steps:
Step S101, input image to be identified and obtain the shape facility of image to be identified, Gradient Features, color characteristic and
Textural characteristics.
Specifically, due to SIFT operator to translation, rotation, scaling, brightness flop, block and noise etc. has good
Invariance, visible change, affine transformation are also kept a certain degree of stability, therefore the present invention extracts SIFT operator conduct
Picture shape feature.LBP has rotational invariance, can be largely overcoming the illumination variation impact on image simultaneously,
Therefore extract LBP operator to represent as image texture characteristic.Additionally, pertinent literature shows, the gradient of image and color are images
The maximally effective description of information, therefore Simultaneous Extracting Image Gradient Features and color characteristic represent image.
Step S102, splits the training sample in image library and generates multiple various sizes of local area image, go forward side by side
Row change of scale, it is thus achieved that image template collection, image template is concentrated and is included multiple image template.
Specifically, the training sample in image library is split and generates multiple various sizes of local area image, constitute the
One image template collection, and each image template of the first image template concentration is carried out change of scale, it is thus achieved that the figure of different scale
As template, pie graph is as template set.Wherein, the size dimension of template image can be adjusted and specification according to actual needs.
Step S103, analyzes the image template in image template set, it is thus achieved that the shape facility of image template, Gradient Features,
Color characteristic and textural characteristics.
Specifically, due to SIFT operator to translation, rotation, scaling, brightness flop, block and noise etc. has good
Invariance, visible change, affine transformation are also kept a certain degree of stability, therefore the present invention extracts SIFT operator conduct
Picture shape feature.LBP has rotational invariance, can be largely overcoming the illumination variation impact on image simultaneously,
Therefore extract LBP operator to represent as image texture characteristic.Additionally, pertinent literature shows, the gradient of image and color are images
The maximally effective description of information, therefore Simultaneous Extracting Image Gradient Features and color characteristic represent image.
Step S104, concentrates the character pair of image to carry out mating and processing with image template image to be identified,
Obtain the image detail information of image to be identified.
Specifically, concentrate the character pair of image to mate with image template image to be identified, obtain respectively and wait to know
The shape response diagram of other image, gradient response diagram, color response figure and texture response diagram.Then, by the shape of image to be identified
Numerical value in every width response diagram of response diagram, gradient response diagram, color response figure and texture response diagram is by sorting from big to small, so
After take in sorted numerical value before multiple data, form the fisrt feature collection of this width response diagram, and by the shape of image to be identified
Every width response diagram of shape response diagram, gradient response diagram, color response figure and texture response diagram is respectively divided into multiple region, and divides
Not by the numerical value in each region in every width response diagram by sorting from big to small, many before taking the most respectively in sorted numerical value
The data that all regions are taken out are formed the second feature collection of this width response diagram by individual data in certain sequence.Afterwards every width is rung
Should the fisrt feature collection of figure and second feature collection connect, composition characteristic vector, respectively by the feature of all shape response diagrams
Vector series connection, the characteristic vector series connection of all gradient response diagrams, the characteristic vector series connection of all colours response diagram and all stricture of vaginas
The characteristic vector series connection of reason response diagram, obtains the shape blending characteristic vector of image to be identified, gradient fusion feature vector, color
Fusion feature vector sum grain table characteristic vector.Finally, the shape blending characteristic vector of image to be identified, gradient are merged spy
Levy vector, color blend characteristic vector and grain table characteristic vector again to merge, generate final fusion feature vector, use
Represent image detail information to be identified.
Step S105, by image representative data and utilize Bagging grader to obtain the classification of image to be identified.
Specifically, the object function of design Bagging grader, its object function is as follows:
Then pass through object function and utilize image detail information to obtain characteristic vector weight sets, and according to multiple graders
The confidence level of characteristic vector weight sets is determined the classification of image to be identified and exports.
Fig. 2 is according to the comparison diagram of difference between difference and class in the class of one embodiment of the invention.As in figure 2 it is shown, in figure
A () and (c) is Herba Taraxaci, (b) is coltsfoot.Respective image can be determined accurately after similar image is classified
Specific category.
Method according to embodiments of the present invention, by input picture is carried out shape facility, Gradient Features, color characteristic and
The extraction of textural characteristics and coupling, effectively overcome the factor images to classification results such as affine, illumination, ensure that image simultaneously
The integrity of expression information, rich and discriminability, it is ensured that the correct classification of similar image.
Fig. 3 is the frame diagram of the similar image categorizing system according to another embodiment of the present invention.As it is shown on figure 3, according to
The similar image categorizing system of the embodiment of the present invention includes input module 100, divides module 200, analysis module 300, coupling mould
Block 400 and acquisition module 500.
Input module 100 is for inputting image to be identified and obtaining the shape facility of image to be identified, Gradient Features, color
Feature and textural characteristics.
In one embodiment of the invention, due to SIFT operator to translation, rotation, scaling, brightness flop, block
With noise etc., there is good invariance, visible change, affine transformation are also kept a certain degree of stability, the therefore present invention
Extract SIFT operator as picture shape feature.LBP has rotational invariance, can be largely overcoming illumination simultaneously and become
Change the impact on image, therefore extract LBP operator and represent as image texture characteristic.Additionally, pertinent literature shows, the ladder of image
Degree and color are the maximally effective descriptions of image information, and therefore Simultaneous Extracting Image Gradient Features and color characteristic represent image.
Divide module 200 and generate multiple various sizes of regional area figures for being split by the training sample in image library
Picture, and local area image is carried out respectively change of scale, it is thus achieved that image template collection, image template is concentrated and is included multiple image mould
Plate.
In one embodiment of the invention, divide module 200 and include extraction unit 210 and change of scale unit 220.
Extraction unit 210 generates multiple various sizes of regional area figures for being split by the training sample in image library
Picture, constitutes the first image template collection.
Change of scale unit 220 carries out change of scale for each image template concentrating the first image template, it is thus achieved that
The image template of different scale, pie graph is as template set.
In one embodiment of the invention, the training sample in image library is split the multiple various sizes of local of generation
Area image, constitutes the first image template collection, and each image template concentrating the first image template carries out change of scale, obtains
Obtaining the image template of different scale, pie graph is as template set.Wherein, the size dimension of template image can enter according to actual needs
Row sum-equal matrix and specification.
Analyze module 300 to be used for analyzing the image template in image template set, and obtain the shape facility of image template, ladder
Degree feature, color characteristic and textural characteristics.
In one embodiment of the invention, due to SIFT operator to translation, rotation, scaling, brightness flop, block
With noise etc., there is good invariance, visible change, affine transformation are also kept a certain degree of stability, the therefore present invention
Extract SIFT operator as picture shape feature.LBP has rotational invariance, can be largely overcoming illumination simultaneously and become
Change the impact on image, therefore extract LBP operator and represent as image texture characteristic.Additionally, pertinent literature shows, the ladder of image
Degree and color are the maximally effective descriptions of image information, and therefore Simultaneous Extracting Image Gradient Features and color characteristic represent image.
Matching module 400, for concentrating the character pair of image to mate with image template image to be identified, is gone forward side by side
Row processes, it is thus achieved that the image detail information of image to be identified.
In one embodiment of the invention, matching module 400 includes matching unit 410, searches unit 420, division list
Unit's the 430, first integrated unit 440 and the second integrated unit 450.
Matching unit 410, for being mated by the character pair of image to be identified with image template collection image, obtains respectively
Obtaining the characteristic response figure of image to be identified, wherein, the characteristic response figure of image to be identified includes that shape response diagram, gradient respond
Figure, color response figure and texture response diagram.
Search unit 420 for the shape response diagram of image to be identified, gradient response diagram, color response figure and texture being rung
Should numerical value in every width response diagram of figure by sorting from big to small, then take multiple data before in sorted numerical value, composition should
The fisrt feature collection of width response diagram.
Division unit 430 is for ringing the shape response diagram of image to be identified, gradient response diagram, color response figure and texture
Every width response diagram of figure should be respectively divided into multiple region, and respectively by the numerical value in each region in every width response diagram by from
Big to little sequence, take multiple data before in sorted numerical value, the data taken out in all regions in certain sequence the most respectively
Form the second feature collection of this width response diagram.
First integrated unit 440, for fisrt feature collection and the second feature collection of every width response diagram being connected, generates
Characteristic vector, connects the characteristic vector of the characteristic vector series connection of all shape response diagrams, all gradient response diagrams, owns respectively
Color response figure characteristic vector series connection and institute textured response diagram characteristic vector connect, obtain the shape of image to be identified
Fusion feature vector, gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector.
Second integrated unit 450 is for by the shape blending characteristic vector of image to be identified, gradient fusion feature vector, face
Color fusion feature vector sum grain table characteristic vector merges again, generates final fusion feature vector, is used for representing and treats
Identify image detail information.
In one embodiment of the invention, image to be identified and image template are concentrated the character pair of image carry out
Join, obtain the shape response diagram of image to be identified, gradient response diagram, color response figure and texture response diagram respectively.Then, will treat
Identify that the numerical value in every width response diagram of the shape response diagram of image, gradient response diagram, color response figure and texture response diagram is pressed
Sort from big to small, then take multiple data before in sorted numerical value, form the fisrt feature collection of this width response diagram, and will
Every width response diagram of the shape response diagram of image to be identified, gradient response diagram, color response figure and texture response diagram is respectively divided
For multiple regions, and respectively by the numerical value in each region in every width response diagram by sorting from big to small, the row of taking the most respectively
The data that all regions are taken out are formed the second feature of this width response diagram by multiple data in certain sequence before in the numerical value of good sequence
Collection.Fisrt feature collection and the second feature collection of every width response diagram being connected afterwards, composition characteristic vector, respectively by tangible for institute
The characteristic vector series connection of shape response diagram, the characteristic vector series connection of all gradient response diagrams, the characteristic vector of all colours response diagram
The characteristic vector of the textured response diagram of series connection and institute is connected, and obtains the shape blending characteristic vector of image to be identified, gradient is melted
Close characteristic vector, color blend characteristic vector and grain table characteristic vector.Finally, by the shape blending feature of image to be identified
Vector, gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector merge again, generate final
Fusion feature vector, is used for representing image detail information to be identified.
Obtain module 500 be used for by image representative data and utilize Bagging grader to obtain the class of image to be identified
Not.
In one embodiment of the invention, it is thus achieved that module 500 includes design cell 510, obtains unit 520 and determine defeated
Go out unit 530.
Design cell 510 is for designing the object function of Bagging grader.Its object function is as follows:
Obtain unit 520 be used for by object function and utilize image detail information to obtain characteristic vector weight sets.
Determine that output unit 530 is for determining figure to be identified according to multiple graders to the confidence level of characteristic vector weight sets
The classification of picture also exports.
Fig. 2 is according to the comparison diagram of difference between difference and class in the class of one embodiment of the invention.As in figure 2 it is shown, in figure
A () and (c) is Herba Taraxaci, (b) is coltsfoot.Respective image can be determined accurately after similar image is classified
Specific category.
Method according to embodiments of the present invention, by input picture is carried out shape facility, Gradient Features, color characteristic and
The extraction of textural characteristics and coupling, effectively overcome the factor images to classification results such as affine, illumination, ensure that image simultaneously
The integrity of expression information, rich and discriminability, it is ensured that the correct classification of similar image.
Although above it has been shown and described that embodiments of the invention, it is to be understood that above-described embodiment is example
Property, it is impossible to be interpreted as limitation of the present invention, those of ordinary skill in the art is without departing from the principle of the present invention and objective
In the case of above-described embodiment can be changed within the scope of the invention, revise, replace and modification.
Claims (6)
1. a similar image sorting technique, it is characterised in that comprise the following steps:
S1: input image to be identified and obtain the shape facility of described image to be identified, Gradient Features, color characteristic and texture spy
Levy；
S2: the training sample in image library is split and generates multiple various sizes of local area image, and carry out change of scale,
Obtaining image template collection, described image template is concentrated and is included multiple image template；
S3: analyze the image template that described image template is concentrated, and obtain the shape facility of image template, Gradient Features, color
Feature and textural characteristics；
S4: the character pair of the image template concentrated with described image template by described image to be identified mates, and carries out
Process, it is thus achieved that the image detail information of described image to be identified；
S5: obtained the class of described image to be identified by the image detail information of described image to be identified and Bagging grader
Not,
Described step S4 specifically includes:
S41: the character pair of the image template concentrated with described image template by described image to be identified mates, it is thus achieved that treat
Identifying the characteristic response figure of image, wherein, the characteristic response figure of described image to be identified includes that shape response diagram, gradient respond
Figure, color response figure and texture response diagram；
S42: by every width of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified
Numerical value in response diagram, by sorting from big to small, then takes multiple data before in sorted numerical value, forms this width response diagram
Fisrt feature collection；
S43: by every width of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified
Response diagram is respectively divided into multiple region, and is pressed from big to small by the numerical value in each region in described every width response diagram respectively
Sequence, takes multiple data before in sorted numerical value the most respectively, and the data taken out in all regions in certain sequence composition should
The second feature collection of width response diagram；
S44: fisrt feature collection and the second feature collection of described every width response diagram are connected, composition characteristic vector, respectively will
The characteristic vector series connection of all shape response diagrams, the characteristic vector series connection of all gradient response diagrams, the spy of all colours response diagram
Levy vector series connection and institute textured response diagram characteristic vector connect, obtain image to be identified shape blending characteristic vector,
Gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector；
S45: by the shape blending characteristic vector of described image to be identified, gradient fusion feature vector, color blend characteristic vector
Again merge with grain table characteristic vector, generate final fusion feature vector, be used for representing that image detail to be identified is believed
Breath.
Similar image sorting technique the most according to claim 1, it is characterised in that described step S2 specifically includes:
S21: the training sample in image library is split and generates multiple various sizes of local area image, constitute the first image mould
Plate collection；
S22: each image template concentrating described first image template carries out change of scale, it is thus achieved that the image mould of different scale
Plate, pie graph is as template set.
Similar image sorting technique the most according to claim 1, it is characterised in that described step S5 specifically includes:
S51: the object function of design Bagging grader；
S52: obtain characteristic vector weight sets with described image detail information by described object function；
S53: the confidence level of described characteristic vector weight sets is determined the classification of described image to be identified also according to multiple graders
Output.
4. a similar image categorizing system, it is characterised in that including:
Input module, for inputting image to be identified and obtaining the shape facility of described image to be identified, Gradient Features, color spy
Seek peace textural characteristics；
Divide module, generate multiple various sizes of local area image for being split by the training sample in image library, and will
Local area image carries out change of scale respectively, it is thus achieved that image template collection, and described image template is concentrated and included multiple image template；
Analyze module, for analyzing the image template that described image template is concentrated, and obtain the shape facility of image template, gradient
Feature, color characteristic and textural characteristics；
Matching module, the character pair of the image template for described image to be identified and described image template being concentrated is carried out
Join, and process, it is thus achieved that the image detail information of described image to be identified；
Obtain module, wait to know described in being obtained by the image detail information of described image to be identified and Bagging grader
The classification of other image,
Described matching module includes:
Matching unit, the character pair of the image template for described image to be identified and described image template being concentrated is carried out
Joining, obtain the characteristic response figure of image to be identified respectively, wherein, the characteristic response figure of described image to be identified includes that shape responds
Figure, gradient response diagram, color response figure and texture response diagram；
Search unit, for by the shape response diagram of described image to be identified, gradient response diagram, color response figure and texture response
Numerical value in every width response diagram of figure, by sorting from big to small, then takes multiple data before in sorted numerical value, forms this width
The fisrt feature collection of response diagram；
Division unit, for by the shape response diagram of described image to be identified, gradient response diagram, color response figure and texture response
Every width response diagram of figure is respectively divided into multiple region, and is pressed by the numerical value in each region in described every width response diagram respectively
Sort from big to small, take multiple data before in sorted numerical value, the number taken out in all regions in certain sequence the most respectively
According to the second feature collection forming this width response diagram；
First integrated unit, for fisrt feature collection and the second feature collection of described every width response diagram being connected, generates spy
Levy vector, respectively by the characteristic vector series connection of all shape response diagrams, the characteristic vector series connection of all gradient response diagrams, all face
Colour response figure characteristic vector series connection and institute textured response diagram characteristic vector connect, the shape obtaining image to be identified is melted
Close characteristic vector, gradient fusion feature vector, color blend characteristic vector and grain table characteristic vector；And
Second integrated unit, for by the shape blending characteristic vector of described image to be identified, gradient fusion feature vector, color
Fusion feature vector sum grain table characteristic vector merges again, generates final fusion feature vector, is used for representing and waits to know
Other image detail information.
Similar image categorizing system the most according to claim 4, it is characterised in that described division module includes:
Extraction unit, generates multiple various sizes of local area image for being split by the training sample in image library, constitutes
First image template collection；
Change of scale unit, carries out change of scale for each image template concentrating described first image template, it is thus achieved that no
With the image template of yardstick, pie graph is as template set.
Similar image categorizing system the most according to claim 4, it is characterised in that described acquisition module includes:
Design cell, for designing the object function of Bagging grader；
Obtain unit, for obtaining characteristic vector weight sets by described object function with described image detail information；
Determine output unit, for determining described to be identified according to multiple graders to the confidence level of described characteristic vector weight sets
The classification of image also exports.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310037741.2A CN103106265B (en) | 2013-01-30 | 2013-01-30 | Similar image sorting technique and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310037741.2A CN103106265B (en) | 2013-01-30 | 2013-01-30 | Similar image sorting technique and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103106265A CN103106265A (en) | 2013-05-15 |
| CN103106265B true CN103106265B (en) | 2016-10-12 |
Family
ID=48314120
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310037741.2A Expired - Fee Related CN103106265B (en) | 2013-01-30 | 2013-01-30 | Similar image sorting technique and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103106265B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530645B (en) * | 2013-09-25 | 2016-06-15 | 北京工商大学 | Texture image classification method and system based on local binary patterns and Zernike square |
| US9384422B2 (en) | 2014-04-04 | 2016-07-05 | Ebay Inc. | Image evaluation |
| CN104217221A (en) * | 2014-08-27 | 2014-12-17 | 重庆大学 | Method for detecting calligraphy and paintings based on textural features |
| CN105809182B (en) * | 2014-12-31 | 2019-12-13 | 中国科学院深圳先进技术研究院 | Image classification method and device |
| CN104680168A (en) * | 2015-03-10 | 2015-06-03 | 无锡桑尼安科技有限公司 | Extraction equipment of feature vectors for highway landform recognition |
| CN104809245A (en) * | 2015-05-13 | 2015-07-29 | 信阳师范学院 | Image retrieval method |
| CN104834732A (en) * | 2015-05-13 | 2015-08-12 | 信阳师范学院 | Texture image retrieving method |
| CN105631451A (en) * | 2016-01-07 | 2016-06-01 | 同济大学 | Plant leave identification method based on android system |
| CN106354850A (en) * | 2016-08-31 | 2017-01-25 | 广州精点计算机科技有限公司 | Image recognition method based on K-nearest neighbor classification |
| CN108319952B (en) * | 2017-01-16 | 2021-02-02 | 浙江宇视科技有限公司 | Vehicle feature extraction method and device |
| CN107341189A (en) * | 2017-06-09 | 2017-11-10 | 国政通科技股份有限公司 | A kind of indirect labor carries out the method and system of examination, classification and storage to image |
| CN108776805A (en) * | 2018-05-03 | 2018-11-09 | 北斗导航位置服务（北京）有限公司 | It is a kind of establish image classification model, characteristics of image classification method and device |
| CN109102264B (en) * | 2018-08-29 | 2021-07-16 | Oppo广东移动通信有限公司 | Electronic red packet detection method and device and terminal equipment |
| CN114235122A (en) * | 2021-12-16 | 2022-03-25 | 广州市超赢信息科技有限公司 | Weighing settlement method and system of electronic scale based on AI image recognition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101763440A (en) * | 2010-03-26 | 2010-06-30 | 上海交通大学 | Method for filtering searched images |
| CN102254043A (en) * | 2011-08-17 | 2011-11-23 | 电子科技大学 | Semantic mapping-based clothing image retrieving method |
| CN102663391A (en) * | 2012-02-27 | 2012-09-12 | 安科智慧城市技术（中国）有限公司 | Image multifeature extraction and fusion method and system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8254678B2 (en) * | 2008-08-27 | 2012-08-28 | Hankuk University Of Foreign Studies Research And Industry-University Cooperation Foundation | Image segmentation |
| US8447119B2 (en) * | 2010-03-16 | 2013-05-21 | Nec Laboratories America, Inc. | Method and system for image classification |
-
2013
- 2013-01-30 CN CN201310037741.2A patent/CN103106265B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101763440A (en) * | 2010-03-26 | 2010-06-30 | 上海交通大学 | Method for filtering searched images |
| CN102254043A (en) * | 2011-08-17 | 2011-11-23 | 电子科技大学 | Semantic mapping-based clothing image retrieving method |
| CN102663391A (en) * | 2012-02-27 | 2012-09-12 | 安科智慧城市技术（中国）有限公司 | Image multifeature extraction and fusion method and system |
Non-Patent Citations (3)
| Title |
|---|
| 图像特征提取方法的综述;王志瑞等;《吉首大学学报（自然科学版）》;20110930;第32卷(第5期);第43-47页 * |
| 基于梯度基元聚合矢量的图像检索算法;张尤赛等;《计算机应用研究》;20120331;第29卷(第3期);第1119-1122页 * |
| 基于结构和纹理特征融合的场景图像分类;程刚等;《计算机工程》;20110331;第37卷(第5期);第227-229页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103106265A (en) | 2013-05-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103106265B (en) | Similar image sorting technique and system | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN106203395B (en) | Face attribute recognition method based on multitask deep learning | |
| CN106650806B (en) | A kind of cooperating type depth net model methodology for pedestrian detection | |
| CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
| CN107766933B (en) | Visualization method for explaining convolutional neural network | |
| CN104504366A (en) | System and method for smiling face recognition based on optical flow features | |
| CN109086792A (en) | Based on the fine granularity image classification method for detecting and identifying the network architecture | |
| CN102054178B (en) | A kind of image of Chinese Painting recognition methods based on local semantic concept | |
| CN103514456B (en) | Image classification method and device based on compressed sensing multi-core learning | |
| CN108549926A (en) | A kind of deep neural network and training method for refining identification vehicle attribute | |
| CN102314614B (en) | Image semantics classification method based on class-shared multiple kernel learning (MKL) | |
| CN104598885B (en) | The detection of word label and localization method in street view image | |
| CN110458077B (en) | Vehicle color identification method and system | |
| CN105117739A (en) | Clothes classifying method based on convolutional neural network | |
| CN110163069B (en) | Lane line detection method for driving assistance | |
| CN104504365A (en) | System and method for smiling face recognition in video sequence | |
| CN105719285A (en) | Pedestrian detection method based on directional chamfering distance characteristics | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN103996018A (en) | Human-face identification method based on 4DLBP | |
| CN106340016A (en) | DNA quantitative analysis method based on cell microscope image | |
| CN109886161A (en) | A kind of road traffic index identification method based on possibility cluster and convolutional neural networks | |
| CN105718866A (en) | Visual target detection and identification method | |
| CN108230330B (en) | Method for quickly segmenting highway pavement and positioning camera | |
| CN105138975B (en) | A kind of area of skin color of human body dividing method based on degree of depth conviction network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Wang Yu Inventor after: Yu Zhongzhong Inventor after: Zhang Huiyan Inventor after: Yan Huaixin Inventor before: Wang Yu Inventor before: Yu Zhongzhong Inventor before: Zhang Huiyan |
|
| COR | Change of bibliographic data | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20161012 Termination date: 20200130 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |